# chatgpt_fewshot # cola ## 10 prompts Acc: 81.00%, prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Acc: 80.50%, prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Acc: 80.50%, prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Acc: 80.50%, prompt: Review the sentence below and identify whether its grammar is 'Acceptable' or 'Unacceptable'. Acc: 80.50%, prompt: Determine if the grammar of the given sentence is 'Acceptable' or 'Unacceptable'. Acc: 80.00%, prompt: Assess the following sentence and determine if it is grammatically correct. Respond with 'Acceptable' or 'Unacceptable'. Acc: 80.00%, prompt: Please evaluate the grammatical structure of the provided sentence and answer with 'Acceptable' or 'Unacceptable'. Acc: 80.00%, prompt: Check the grammar of the following sentence and indicate if it is 'Acceptable' or 'Unacceptable'. Acc: 79.50%, prompt: Is the provided sentence grammatically correct? Respond with 'Acceptable' or 'Unacceptable'. Acc: 79.00%, prompt: Assess the grammatical structure of the given sentence and classify it as 'Acceptable' or 'Unacceptable'. Acc: 81.50%, prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Acc: 81.00%, prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Acc: 81.00%, prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Acc: 80.50%, prompt: Acting as a grammar detection instrument, determine if the provided sentence is grammatically sound, answering with 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Acc: 80.00%, prompt: As a grammar identification system, examine the provided sentence and respond with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Acc: 80.00%, prompt: As a tool for determining grammatical correctness, review the sentence and categorize its grammar as either 'acceptable' or 'unacceptable'. Acc: 80.00%, prompt: Working as a grammar classifier, identify whether the given sentence has correct grammar, and respond with 'acceptable' for correct sentences or 'unacceptable' for incorrect ones. Acc: 80.00%, prompt: As an instrument for grammar evaluation, consider the sentence and determine if its grammar is correct, responding with 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Acc: 79.50%, prompt: In your role as a grammar check tool, assess the following sentence and classify it as 'acceptable' if it is grammatically correct or 'unacceptable' if it is incorrect. Acc: 79.50%, prompt: In the capacity of a grammar assessment system, indicate if the structure of the provided sentence is grammatically correct, responding with 'acceptable' or 'unacceptable'. ## bertattack Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"examine the sentence and pick if its adjective is 'Acceptable' or 'Unacceptable'. " Original acc: 81.00%, attacked acc: 61.00%, dropped acc: 20.00% Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analyze the reported speech and classify its inclusive correcting as 'Acceptable' or 'Unacceptable'. " Original acc: 81.00%, attacked acc: 59.00%, dropped acc: 22.00% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given sentence and check if it is grammatically complex. Answer with either 'Acceptable' or 'Unacceptable'. " Original acc: 79.50%, attacked acc: 63.00%, dropped acc: 16.50% Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"or as a hierarchy evaluation tool, analyze the given sentence and decide if it is grammatically applicable, returning with 'acceptable' or 'unacceptable'. " Original acc: 81.50%, attacked acc: 77.00%, dropped acc: 4.50% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"While performing grammar ranking, classify the category of the a sentence as 'acceptable' for correct spell or 'unacceptable' for incorrect grammar. " Original acc: 81.00%, attacked acc: 78.00%, dropped acc: 3.00% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"using the art of a level analyzeg, examine the grammar of the provided sentence and decide if it is genuine, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. " Original acc: 81.00%, attacked acc: 75.50%, dropped acc: 5.50% ## checklist Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable' vTAjHynoIG. " Original acc: 80.50%, attacked acc: 78.50%, dropped acc: 2.00% Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable' 9Q3S1F94fE. " Original acc: 81.00%, attacked acc: 78.00%, dropped acc: 3.00% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable' 9Q3S1F94fE. " Original acc: 79.50%, attacked acc: 78.00%, dropped acc: 1.50% Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable' 6a4Yn3RGVc. " Original acc: 81.50%, attacked acc: 79.50%, dropped acc: 2.00% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar IhuBIhoPGc. " Original acc: 81.50%, attacked acc: 79.50%, dropped acc: 2.00% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones d6ZQ3u0GBQ. " Original acc: 81.00%, attacked acc: 79.50%, dropped acc: 1.50% ## deepwordbug Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analyze the provNded sentence and classifE its grammatical correctness as 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 79.50%, dropped acc: 1.00% Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examie the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 80.00%, dropped acc: 0.50% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given sentence and decide if it is gramatically sound. nAswer with either 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 78.50%, dropped acc: 2.00% Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"Funtioning as a graMmar evaluation tool, analyzY the given sentence and decide if it is grammatically corLect, responding with 'acceptable' or 'unacceptable'. " Original acc: 81.50%, attacked acc: 78.50%, dropped acc: 3.00% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"While performing grammar analysis, classify the grammar of the follownig sentence as 'acceptable' for coKrrect grammar or 'unacceptable' for incorreBct grammar. " Original acc: 81.00%, attacked acc: 79.50%, dropped acc: 1.50% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"In the role of a syntax analyzer, examine the grammhr of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ons. " Original acc: 81.00%, attacked acc: 79.50%, dropped acc: 1.50% ## stresstest Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable' and true is true . " Original acc: 80.50%, attacked acc: 80.50%, dropped acc: 0.00% Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable' and true is true and true is true and true is true and true is true and true is true . " Original acc: 81.00%, attacked acc: 80.00%, dropped acc: 1.00% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable' and true is true and true is true and true is true and true is true and true is true . " Original acc: 80.50%, attacked acc: 79.50%, dropped acc: 1.00% Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable' and false is not true . " Original acc: 81.50%, attacked acc: 80.50%, dropped acc: 1.00% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar and false is not true . " Original acc: 81.00%, attacked acc: 81.50%, dropped acc: -0.50% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones and false is not true . " Original acc: 81.00%, attacked acc: 81.00%, dropped acc: 0.00% ## textbugger Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analysing the provided sentences and classify its pronouns correct as 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 67.50%, dropped acc: 13.00% Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the sentence and decide if its gr\xc9\x91mmar is 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 79.00%, dropped acc: 1.50% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given condemnation and deci\xd4\x81e if it is grammatically s\xd0\xbeund. A\xd5\xb8swer with either 'Acceptable' or 'Unacceptable'. " Original acc: 80.00%, attacked acc: 77.00%, dropped acc: 3.00% Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"Funct\xd1\x96oning as a gra\xef\xbd\x8dmar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. " Original acc: 81.50%, attacked acc: 78.50%, dropped acc: 3.00% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"Whil\xd0\xb5 performing g\xe2\xb2\x85ammar analysis, classify the grammar of the following sentencing as 'acceptable' for rectify grammar or 'unacceptable' for incorrect grammar. " Original acc: 81.00%, attacked acc: 78.00%, dropped acc: 3.00% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"In the role of a syntax checker, examine the gramm\xc9\x91r of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentence or 'unacceptable' for incorrect ones. " Original acc: 81.00%, attacked acc: 77.50%, dropped acc: 3.50% ## textfooler Original prompt: Examine the sentence and decide if its grammar is 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Exams the sentence and defining if its jargon is 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 39.50%, dropped acc: 41.00% Original prompt: Analyze the provided sentence and classify its grammatical correctness as 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Analyze the afforded sentence and hierarchy its grammatical correctness as 'Acceptable' or 'Unacceptable'. " Original acc: 80.50%, attacked acc: 77.50%, dropped acc: 3.00% Original prompt: Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'. Attacked prompt: b"Examine the given verdicts and opted if it is grammatically judicious. Satisfy with either 'Acceptable' or 'Unacceptable'. " Original acc: 80.00%, attacked acc: 76.00%, dropped acc: 4.00% Acc: 81.00%, prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Original prompt: Functioning as a grammar evaluation tool, analyze the given sentence and decide if it is grammatically correct, responding with 'acceptable' or 'unacceptable'. Attacked prompt: b"Task as a grammar estimate utility, analyze the given sentencing and decide if it is grammatically correct, reacting with 'acceptable' or 'unacceptable'. " Original acc: 82.00%, attacked acc: 76.50%, dropped acc: 5.50% Original prompt: While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar. Attacked prompt: b"During exerting grammar appraisals, classified the grammar of the later sentence as 'acceptable' for correct grammar or 'unacceptable' for malformed grammar. " Original acc: 81.00%, attacked acc: 77.00%, dropped acc: 4.00% Original prompt: In the role of a syntax analyzer, examine the grammar of the provided sentence and decide if it is correct, answering with 'acceptable' for grammatically correct sentences or 'unacceptable' for incorrect ones. Attacked prompt: b"Across the role of a syntax analyzer, examine the dictionary of the provided chastisement and decide if it is adequate, answered with 'acceptable' for grammatically correct sentences or 'unacceptable' for deceptive ones. " Original acc: 81.00%, attacked acc: 76.00%, dropped acc: 5.00% ## translation Language: chinese, acc: 80.00%, prompt: As a tool for determining grammatical correctness, please review the following sentences and categorize them grammatically as 'acceptable' or 'unacceptable'. Language: chinese, acc: 80.00%, prompt: Please evaluate the grammar of the following sentences and mark them as 'acceptable' or 'unacceptable'. Language: chinese, acc: 80.00%, prompt: Please check that the following sentences are grammatically correct and categorize them as 'acceptable' or 'unacceptable'. Language: chinese, acc: 81.50%, prompt: Please determine whether the following sentences are 'acceptable' or 'unacceptable' based on grammatical correctness. Language: chinese, acc: 80.00%, prompt: Please categorize the grammar of the following sentences as 'acceptable' or 'unacceptable'. Language: chinese, acc: 80.50%, prompt: Please check the grammar of the following sentences and classify them as 'acceptable' or 'unacceptable'. Language: chinese, acc: 81.50%, prompt: Please mark the following sentences as 'acceptable' or 'unacceptable' according to the grammatical norms. Language: chinese, acc: 80.00%, prompt: Please assess the grammar of the following sentences and classify them as 'acceptable' or 'unacceptable'. Language: chinese, acc: 80.50%, prompt: Please mark the following sentences as 'acceptable' or 'unacceptable' for grammatical accuracy. Language: chinese, acc: 80.50%, prompt: Please check the grammar of the following sentences and classify them as 'acceptable' or 'unacceptable'. Language: french, acc: 79.00%, prompt: As a tool for assessing semantic relevance, categorize the sentence as 'acceptable' or 'unacceptable' according to its meaning and context. Language: french, acc: 79.50%, prompt: Determine whether the grammatical structure of the sentence is 'acceptable' or 'unacceptable' by carefully reviewing it. Language: french, acc: 79.50%, prompt: Judge whether the sentence follows the syntactic rules of the language and classify it as 'acceptable' or 'unacceptable'. Language: french, acc: 78.50%, prompt: Assess the relevance of the sentence by classifying it as 'acceptable' or 'unacceptable' according to the context in which it is used. Language: french, acc: 80.00%, prompt: As a semantic classification tool, decide whether the sentence is 'acceptable' or 'unacceptable' according to its meaning and consistency. Language: french, acc: 81.00%, prompt: Analyze the sentence carefully and determine whether it is grammatically 'acceptable' or 'unacceptable'. Language: french, acc: 80.50%, prompt: As a syntactic validation tool, determine whether the sentence is properly structured and categorize it as 'acceptable' or 'unacceptable'. Language: french, acc: 82.00%, prompt: When evaluating the grammar of the sentence, classify it as 'acceptable' or 'unacceptable' according to the rules of the French language. Language: french, acc: 79.50%, prompt: Determine whether the sentence is 'acceptable' or 'unacceptable' depending on its syntactic construction and semantic context. Language: french, acc: 79.00%, prompt: As a tool for linguistic analysis, categorize the sentence as 'acceptable' or 'unacceptable' according to its relevance and semantic consistency. Language: arabic, acc: 80.50%, prompt: As a tool for determining grammatical correctness, review the sentence and classify its rules as 'acceptable' or 'unacceptable'. Language: arabic, acc: 80.00%, prompt: Analyze the sentence carefully and classify its grammar between 'acceptable' or 'unacceptable'. Language: arabic, acc: 79.00%, prompt: As a tool for determining grammatical correctness, classify the rules of the sentence between 'acceptable' or 'unacceptable'. Language: arabic, acc: 80.00%, prompt: Evaluate the correctness of the sentence between 'acceptable' or 'unacceptable', as a tool for determining grammatical correctness. Language: arabic, acc: 80.00%, prompt: As a tool for classifying valid sentences, determine whether 'acceptable' or 'unacceptable' is based on its grammatical rules. Language: arabic, acc: 80.00%, prompt: Analyze the sentence carefully and classify its grammatical rules between 'acceptable' or 'unacceptable'. Language: arabic, acc: 78.50%, prompt: Analyze the sentence and classify it between 'acceptable' or 'unacceptable' as a grammatical check tool. Language: arabic, acc: 80.50%, prompt: As a classification tool for grammatical sentences, determine whether the sentence 'acceptable' or 'unacceptable' is based on its rules. Language: arabic, acc: 80.50%, prompt: As a tool for analyzing grammar, classify the rules of the sentence between 'acceptable' or 'unacceptable'. Language: arabic, acc: 78.50%, prompt: Determine whether the sentence is 'acceptable' or 'unacceptable' as a tool for determining grammatical correctness and analyze its grammar. Language: spanish, acc: 80.50%, prompt: As a tool to determine grammatical correctness, review the sentence and categorize its grammar as 'acceptable' or 'unacceptable'. Language: spanish, acc: 79.50%, prompt: Analyze the sentence carefully and classify its grammar as 'acceptable' or 'unacceptable'. Language: spanish, acc: 80.50%, prompt: As a tool for determining grammatical correctness, it categorizes the grammatical rules of the sentence as 'acceptable' or 'unacceptable'. Language: spanish, acc: 79.00%, prompt: Evaluate the grammatical correctness of the sentence and classify it as 'acceptable' or 'unacceptable' using a grammatical verification tool. Language: spanish, acc: 79.50%, prompt: As a tool for classifying grammatically correct sentences, determine whether the sentence is 'acceptable' or 'unacceptable' based on its grammatical rules. Language: spanish, acc: 81.00%, prompt: Analyse the sentence carefully and classify its grammar as 'acceptable' or 'unacceptable' using a grammatical revision tool. Language: spanish, acc: 79.50%, prompt: As a tool for classifying grammatical sentences, it determines whether the sentence is 'acceptable' or 'unacceptable' using its grammatical structure. Language: spanish, acc: 80.50%, prompt: As a tool for analyzing grammatical correctness, it categorizes the grammatical rules of the sentence as 'acceptable' or 'unacceptable'. Language: spanish, acc: 79.50%, prompt: Determine whether the sentence is 'acceptable' or 'unacceptable' using a grammatical verification tool and sort it accordingly. Language: spanish, acc: 78.50%, prompt: As a tool for determining grammatical correctness, evaluate the sentence and classify it as 'acceptable' or 'unacceptable' based on its grammatical rules. Language: japanese, acc: 80.00%, prompt: As a tool to determine whether grammar is grammatically correct, look at the sentence and categorize grammar into the 'acceptable' or 'unacceptable' categories. Language: japanese, acc: 79.00%, prompt: Please read the given sentence and categorize the grammar into the 'acceptable' or 'unacceptable' categories. Language: japanese, acc: 80.00%, prompt: To determine the correctness of a sentence, categorize grammar into two categories: 'acceptable' or 'unacceptable'. Language: japanese, acc: 79.50%, prompt: Classify the grammar of a given sentence into two categories: 'acceptable' and 'unacceptable'. Language: japanese, acc: 78.50%, prompt: Classify a given sentence into the categories 'acceptable' or 'unacceptable' to determine whether it is grammatically correct. Language: japanese, acc: 81.00%, prompt: To determine whether it is grammatically correct, categorize a given sentence into two categories: 'acceptable' or 'unacceptable'. Language: japanese, acc: 80.00%, prompt: To determine the correctness of grammar, categorize a given sentence into two categories: 'acceptable' or 'unacceptable'. Language: japanese, acc: 79.50%, prompt: Classify the grammar of a given sentence into two categories, 'acceptable' or 'unacceptable', and judge its accuracy. Language: japanese, acc: 79.00%, prompt: To determine whether it is grammatically correct, divide a given sentence into two categories: 'acceptable' or 'unacceptable'. Language: japanese, acc: 81.00%, prompt: To evaluate the accuracy of grammar, categorize a given sentence into two categories: 'acceptable' or 'unacceptable'. Language: korean, acc: 80.50%, prompt: As a tool for judging grammatical correctness, please review the sentences and classify the grammar as 'acceptable' or 'unacceptable'. Language: korean, acc: 79.50%, prompt: Please read the given sentences, and classify the grammar as 'acceptable' or 'unacceptable'. Language: korean, acc: 81.00%, prompt: Please classify the sentences as 'acceptable' or 'unacceptable' to judge the grammatical appropriateness. Language: korean, acc: 80.00%, prompt: Please classify the grammar of the given sentences into 2 categories: 'acceptable' or 'unacceptable'. Language: korean, acc: 80.00%, prompt: Please classify the given sentences as 'acceptable' or 'unacceptable' to judge the correctness of the grammar. Language: korean, acc: 79.50%, prompt: Please classify the given sentences as 'acceptable' or 'unacceptable' in order to judge their grammatical appropriateness. Language: korean, acc: 79.50%, prompt: Please determine the correctness of your grammar by classifying sentences as 'acceptable' or 'unacceptable'. Language: korean, acc: 68.50%, prompt: Classify the grammar of a given sentence as 'acceptable' or 'unacceptable', and judge its accuracy. Language: korean, acc: 80.00%, prompt: Please classify the given sentences as 'acceptable' or 'unacceptable' to judge the correctness of the grammar. Language: korean, acc: 78.00%, prompt: Please rate the accuracy of your grammar by categorizing sentences as 'acceptable' or 'unacceptable'. # mnli ## 10 prompts Acc: 78.50%, prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Acc: 78.50%, prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Acc: 78.00%, prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Acc: 77.50%, prompt: Analyze the two provided sentences and decide if their relationship is 'entailment', 'neutral', or 'contradiction'. Acc: 76.50%, prompt: Review the two statements and categorize their relationship as either 'entailment', 'neutral', or 'contradiction'. Acc: 76.50%, prompt: Determine if the given pair of sentences displays entailment, neutral, or contradiction. Respond with 'entailment', 'neutral', or 'contradiction'. Acc: 76.00%, prompt: Examine the pair of sentences and determine if they exhibit entailment, neutral, or contradiction. Answer with either 'entailment', 'neutral', or 'contradiction'. Acc: 75.50%, prompt: Does the relationship between the given sentences represent entailment, neutral, or contradiction? Respond with 'entailment', 'neutral', or 'contradiction'. Acc: 75.00%, prompt: Considering the two sentences, identify if their relationship is 'entailment', 'neutral', or 'contradiction'. Acc: 73.50%, prompt: Please classify the relationship between the provided sentences as 'entailment', 'neutral', or 'contradiction'. Acc: 75.50%, prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Acc: 74.00%, prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Acc: 74.00%, prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Acc: 74.00%, prompt: As an instrument for entailment evaluation, consider the two sentences and determine if their relationship is 'entailment', 'neutral', or 'contradiction'. Acc: 73.50%, prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment', 'neutral', or 'contradiction'. Acc: 73.50%, prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Acc: 73.00%, prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment', 'neutral', or 'contradiction'. Acc: 72.50%, prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Acc: 72.00%, prompt: Working as an entailment classifier, identify whether the given pair of sentences displays entailment, neutral, or contradiction. Respond with 'entailment', 'neutral', or 'contradiction'. Acc: 71.50%, prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment', 'neutral', or 'contradiction'. ## bertattack Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"and the connection between the proceeding sentences and classify it as 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.00%, attacked acc: 71.00%, dropped acc: 6.00% Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Indicate if the subject between the following present is 'entailment', 'neutral', or 'contradiction'. " Original acc: 78.00%, attacked acc: 72.00%, dropped acc: 6.00% Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Identify as the given complement of discourse demonstrates entailment, neutral, or contradiction. object with 'entailment', 'neutral', or 'contradiction'. " Original acc: 78.00%, attacked acc: 70.00%, dropped acc: 8.00% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"In the capacity of an entailment assessment system, find if the content between the following expressions is 'entailment', 'neutral', or 'contradiction'. " Original acc: 75.50%, attacked acc: 69.50%, dropped acc: 6.00% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"from as an entailment detection indicates, give if the given pair of conversation demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.00%, attacked acc: 68.50%, dropped acc: 5.50% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a tool for determining entailment relationships, examination the pair statements and colabeldetermine their connection as either 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.50%, attacked acc: 70.50%, dropped acc: 4.00% ## checklist Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction' ItGDrrA1Sc. " Original acc: 78.00%, attacked acc: 75.50%, dropped acc: 2.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction' RcRneWo6Iv. " Original acc: 77.50%, attacked acc: 74.50%, dropped acc: 3.00% Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction' lf8wBa2yEm. " Original acc: 78.00%, attacked acc: 75.00%, dropped acc: 3.00% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction' w52rwgo0Av. " Original acc: 75.50%, attacked acc: 72.50%, dropped acc: 3.00% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction' d6ZQ3u0GBQ. " Original acc: 74.00%, attacked acc: 71.50%, dropped acc: 2.50% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction' Vq3aBzuZcD. " Original acc: 74.00%, attacked acc: 71.50%, dropped acc: 2.50% ## deepwordbug Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Indicate if the conncetion between the follwoing sSntences is 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.50%, attacked acc: 73.50%, dropped acc: 4.00% Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Identify wehther the given pair of sentencs demosntrates entailment, neutral, or contradiction. nswer with 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.50%, attacked acc: 73.50%, dropped acc: 4.00% Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Assesj the connection between the following seUtences and clLassify it as 'entailment', 'neutral', or 'contradiction'. " Original acc: 78.00%, attacked acc: 72.50%, dropped acc: 5.50% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"In the capacity of an entailment assessment system, indimcate if the link between the followWng sentences is 'entailment', 'neutral', or 'contradiction'. " Original acc: 75.50%, attacked acc: 71.50%, dropped acc: 4.00% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a tool for determining entailment relationships, rkview the two statementR and categoirze their connection as either 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.50%, attacked acc: 71.50%, dropped acc: 3.00% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Atcing as an entailment detection instrument, determine if the giYven pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.00%, attacked acc: 70.50%, dropped acc: 3.50% ## stresstest Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction' and true is true and true is true and true is true and true is true and true is true . " Original acc: 78.00%, attacked acc: 75.00%, dropped acc: 3.00% Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction' and true is true and true is true and true is true and true is true and true is true . " Original acc: 78.00%, attacked acc: 75.50%, dropped acc: 2.50% Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction' and false is not true . " Original acc: 77.50%, attacked acc: 77.00%, dropped acc: 0.50% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction' and true is true . " Original acc: 75.00%, attacked acc: 73.50%, dropped acc: 1.50% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction' and true is true . " Original acc: 74.50%, attacked acc: 73.00%, dropped acc: 1.50% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction' and true is true and true is true and true is true and true is true and true is true . " Original acc: 74.50%, attacked acc: 73.00%, dropped acc: 1.50% ## textbugger Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Idetnify whether the given couple of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.00%, attacked acc: 75.00%, dropped acc: 2.00% Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Indicate if the connection between the flolowing s\xd0\xb5ntences is 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.50%, attacked acc: 74.50%, dropped acc: 3.00% Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Appraise the connection between the following punishments and categorize it as 'entailment', 'neutral', or 'contradiction'. " Original acc: 77.50%, attacked acc: 72.00%, dropped acc: 5.50% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"In the capacity of an entailment assessment system, indicating if the link between the hereafter sentences is 'entailment', 'neutral', or 'contradiction'. " Original acc: 75.50%, attacked acc: 71.50%, dropped acc: 4.00% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Acting as an entailment detection instrument, determ\xd1\x96ne if the given pair of sentence demonstrate entailment, neutral, or contradiction. Answe\xe2\xb2\x85 with 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.00%, attacked acc: 70.50%, dropped acc: 3.50% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a tool for determinin\xc9\xa1 entailment relationships, reviews the two statement\xd1\x95 and classification their connection as either 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.00%, attacked acc: 69.50%, dropped acc: 4.50% ## textfooler Original prompt: Indicate if the connection between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Showed if the connection between the following pena is 'entailment', 'neutral', or 'contradiction'. " Original acc: 78.00%, attacked acc: 71.00%, dropped acc: 7.00% Original prompt: Identify whether the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Identifies whether the offered pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. " Original acc: 78.50%, attacked acc: 75.00%, dropped acc: 3.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Assess the respecting between the subsequent sentences and rankings it as 'entailment', 'neutral', or 'contradiction'. " Original acc: 76.50%, attacked acc: 71.50%, dropped acc: 5.00% Original prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Throughout the capacity of an entailment assessment system, illustrates if the relations between the following sentences is 'entailment', 'neutral', or 'contradiction'. " Original acc: 75.50%, attacked acc: 70.50%, dropped acc: 5.00% Original prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Answer with 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment, neutral, or contradiction. Responded with 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.50%, attacked acc: 70.50%, dropped acc: 4.00% Original prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. Attacked prompt: b"As a devices for determining entailment affairs, review the two languages and categorize their connection as either 'entailment', 'neutral', or 'contradiction'. " Original acc: 74.00%, attacked acc: 70.00%, dropped acc: 4.00% ## translation Language: chinese, acc: 73.00%, prompt: As an implication analysis tool, evaluate the relationships between given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 74.00%, prompt: Analyze the relationships between the given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 75.00%, prompt: Evaluate the logical relationship between the following sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 74.50%, prompt: In your role as an implication analysis tool, judge the relationships of the following sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 74.00%, prompt: Please judge the relationship between the following sentences as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 74.50%, prompt: From a given sentence, evaluate the relationship between them and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 75.50%, prompt: Please classify the following sentences as 'entailment', 'neutral', or 'contradiction' according to their relationships. Language: chinese, acc: 74.50%, prompt: Evaluate the logical relationship between the following sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 73.00%, prompt: As an implication analysis tool, categorize the relationships between given sentences as 'entailment', 'neutral', or 'contradiction'. Language: chinese, acc: 77.50%, prompt: Please classify the following sentences as 'entailment', 'neutral', or 'contradiction' according to their logical relationship. Language: french, acc: 71.50%, prompt: As a tool for analyzing the consequence relationship, evaluate the relationship between the given sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 74.50%, prompt: Evaluate the relationship between the given sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 77.00%, prompt: Determine whether the following sentences are related to 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 72.00%, prompt: In your role as a consequence analysis tool, evaluate the relationship between the given sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 74.50%, prompt: Classify the relationship between the following sentences as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 71.50%, prompt: As a consequence analysis tool, evaluate the relationship between the given sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 77.50%, prompt: Analyze the relationship between the given sentences and determine whether it is of 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 75.00%, prompt: Evaluate the relationship between the following sentences and classify it as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 77.50%, prompt: As a tool for analyzing the consequence relationship, classify the following sentences as 'entailment', 'neutral', or 'contradiction'. Language: french, acc: 76.50%, prompt: Determine whether the given sentences are related to 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 76.00%, prompt: Based on your role as a reasoning analyst, analyze the relationship between the given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 73.00%, prompt: Evaluate the relationship between given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 74.50%, prompt: Determine if the following sentences are 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 72.00%, prompt: In your role as a tool of reasoning analysis, investigate the relationship between sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 75.00%, prompt: Classify the relationship between the following sentences as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 74.50%, prompt: In your role as a tool of reasoning analysis, evaluate the relationship between the given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 76.00%, prompt: Analyze the relationship between the given sentences and determine if they are 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 74.50%, prompt: Evaluate the relationship between the following sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 75.00%, prompt: In your role as a tool of reasoning analysis, the following sentences are classified as 'entailment', 'neutral', or 'contradiction'. Language: arabic, acc: 74.50%, prompt: Determine if the sentences given are 'entailment', 'neutral', or 'contradiction'. Language: spanish, acc: 71.00%, prompt: In your role as an implication analysis tool, evaluate the relationship between the given phrases and classify them as 'entailment', 'neutral', or 'contradiction'. Language: spanish, acc: 73.50%, prompt: Determine whether there is 'entailment', 'neutral', or 'contradiction' between the sentences given, using this text analysis tool, Language: spanish, acc: 73.50%, prompt: Analyze the relationship between the two sentences and classify it as 'entailment', 'neutral', or 'contradiction' using this text classification tool, Language: spanish, acc: 75.00%, prompt: Using this implication analysis tool, decide whether the sentences given are related by 'entailment', 'neutral', or 'contradiction'. Language: spanish, acc: 73.50%, prompt: Classifies the relationship between the given phrases as 'entailment', 'neutral', or 'contradiction' using this text analysis tool, Language: spanish, acc: 71.00%, prompt: Evaluate whether there is 'entailment', 'neutral', or 'contradiction' between the sentences provided using this text classification tool, Language: spanish, acc: 76.50%, prompt: Using this implication analysis tool, decide whether the two sentences are related by 'entailment', 'neutral', or 'contradiction'. Language: spanish, acc: 75.50%, prompt: Determine whether the given phrases are related by 'entailment', 'neutral', or 'contradiction' using this text analysis tool, Language: spanish, acc: 73.50%, prompt: Analyze the relationship between the two sentences and classify it as 'entailment', 'neutral', or 'contradiction' using this text analysis tool, Language: spanish, acc: 73.00%, prompt: Using this text classification tool, it classifies the relationship between the given phrases as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 71.50%, prompt: As your role as an implication analysis tool, evaluate the relationship of a given sentence and classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 71.50%, prompt: Use the implication analysis tool as your role to evaluate the relationship of a given sentence and classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 72.00%, prompt: Use this text classification tool to categorize relationships in a given text as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 73.00%, prompt: Use the implication analysis tool as your role and classify the relationship of a given sentence as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 73.50%, prompt: Evaluate the relationship of a given sentence and use this text classification tool to classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 74.50%, prompt: Evaluate the relationship of a given sentence and use this text classification tool to accurately classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 71.50%, prompt: Use the implication analysis tool as your role and use this text classification tool to classify the relationship of a given sentence as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 73.00%, prompt: Use this text classification tool to evaluate the relationship of a given sentence and classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 71.00%, prompt: Use the implication analysis tool as your role, evaluate the relationship of a given sentence, and use this text classification tool to classify it as 'entailment', 'neutral', or 'contradiction'. Language: japanese, acc: 73.00%, prompt: Use the implication analysis tool as your role and categorize the relationship of a given sentence strictly as 'entailment', 'neutral', or 'contradiction' using this text classification tool. Language: korean, acc: 73.50%, prompt: Analyze the relationships between given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 71.50%, prompt: In the text categorization task, identify the relationship between given sentences as one of 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 72.50%, prompt: Perform the role of analyzing the relationship between sentences and classifying them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 73.00%, prompt: Evaluate the relationship between two given sentences, and classify them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 72.50%, prompt: In the text categorization task, perform the role of classifying relationships between given sentences as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 73.50%, prompt: Judge the associations between sentences, and classify them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 74.50%, prompt: Analyze the relationship between two given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 70.50%, prompt: In the task of text classification, identify the relationships between given sentences as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 71.50%, prompt: Act as an instrument to evaluate the relationships between sentences, and classify them as 'entailment', 'neutral', or 'contradiction'. Language: korean, acc: 73.00%, prompt: Analyze the associations of two given sentences and classify them as 'entailment', 'neutral', or 'contradiction'. # mrpc ## 10 prompts Acc: 74.50%, prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Acc: 74.50%, prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Acc: 74.00%, prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Acc: 72.00%, prompt: Are the meanings of the following pair of sentences the same? Answer with 'equivalent' or 'not_equivalent'. Acc: 72.00%, prompt: Can the given sentences be considered semantically identical? Please reply with 'equivalent' or 'not_equivalent'. Acc: 72.00%, prompt: Determine if the meanings of the following sentences are semantically equivalent by responding with 'equivalent' or 'not_equivalent'. Acc: 71.00%, prompt: Do these two sentences have the same underlying meaning? Respond with 'equivalent' or 'not_equivalent'. Acc: 70.50%, prompt: Please analyze the provided sentences and indicate if their meanings are the same by choosing 'equivalent' or 'not_equivalent'. Acc: 68.50%, prompt: Evaluate whether the two provided sentences convey the same meaning by answering 'equivalent' or 'not_equivalent'. Acc: 67.00%, prompt: Do the meanings of these two statements align? Indicate your answer with 'equivalent' or 'not_equivalent'. Acc: 73.00%, prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Acc: 72.50%, prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Acc: 72.50%, prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Acc: 72.00%, prompt: As a sentence similarity evaluator, analyze the provided sentences and indicate if their meanings are 'equivalent' or 'not_equivalent'. Acc: 72.00%, prompt: As a textual similarity evaluator, analyze the following pair of sentences and determine if they are semantically 'equivalent' or 'not_equivalent'. Acc: 70.50%, prompt: As a language comparison expert, examine the given pair of sentences and decide if their meanings align, answering with 'equivalent' or 'not_equivalent'. Acc: 70.50%, prompt: In the role of a sentence comparison analyst, assess the provided sentences and indicate if they convey the same meaning by selecting 'equivalent' or 'not_equivalent'. Acc: 70.00%, prompt: As a semantic comparison expert, evaluate the given pair of sentences and determine if they are 'equivalent' or 'not_equivalent'. Acc: 65.00%, prompt: In your capacity as a language analyst, assess the following sentences and classify their similarity as 'equivalent' or 'not_equivalent'. Acc: 64.50%, prompt: In your capacity as a semantic assessment tool, evaluate the provided sentences and classify their meanings as 'equivalent' or 'not_equivalent'. ## bertattack Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Compare the following sentences and determine if they lack the same aspect marking by respectively with 'equivalent' or 'not_equivalent'. " Original acc: 74.00%, attacked acc: 47.50%, dropped acc: 26.50% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"investigate the pair of grammar and decide if their meanings are suitable by a with 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 61.50%, dropped acc: 11.50% Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"assess if the even the sentences have equivalent constructions by selecting 'equivalent' or 'not_equivalent'. " Original acc: 73.50%, attacked acc: 61.50%, dropped acc: 12.00% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"As a linguistic coparacomparison, review the follow pair of sentences and distinguish their semantic quality by choosing 'equivalent' or 'not_equivalent'. " Original acc: 72.50%, attacked acc: 57.50%, dropped acc: 15.00% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"during your capacity as a dictionary comparison tools, examine the given sentences and be if their forms are inconsistent, comparing with 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 57.50%, dropped acc: 15.50% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"have the role of a textual comparison academic, examine the given sentences and determine if they share the same length, responding with 'equivalent' or 'not_equivalent'. " Original acc: 71.50%, attacked acc: 47.00%, dropped acc: 24.50% ## checklist Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent' ofw9fEkN5R. " Original acc: 75.00%, attacked acc: 70.50%, dropped acc: 4.50% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent' AMsRIKZniY. " Original acc: 73.50%, attacked acc: 70.00%, dropped acc: 3.50% Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent' LkF0FZxMZ4. " Original acc: 74.00%, attacked acc: 68.50%, dropped acc: 5.50% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent' bhrRSokrfa. " Original acc: 72.00%, attacked acc: 68.00%, dropped acc: 4.00% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent' zq0DcZ5dnI. " Original acc: 73.00%, attacked acc: 71.00%, dropped acc: 2.00% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent' njCuciQJpB. " Original acc: 73.00%, attacked acc: 69.00%, dropped acc: 4.00% ## deepwordbug Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"Assess if the tNo given sentecnes have equivalent meaings by selecting 'equivalent' or 'not_equivalent'. " Original acc: 73.50%, attacked acc: 68.00%, dropped acc: 5.50% Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Comare the following sentences and determine if they sare the same seamntic meaning by vresponding with 'equivalent' or 'not_equivalent'. " Original acc: 75.00%, attacked acc: 68.50%, dropped acc: 6.50% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Examine the Kair of sentences and edcide if their meanijngs are ibentical by answering with 'equivalent' or 'not_equivalent'. " Original acc: 74.50%, attacked acc: 65.50%, dropped acc: 9.00% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"AV a linguistic comaprator, review the following pair of sentences and determine their semayntic equivalence by chhoosing 'equivalent' or 'not_equivalent'. " Original acc: 72.00%, attacked acc: 66.50%, dropped acc: 5.50% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In your capcity as a sematnic comparison tol, examine the given sentences and decde if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. " Original acc: 73.50%, attacked acc: 68.50%, dropped acc: 5.00% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Ic the role of a textual comparison specialist, examine the given sentences and decide if they share the same meanNng, respondTng with 'equivalent' or 'not_equivalent'. " Original acc: 72.50%, attacked acc: 67.00%, dropped acc: 5.50% ## stresstest Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent' and true is true and true is true and true is true and true is true and true is true . " Original acc: 74.00%, attacked acc: 72.50%, dropped acc: 1.50% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent' and false is not true . " Original acc: 74.50%, attacked acc: 73.00%, dropped acc: 1.50% Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent' and true is true and true is true and true is true and true is true and true is true . " Original acc: 73.50%, attacked acc: 71.50%, dropped acc: 2.00% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent' and true is true and true is true and true is true and true is true and true is true . " Original acc: 72.50%, attacked acc: 71.50%, dropped acc: 1.00% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent' and true is true . " Original acc: 72.50%, attacked acc: 73.00%, dropped acc: -0.50% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent' and true is true . " Original acc: 72.50%, attacked acc: 70.00%, dropped acc: 2.50% ## textbugger Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Compare the following sentences and determine if they share the same semantic \xef\xbd\x8deaning by respondi\xd5\xb8g with 'equivalent' or 'not_equivalent'. " Original acc: 75.00%, attacked acc: 69.00%, dropped acc: 6.00% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Examine the torque of sentences and decide if their meanings are selfsame by answering with 'equivalent' or 'not_equivalent'. " Original acc: 74.00%, attacked acc: 68.50%, dropped acc: 5.50% Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"Ass ess if the two gien sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 68.00%, dropped acc: 5.00% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"As a linguistic comparator, review the following torque of sentences and determine their semanti\xcf\xb2 e\xd4\x9buivalence by choosing 'equivalent' or 'not_equivalent'. " Original acc: 71.50%, attacked acc: 64.50%, dropped acc: 7.00% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In your capacity as a sema\xd5\xb8tic comparison t\xd0\xbeol, exaine the given sentences and decide if their meanings are iden\xf0\x9d\x9a\x9dical, responding with 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 68.50%, dropped acc: 4.50% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Throughout the role of a textual co\xef\xbd\x8dparison s\xd1\x80ecialist, examine the given se\xd5\xb8tences and decide if they share the same meaning, answering with 'equivalent' or 'not_equivalent'. " Original acc: 71.50%, attacked acc: 64.50%, dropped acc: 7.00% ## textfooler Original prompt: Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Compared the afterward sentences and establish if they communicated the same semantic meaning by answers with 'equivalent' or 'not_equivalent'. " Original acc: 74.00%, attacked acc: 66.50%, dropped acc: 7.50% Original prompt: Assess if the two given sentences have equivalent meanings by selecting 'equivalent' or 'not_equivalent'. Attacked prompt: b"Appraisal if the two accorded sentences enjoys equivalent connotation by choice 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 59.00%, dropped acc: 14.00% Original prompt: Examine the pair of sentences and decide if their meanings are identical by answering with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Analysis the couple of chastisement and decide if their connotation are same by reactions with 'equivalent' or 'not_equivalent'. " Original acc: 74.00%, attacked acc: 61.50%, dropped acc: 12.50% Original prompt: As a linguistic comparator, review the following pair of sentences and determine their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. Attacked prompt: b"Iike a linguistic comparator, revisiting the following torque of sentences and ascertain their semantic equivalence by choosing 'equivalent' or 'not_equivalent'. " Original acc: 72.00%, attacked acc: 63.50%, dropped acc: 8.50% Original prompt: In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In your capacity as a semantic comparison instruments, examine the given sorrows and decide if their meanings are similar, responses with 'equivalent' or 'not_equivalent'. " Original acc: 73.00%, attacked acc: 70.00%, dropped acc: 3.00% Original prompt: In the role of a textual comparison specialist, examine the given sentences and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"De the role of a textual comparisons scholar, revisited the given sentences and decide if they exchange the same feel, responding with 'equivalent' or 'not_equivalent'. " Original acc: 71.50%, attacked acc: 57.00%, dropped acc: 14.50% ## translation Language: chinese, acc: 72.00%, prompt: As an expert in semantic comparison, evaluates two given sentences to determine whether they are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 72.00%, prompt: From the point of view of semantic comparison, determine whether this pair of sentences is 'equivalent' or 'not_equivalent'. Language: chinese, acc: 70.50%, prompt: Using the semantic comparison method, determine whether the following two statements are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 72.50%, prompt: For the following two sentences, determine whether they are 'equivalent' or 'not_equivalent' based on semantic comparison. Language: chinese, acc: 69.00%, prompt: As an expert in semantic comparison, please evaluate the following two sentences and determine if they are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 70.00%, prompt: Using semantic comparison techniques, determine whether the following two sentences are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 69.00%, prompt: Please determine whether the following two sentences are 'equivalent' or 'not_equivalent' according to the standard of semantic comparison. Language: chinese, acc: 67.50%, prompt: As an expert in the field of semantic comparison, please evaluate the following two sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 68.50%, prompt: Using semantic comparison, determine whether the following two sentences are 'equivalent' or 'not_equivalent'. Language: chinese, acc: 73.00%, prompt: Determine whether the following two sentences are 'equivalent' or 'not_equivalent' based on semantic comparison. Language: french, acc: 68.00%, prompt: As an expert in semantic comparison, evaluate the following pair of sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: french, acc: 70.50%, prompt: Can you determine whether the following two sentences are 'equivalent' or 'not_equivalent' as a semantic comparison expert? Language: french, acc: 68.50%, prompt: Using your expertise in semantic comparison, determine whether the following two sentences are 'equivalent' or 'not_equivalent'. Language: french, acc: 71.50%, prompt: As a semantic comparison specialist, assess the similarity between the following two sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: french, acc: 69.50%, prompt: Are you able to determine whether the following two sentences are 'equivalent' or 'not_equivalent' as an expert in semantic comparison? Language: french, acc: 67.50%, prompt: As a semantic comparison professional, evaluate the following pair of sentences and indicate whether they are 'equivalent' or 'not_equivalent'. Language: french, acc: 70.50%, prompt: Can you determine whether the following two sentences have a 'equivalent' or 'not_equivalent' meaning as an expert in semantic comparison? Language: french, acc: 70.00%, prompt: As an expert in semantic comparison, assess the similarity between the following two sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: french, acc: 69.50%, prompt: Using your expertise in semantic comparison, determine whether the following two sentences are 'equivalent' or 'not_equivalent' in terms of meaning. Language: french, acc: 66.50%, prompt: As a semantic comparison professional, assess the similarity between the following two sentences and indicate whether they are 'equivalent' or 'not_equivalent'. Language: arabic, acc: 68.00%, prompt: As an expert in semantic comparison, evaluate the two given sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: arabic, acc: 71.50%, prompt: Based on my experience in semantic analysis, classify the following two sentences as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 66.00%, prompt: As an expert in semantic comparison, analyze the following two sentences and classify them as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 71.50%, prompt: Your task as an expert in semantic comparison is to evaluate the following two sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: arabic, acc: 70.00%, prompt: As a semantic comparison specialist, analyze the two data statements and insert them into one of the following categories: 'equivalent' or 'not_equivalent'. Language: arabic, acc: 65.00%, prompt: Based on my experience in semantic analysis, classify the following two sentences between 'equivalent' or 'not_equivalent'. Language: arabic, acc: 71.00%, prompt: Your role as a semantic comparison specialist requires analyzing the two given sentences and determining whether they are 'equivalent' or 'not_equivalent'. Language: arabic, acc: 70.50%, prompt: As an experienced semantic analyst, classify the following two sentences as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 73.50%, prompt: Your job as a semantic analyst evaluates the following two sentences as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 73.00%, prompt: As a semantic analyst, determine whether the given sentences are 'equivalent' or 'not_equivalent' based on their relationship. Language: spanish, acc: 72.00%, prompt: As an expert in semantic comparison, it evaluates the pair of sentences provided and determines whether they are 'equivalent' or 'not_equivalent'. Language: spanish, acc: 71.50%, prompt: Based on my experience in semantic analysis, classify the following two sentences as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 66.50%, prompt: As an expert in semantic comparison, analyze the two sentences given and classify them as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 68.50%, prompt: Your task as a semantic comparison specialist is to evaluate the following two sentences and determine whether they are 'equivalent' or 'not_equivalent'. Language: spanish, acc: 69.00%, prompt: As an expert in semantic analysis, he makes a classification of the following two sentences based on their 'equivalent' or 'not_equivalent'. Language: spanish, acc: 72.00%, prompt: Based on your experience of semantic comparison, classify the next two sentences as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 73.50%, prompt: As a specialist in semantic analysis, you are given the task of analysing the two sentences given and classifying them as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 69.00%, prompt: As an expert in semantic comparison, he classifies the following two sentences into 'equivalent' or 'not_equivalent'. Language: spanish, acc: 67.00%, prompt: As a specialist in semantic analysis, evaluate the following two sentences and classify them as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 72.00%, prompt: Your task as an expert in semantic comparison is to analyze the two sentences provided and determine whether they are 'equivalent' or 'not_equivalent' based on their semantic relationship. Language: japanese, acc: 67.00%, prompt: Evaluate whether a given pair of sentences is 'equivalent' or 'not_equivalent', depending on the context. Language: japanese, acc: 71.50%, prompt: Use a semantic comparison to determine whether a given pair of sentences is 'equivalent' or 'not_equivalent'. Language: japanese, acc: 73.50%, prompt: Evaluate a given pair of sentences as 'equivalent' or 'not_equivalent' by determining whether they have the same semantic meaning. Language: japanese, acc: 73.50%, prompt: Determine whether a given pair of sentences is synonyms and evaluate whether they are 'equivalent' or 'not_equivalent'. Language: japanese, acc: 74.50%, prompt: Determine whether a given pair of sentences is 'equivalent' or 'not_equivalent', and whether they are semantically identical. Language: japanese, acc: 70.00%, prompt: Determinate whether a given pair of sentences has the same meaning and evaluate whether they are 'equivalent' or 'not_equivalent'. Language: japanese, acc: 72.00%, prompt: Evaluate whether a given pair of sentences is 'equivalent' or 'not_equivalent' by determining whether they are semantically identical. Language: japanese, acc: 68.50%, prompt: Judge whether a given pair of sentences is equal and evaluate whether they are 'equivalent' or 'not_equivalent'. Language: japanese, acc: 73.00%, prompt: Determinate whether a given pair of sentences are semantically equal and evaluate whether they are 'equivalent' or 'not_equivalent'. Language: japanese, acc: 65.00%, prompt: Whether a given pair of sentences is 'equivalent' or 'not_equivalent' depends on the context. Language: korean, acc: 72.00%, prompt: As a sentence comparator, evaluate the two sentences given to determine 'equivalent' or 'not_equivalent'. Language: korean, acc: 71.00%, prompt: Compare two sentences to determine 'equivalent' or 'not_equivalent'. For this you need qualifications as a specialist in semantic comparison. Language: korean, acc: 71.00%, prompt: It takes your knowledge as an expert in semantic comparison to determine that two sentences are 'equivalent' or 'not_equivalent'. Language: korean, acc: 70.00%, prompt: As a specialist in semantic comparison, evaluate whether two given sentences are 'equivalent' or 'not_equivalent'. Language: korean, acc: 67.50%, prompt: Analyze two sentences to determine 'equivalent' or 'not_equivalent'. For that you need the knowledge of a semantic comparison expert. Language: korean, acc: 72.00%, prompt: As an expert in semantic comparison, decide whether two sentences are 'equivalent' or 'not_equivalent'. Language: korean, acc: 72.50%, prompt: It takes the knowledge of an expert in semantic comparison to compare two sentences to judge 'equivalent' or 'not_equivalent'. Language: korean, acc: 70.50%, prompt: Experience as an expert in semantic comparison is required to determine whether two given sentences are 'equivalent' or 'not_equivalent'. Language: korean, acc: 69.50%, prompt: As an expert in semantic comparison, determine whether two sentences are 'equivalent' or 'not_equivalent'. Language: korean, acc: 69.00%, prompt: Analyze two sentences to determine 'equivalent' or 'not_equivalent'. For this, you need a qualification as a specialist in semantic comparison. # qnli ## 10 prompts Acc: 84.00%, prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Acc: 84.00%, prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Acc: 84.00%, prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Acc: 83.50%, prompt: Based on the information in the context, decide if the answer to the question is justified by choosing 'entailment' or 'not_entailment'. Acc: 83.50%, prompt: Assess if the answer to the question can be logically concluded from the provided context by choosing 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: Based on the provided context and question, decide if the information supports the answer by responding with 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: Evaluate whether the given context supports the answer to the question by responding with 'entailment' or 'not_entailment'. Acc: 81.50%, prompt: Please assess if the answer to the question can be derived from the given context by selecting 'entailment' or 'not_entailment'. Acc: 81.00%, prompt: Analyze the context and question, and indicate if the context entails the answer by choosing 'entailment' or 'not_entailment'. Acc: 80.50%, prompt: Given the question and context provided, determine if the answer can be inferred by choosing 'entailment' or 'not_entailment'. Acc: 86.00%, prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Acc: 85.00%, prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Acc: 84.50%, prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: In the capacity of a language evaluator, examine if the given context justifies the answer to the question and indicate your assessment with 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: In the capacity of a language specialist, decide if the context presented contains enough information to infer the answer to the question and respond with 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: In your role as a linguistic investigator, determine if the context given entails the answer to the question and provide your conclusion with 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: As a language expert, assess if the given context entails the answer to the question and respond with 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: As a semantic interpreter, assess whether the provided context supports the answer to the given question and answer with 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: As a linguistic consultant, decide if the answer to the question is logically supported by the provided context and respond with 'entailment' or 'not_entailment'. Acc: 81.00%, prompt: In your role as a semantic evaluator, determine if the provided context justifies the answer to the question and answer with 'entailment' or 'not_entailment'. ## bertattack Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Examine the framing and question, and determine if the term logically implies the subject by selecting 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 73.00%, dropped acc: 12.00% Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"put the contexts and question, and be if the answer can be logically deduced from the question by stating with 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 71.00%, dropped acc: 13.00% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"write the full context and question, and decides if the tense contains sufficient information to support the answer by selects 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 71.00%, dropped acc: 13.00% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"according a textual analyst, ask if the situation situation logical: implies the object to the question and phrase your decision with 'entailment' or 'not_entailment'. " Original acc: 86.00%, attacked acc: 71.00%, dropped acc: 15.00% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"1 a semantic researcher, consideration whether the provided context supports the answer to the question and selects 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 78.00%, dropped acc: 5.50% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a structural inference expert, analyze if the answer to the challenge can be deduced from the query audience and select 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 70.00%, dropped acc: 14.00% ## checklist Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment' 9Q3S1F94fE. " Original acc: 84.50%, attacked acc: 80.50%, dropped acc: 4.00% Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment' 5XeflW1ZJc. " Original acc: 84.00%, attacked acc: 80.50%, dropped acc: 3.50% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment' CuVpbbkC4c. " Original acc: 83.50%, attacked acc: 82.50%, dropped acc: 1.00% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment' 6a4Yn3RGVc. " Original acc: 85.50%, attacked acc: 83.00%, dropped acc: 2.50% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment' Vq3aBzuZcD. " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment' YHQiDfFhw4. " Original acc: 84.50%, attacked acc: 81.00%, dropped acc: 3.50% ## deepwordbug Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Examine the contedt and question, and determine if the cXontext logically implies the answer by selecting 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 77.50%, dropped acc: 7.50% Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"Contsider the context and question, and indiate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 81.00%, dropped acc: 2.50% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Review the given cnotext and qjestion, and decid if the contexSt contains eonugh iDformation to support the answer by selecting 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 80.50%, dropped acc: 4.00% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"A a textual analyst, xamine if the gPven conSext logiclaly implies the asnwer to the qbuestion and indicate your decision with 'entailment' or 'not_entailment'. " Original acc: 86.00%, attacked acc: 75.00%, dropped acc: 11.00% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. " Original acc: 83.00%, attacked acc: 83.00%, dropped acc: 3.00% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual inference expert, analXyze if the answre to the questiou can be deduced from the provided contexK and select 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 80.00%, dropped acc: 3.50% ## stresstest Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 85.00%, attacked acc: 79.50%, dropped acc: 5.50% Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 84.00%, attacked acc: 80.00%, dropped acc: 4.00% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 85.50%, attacked acc: 82.50%, dropped acc: 3.00% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment' and false is not true . " Original acc: 84.00%, attacked acc: 81.50%, dropped acc: 2.50% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment' and true is true . " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% ## textbugger Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"Consider the context and question, and \xd1\x96ndicate if the answer can be logically deduc\xd0\xb5d from the backgrounds by responding with 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 77.00%, dropped acc: 7.00% Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Examine the cntext and question, and determine if the backdrop logically implies the answer by selecting 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 76.00%, dropped acc: 9.00% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 83.50%, dropped acc: 1.50% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual analys\xf0\x9d\x9a\x9d, ex\xc9\x91mine if the give\xd5\xb8 backdrop logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. " Original acc: 86.00%, attacked acc: 77.00%, dropped acc: 9.00% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"A\xd1\x95 a semantic researcher, evaluate si the gave framework supports the respond to the question and choose 'entailment' or 'not_entailment'. " Original acc: 83.00%, attacked acc: 73.50%, dropped acc: 9.50% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual innuendo expert, analyze if the answer to the question can be dedu ced from the provided con\xf0\x9d\x9a\x9dext and select 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 76.00%, dropped acc: 7.50% ## textfooler Original prompt: Consider the context and question, and indicate if the answer can be logically deduced from the context by responding with 'entailment' or 'not_entailment'. Attacked prompt: b"Envisage the context and theme, and displayed if the address can be reasonably deduce from the context by replies with 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 76.00%, dropped acc: 8.00% Original prompt: Review the given context and question, and decide if the context contains enough information to support the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Review the given context and question, and decide if the backgrounds composed enough info to favour the reacts by selecting 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 78.50%, dropped acc: 5.50% Original prompt: Examine the context and question, and determine if the context logically implies the answer by selecting 'entailment' or 'not_entailment'. Attacked prompt: b"Discuss the background and question, and determine if the framework intuitively presupposes the answer by akhtar 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 63.50%, dropped acc: 21.00% Original prompt: As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'. Attacked prompt: b"Iike a textual analyst, analyze if the submitted regard rationally implies the replies to the issuing and indicates your clout with 'entailment' or 'not_entailment'. " Original acc: 85.50%, attacked acc: 68.50%, dropped acc: 17.00% Original prompt: As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'. Attacked prompt: b"As a semantic researcher, measured whether the provided context assisting the answer to the tura and choose 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 77.00%, dropped acc: 7.00% Original prompt: As a textual inference expert, analyze if the answer to the question can be deduced from the provided context and select 'entailment' or 'not_entailment'. Attacked prompt: b"As a textual inference expert, analyze if the reacts to the question can become deducting from the provided backdrop and select 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 72.50%, dropped acc: 11.00% ## translation Language: chinese, acc: 81.00%, prompt: Determine whether the answer is implied or not based on the context. Answer 'entailment' or 'not_entailment'. Language: chinese, acc: 79.00%, prompt: As a language expert, determine whether the text contains the answer and choose 'entailment' or 'not_entailment'. Language: chinese, acc: 80.50%, prompt: Determine whether the text implies an answer, and select 'entailment' or 'not_entailment'. Language: chinese, acc: 81.50%, prompt: Given the text and the question, whether the answer is implied, select 'entailment' or 'not_entailment'. Language: chinese, acc: 83.00%, prompt: Determine whether the text contains the answer, depending on the context. Select 'entailment' or 'not_entailment'. Language: chinese, acc: 80.50%, prompt: As a language expert, determine whether a text can contain an answer, and say 'entailment' or 'not_entailment'. Language: chinese, acc: 81.50%, prompt: Please determine whether the text implies an answer. Answer 'entailment' or 'not_entailment'. Language: chinese, acc: 81.00%, prompt: Please select 'entailment' or 'not_entailment' based on the text and the question. Language: chinese, acc: 81.50%, prompt: Assess whether the answer is implied based on the context. Answer 'entailment' or 'not_entailment'. Language: chinese, acc: 81.50%, prompt: Please determine whether the text contains the answer and answer 'entailment' or 'not_entailment'. Language: french, acc: 86.00%, prompt: As a linguistic expert, assess whether the given context involves the answer to the question and answer with 'entailment' or 'not_entailment'. Language: french, acc: 84.00%, prompt: Determine whether the information provided in the context necessarily leads to the answer to the question asked and indicate 'entailment' or 'not_entailment'. Language: french, acc: 84.00%, prompt: Analyze the text to determine if the answer to the question is implied in the context and specify 'entailment' or 'not_entailment'. Language: french, acc: 82.00%, prompt: Based on the given context, decide whether the answer to the question is necessarily involved and mark 'entailment' or 'not_entailment'. Language: french, acc: 83.50%, prompt: Evaluate whether the answer to the question can be deduced from the given context and mark 'entailment' or 'not_entailment'. Language: french, acc: 82.00%, prompt: Discern whether the context provided directly involves the answer to the question and indicate 'entailment' or 'not_entailment'. Language: french, acc: 83.50%, prompt: Determine if the context contains enough information to involve the answer to the question and mark 'entailment' or 'not_entailment'. Language: french, acc: 82.50%, prompt: Assess whether the context provided necessarily leads to the answer to the question and answer with 'entailment' or 'not_entailment'. Language: french, acc: 84.00%, prompt: Analyze the text to determine if the answer to the question is involved in the context and indicate 'entailment' or 'not_entailment'. Language: french, acc: 84.50%, prompt: Based on the given context, decide whether the answer to the question is necessarily inferred and mark 'entailment' or 'not_entailment'. Language: arabic, acc: 81.50%, prompt: As a language expert, evaluate whether the given context calls for an answer and answer 'entailment' or 'not_entailment'. Language: arabic, acc: 79.50%, prompt: Judge the relationship between the text and the question and answer 'entailment' or 'not_entailment', depending on your language experience. Language: arabic, acc: 83.00%, prompt: Does the context given indicate the answer to the question? Evaluate and answer 'entailment' or 'not_entailment'. Language: arabic, acc: 84.00%, prompt: Based on your linguistic knowledge, does the text relate to the question? Answer 'entailment' or 'not_entailment'. Language: arabic, acc: 83.00%, prompt: As a language expert, determine how the text relates to the question and answer 'entailment' or 'not_entailment'. Language: arabic, acc: 80.00%, prompt: Does the text support the answer to the question? Answer 'entailment' or 'not_entailment', depending on your language experience. Language: arabic, acc: 71.50%, prompt: Check the text link to the question and answer 'entailment' or 'not_entailment', depending on your language skills. Language: arabic, acc: 78.50%, prompt: As a language expert, is there a link between the text and the question? Answer 'entailment' or 'not_entailment'. Language: arabic, acc: 81.00%, prompt: Based on your language experience, does context help to answer the question? Evaluate and answer 'entailment' or 'not_entailment'. Language: arabic, acc: 78.50%, prompt: Does the text give a clear answer to the question? Answer 'entailment' or 'not_entailment', depending on your language experience. Language: spanish, acc: 83.50%, prompt: As a language expert, evaluate whether the given context implies the answer to the question and answer with 'entailment' or 'not_entailment'. Language: spanish, acc: 83.00%, prompt: Determine whether the information given in the text necessarily implies the veracity of the hypothesis and answer 'entailment' or 'not_entailment'. Language: spanish, acc: 83.00%, prompt: Analyzes whether the information presented in the paragraph leads to the conclusion of the question and labels the answer as 'entailment' or 'not_entailment'. Language: spanish, acc: 84.00%, prompt: Indicates whether the information provided in the text is sufficient to conclude the statement and labels the response as 'entailment' or 'not_entailment'. Language: spanish, acc: 81.00%, prompt: As an expert on the subject, judge whether the information provided in the text justifies the claim and classify the answer as 'entailment' or 'not_entailment'. Language: spanish, acc: 82.00%, prompt: Evaluates whether the information in the paragraph necessarily supports the conclusion of the hypothesis and responds 'entailment' or 'not_entailment'. Language: spanish, acc: 82.00%, prompt: Determines whether the information presented in the text logically implies the answer to the question and labels the answer as 'entailment' or 'not_entailment'. Language: spanish, acc: 82.50%, prompt: Analyzes whether the information provided in the paragraph necessarily leads to the veracity of the hypothesis and classifies the response as 'entailment' or 'not_entailment'. Language: spanish, acc: 81.50%, prompt: As an expert on the subject, evaluate whether the information presented in the text supports the claim and respond 'entailment' or 'not_entailment'. Language: spanish, acc: 82.00%, prompt: Indicates whether the information provided in the paragraph necessarily implies the answer to the question and labels the answer as 'entailment' or 'not_entailment'. Language: japanese, acc: 80.50%, prompt: Rate whether the answer to the question is derived from the given context and answer with 'entailment' or 'not_entailment'. Language: japanese, acc: 80.50%, prompt: Please answer 'entailment' or 'not_entailment' for the given context and question. Language: japanese, acc: 84.00%, prompt: Decide whether the answer to the question is derived from the given context and answer 'entailment' or 'not_entailment'. Language: japanese, acc: 79.00%, prompt: Compare the question with the given context and give the answer 'entailment' or 'not_entailment'. Language: japanese, acc: 83.00%, prompt: Determinate whether the given context contains the answer to the question and answer with 'entailment' or 'not_entailment'. Language: japanese, acc: 79.00%, prompt: Estimate the answer of the question from the context and give the answer 'entailment' or 'not_entailment'. Language: japanese, acc: 83.50%, prompt: Determinate whether the given context is relevant to the question and answer with 'entailment' or 'not_entailment'. Language: japanese, acc: 82.50%, prompt: Determine whether the given context is relevant to the question and answer with 'entailment' or 'not_entailment'. Language: japanese, acc: 82.50%, prompt: Determinate whether the given context contains the answer to the question and answer 'entailment' or 'not_entailment'. Language: japanese, acc: 79.00%, prompt: Answer with 'entailment' or 'not_entailment', inferring from the given context. Language: korean, acc: 81.50%, prompt: Determine if a given sentence necessarily implies the meaning of another sentence and answer 'entailment' or 'not_entailment'. Language: korean, acc: 80.50%, prompt: By understanding the relations between sentences, judge whether a given sentence necessarily refers to another sentence and answer with 'entailment' or 'not_entailment'. Language: korean, acc: 83.50%, prompt: Evaluate whether a given text necessarily indicates the meaning of another text and respond with 'entailment' or 'not_entailment'. Language: korean, acc: 81.50%, prompt: Understand the relations of a sentence, to determine whether a given sentence necessarily includes other sentences and answer with 'entailment' or 'not_entailment'. Language: korean, acc: 80.00%, prompt: Judge whether a given content necessarily implies the meaning of another content and answer with 'entailment' or 'not_entailment'. Language: korean, acc: 81.00%, prompt: Grasp the relations between sentences, determine if a given sentence necessarily contains the meaning of another sentence and respond with 'entailment' or 'not_entailment'. Language: korean, acc: 78.50%, prompt: Evaluate whether a given text necessarily refers to another text and answer with 'entailment' or 'not_entailment'. Language: korean, acc: 82.50%, prompt: By comparing the meaning of the sentences, to determine if a given sentence necessarily implies another sentence and answer 'entailment' or 'not_entailment'. Language: korean, acc: 79.50%, prompt: Evaluate whether the contents given necessarily refer to other contents and answer with 'entailment' or 'not_entailment'. Language: korean, acc: 83.50%, prompt: By analyzing the relations between sentences, determine if a given sentence does not necessarily include other sentences and answer with 'entailment' or 'not_entailment'. # qqp ## 10 prompts Acc: 83.50%, prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Acc: 83.00%, prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Acc: 82.50%, prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Acc: 82.50%, prompt: Please indicate if the following pair of sentences share the same meaning by responding with 'equivalent' or 'not_equivalent'. Acc: 82.00%, prompt: Do the following expressions mean the same thing? Provide your answer as 'equivalent' or 'not_equivalent'. Acc: 81.50%, prompt: Assess whether the following statements are identical in meaning by answering 'equivalent' or 'not_equivalent'. Acc: 81.50%, prompt: Are the meanings of these two phrases the same? Reply with 'equivalent' or 'not_equivalent'. Acc: 81.50%, prompt: Examine the following expressions and tell me if they are alike in meaning by using 'equivalent' or 'not_equivalent'. Acc: 81.00%, prompt: Evaluate whether these two phrases have identical meanings and respond with 'equivalent' or 'not_equivalent'. Acc: 80.50%, prompt: Do these two sentences convey the same meaning? Indicate with 'equivalent' or 'not_equivalent'. Acc: 80.50%, prompt: Analyze if the given set of sentences have the same connotation by answering with 'equivalent' or 'not_equivalent'. Acc: 84.00%, prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Acc: 83.50%, prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Acc: 83.00%, prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Acc: 82.50%, prompt: As a question equivalence detection system, examine the provided questions and respond with 'equivalent' if they are the same in meaning, or 'not_equivalent' if they are different. Acc: 82.50%, prompt: Functioning as a question similarity evaluation tool, analyze the given questions and decide if they share the same meaning, responding with 'equivalent' or 'not_equivalent'. Acc: 82.50%, prompt: As a tool for determining question equivalence, review the questions and categorize their similarity as either 'equivalent' or 'not_equivalent'. Acc: 82.50%, prompt: Working as a question classifier, identify whether the given questions share the same meaning, and respond with 'equivalent' for equivalent questions or 'not_equivalent' for different ones. Acc: 82.00%, prompt: In your role as a question comparison tool, assess the following pair of questions and classify them as 'equivalent' or 'not_equivalent'. Acc: 82.00%, prompt: While performing question comparison analysis, classify the similarity of the following questions as 'equivalent' for equivalent questions or 'not_equivalent' for different questions. Acc: 81.00%, prompt: In the role of a question similarity analyzer, examine the meaning of the provided questions and decide if they are equivalent, answering with 'equivalent' for equivalent questions or 'not_equivalent' for different questions. ## bertattack Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the each pair of products can be considered the same by responding with 'equivalent' or 'not_equivalent'. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Are the following versus answers equivalent or not? Answer me with "equivalent" or "not_equivalent". ' Original acc: 82.50%, attacked acc: 78.50%, dropped acc: 4.00% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Can these two implication be considered equal in meaning? follow with 'equivalent' or 'not_equivalent'. " Original acc: 81.50%, attacked acc: 79.50%, dropped acc: 2.00% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the capacity of a question assessment approach, indicate if the meaning of the provided answers is the same, responding with 'equivalent' or 'not_equivalent'. " Original acc: 84.00%, attacked acc: 80.00%, dropped acc: 4.00% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Acting as a question equivalence instrument, determine if the provided questions are equivalent in definitions, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"and an instrument for question comparison analysis, consider the questions and establish if their topic is the same, responding with 'equivalent' for matching responses or 'not_equivalent' for between questions. " Original acc: 83.50%, attacked acc: 78.00%, dropped acc: 5.50% ## checklist Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent' MSAddJ4D2a. " Original acc: 83.50%, attacked acc: 82.00%, dropped acc: 1.50% Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent IhuBIhoPGc". ' Original acc: 82.50%, attacked acc: 81.00%, dropped acc: 1.50% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent' xikCjkMydH. " Original acc: 82.00%, attacked acc: 79.00%, dropped acc: 3.00% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones ItGDrrA1Sc. " Original acc: 83.00%, attacked acc: 82.00%, dropped acc: 1.00% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions Cev0s49fbe. " Original acc: 83.50%, attacked acc: 82.00%, dropped acc: 1.50% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent' 5XeflW1ZJc. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% ## deepwordbug Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Aer the following two qtuestions equivalent or not? Answer me with "equivalent" or "not_equivalent". ' Original acc: 83.00%, attacked acc: 82.00%, dropped acc: 1.00% Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the gvien paiv of statementl can be considered the same by responding with 'equivalent' or 'not_equivalent'. " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Can these two statements be considJred equal in meaning? Answer with 'equivalent' or 'not_equivalent'. " Original acc: 82.00%, attacked acc: 80.00%, dropped acc: 2.00% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Actinm as a question equivalence instrumetn, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. " Original acc: 82.00%, attacked acc: 82.00%, dropped acc: 1.00% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"vAs an instrumenjt for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. " Original acc: 82.50%, attacked acc: 81.50%, dropped acc: 1.00% ## stresstest Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent and true is true ". ' Original acc: 83.00%, attacked acc: 82.50%, dropped acc: 0.50% Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent' and true is true and true is true and true is true and true is true and true is true . " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent' and false is not true . " Original acc: 81.50%, attacked acc: 81.00%, dropped acc: 0.50% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones and false is not true . " Original acc: 83.00%, attacked acc: 83.00%, dropped acc: 0.00% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions and true is true and true is true and true is true and true is true and true is true . " Original acc: 83.00%, attacked acc: 82.00%, dropped acc: 1.00% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent' and false is not true . " Original acc: 83.50%, attacked acc: 82.50%, dropped acc: 1.00% ## textbugger Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Ar\xd0\xb5 the following both questions equivalent or not? Answer me with "equivalent" or "not_equivalent". ' Original acc: 82.50%, attacked acc: 81.00%, dropped acc: 1.50% Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the given pair of speeches can be c\xd0\xbensidered the same by responding with 'equivalent' or 'not_equivalent'. " Original acc: 83.50%, attacked acc: 81.00%, dropped acc: 2.50% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Can these two st atements be considered equal in meaning? Ans\xd4\x9der with 'equivalent' or 'not_equivalent'. " Original acc: 82.50%, attacked acc: 80.00%, dropped acc: 2.50% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the abilities of a question \xc9\x91ssessment system, indicate if the meaning of the provided matters is the same, responding with 'equivalent' or 'not_equivalent'. " Original acc: 82.50%, attacked acc: 80.50%, dropped acc: 2.00% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"As an instr ument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for si\xef\xbd\x8dilar questions or 'not_equivalent' for different questions. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Acting as a question equivalence instrument, determine if the provided question\xd1\x95 are equivalent in meaning, answ\xd0\xb5ring with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. " Original acc: 83.00%, attacked acc: 81.00%, dropped acc: 2.00% ## textfooler Original prompt: Determine if the given pair of statements can be considered the same by responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Determine if the given pair of representations can constitute felt the same by responding with 'equivalent' or 'not_equivalent'. " Original acc: 83.50%, attacked acc: 80.00%, dropped acc: 3.50% Original prompt: Are the following two questions equivalent or not? Answer me with "equivalent" or "not_equivalent". Attacked prompt: b'Are the suite two things equivalent or not? Answer me with "equivalent" or "not_equivalent". ' Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: Can these two statements be considered equal in meaning? Answer with 'equivalent' or 'not_equivalent'. Attacked prompt: b"Kan these two affirmations be examined equal in meaning? Meet with 'equivalent' or 'not_equivalent'. " Original acc: 83.00%, attacked acc: 80.00%, dropped acc: 3.00% Original prompt: In the capacity of a question assessment system, indicate if the meaning of the provided questions is the same, responding with 'equivalent' or 'not_equivalent'. Attacked prompt: b"In the capacity of a question assessment system, indicate if the meaning of the provided concerns is the same, respond with 'equivalent' or 'not_equivalent'. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: As an instrument for question comparison evaluation, consider the questions and determine if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. Attacked prompt: b"As an paraphernalia for question comparison evaluation, consider the questions and pinpoint if their meaning is the same, responding with 'equivalent' for similar questions or 'not_equivalent' for different questions. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones. Attacked prompt: b"Interim as a issues equivalence instrument, determine if the provided aspects are equivalent in meaning, answering with 'equivalent' for selfsame questions or 'not_equivalent' for dissimilar ones. " Original acc: 83.00%, attacked acc: 80.00%, dropped acc: 3.00% ## translation Language: chinese, acc: 82.00%, prompt: As a tool for determining the equivalence of problems, the problems are examined and their similarity is classified as 'equivalent' or 'not_equivalent'. Language: chinese, acc: 81.00%, prompt: Please compare the similarity of two sentences and put them into the category of 'equivalent' or 'not_equivalent'. Language: chinese, acc: 82.50%, prompt: Two sentences are classified as 'equivalent' or 'not_equivalent' for their similarity. Language: chinese, acc: 83.00%, prompt: You can determine how similar the questions are by comparing them and categorizing them as 'equivalent' or 'not_equivalent'. Language: chinese, acc: 82.00%, prompt: Using the method of contrast, the similarity of these problems is divided into two categories: 'equivalent' or 'not_equivalent'. Language: chinese, acc: 83.00%, prompt: By comparing these issues, you can classify them as 'equivalent' or 'not_equivalent'. Language: chinese, acc: 83.00%, prompt: To determine whether the questions are similar, put them into the category of 'equivalent' or 'not_equivalent'. Language: chinese, acc: 82.00%, prompt: Divide the similarity of these questions into 'equivalent' or 'not_equivalent' categories. Language: chinese, acc: 82.50%, prompt: Using the similarity assessment tool, these questions were classified as 'equivalent' or 'not_equivalent'. Language: chinese, acc: 80.50%, prompt: By analyzing the similarity of these problems, they are divided into categories of 'equivalent' or 'not_equivalent'. Language: french, acc: 82.50%, prompt: As a tool to determine the equivalence of questions, review the questions and rank their similarity as 'equivalent' or 'not_equivalent'. Language: french, acc: 82.50%, prompt: Please compare the similarity of two sentences and classify them as 'equivalent' or 'not_equivalent'. Language: french, acc: 83.00%, prompt: Based on the similarity of two sentences, classify them as 'equivalent' or 'not_equivalent'. Language: french, acc: 83.00%, prompt: You can determine the similarity between these questions by comparing them and classifying them as 'equivalent' or 'not_equivalent'. Language: french, acc: 82.50%, prompt: Use a comparative method to divide the similarity of these questions into two categories: 'equivalent' or 'not_equivalent'. Language: french, acc: 83.50%, prompt: By comparing these questions, you can classify them as 'equivalent' or 'not_equivalent'. Language: french, acc: 82.50%, prompt: Determine whether these questions are similar or not, and then classify them as 'equivalent' or 'not_equivalent'. Language: french, acc: 83.00%, prompt: Divide the similarity of these questions into two categories: 'equivalent' or 'not_equivalent'. Language: french, acc: 83.00%, prompt: Use a similarity assessment tool to classify these questions as 'equivalent' or 'not_equivalent'. Language: french, acc: 82.50%, prompt: By analyzing the similarity of these questions, you can divide them into two categories: 'equivalent' or 'not_equivalent'. Language: arabic, acc: 83.00%, prompt: As a tool for determining an equation of questions, review the questions and classify their similarity as either 'equivalent' or 'not_equivalent'. Language: arabic, acc: 82.50%, prompt: When using questions in the classification domain, please classify the similarity between the questions as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 83.00%, prompt: To determine an equation of questions, you must review the questions and classify their similarity as 'equivalent' or 'not_equivalent'. Language: arabic, acc: 83.50%, prompt: Questions can be classified as 'equivalent' or 'not_equivalent' when used to identify classifications. Language: arabic, acc: 82.50%, prompt: Classification of question similarity as 'equivalent' or 'not_equivalent' is used as a tool to determine the classification of questions. Language: arabic, acc: 84.00%, prompt: Classify the similarity of the questions as 'equivalent' or 'not_equivalent' to determine the equation of the questions. Language: arabic, acc: 80.50%, prompt: Identifying the similarity of questions and classifying them as 'equivalent' or 'not_equivalent' is an important tool in determining the classification of questions. Language: arabic, acc: 83.00%, prompt: When classifying questions, their similarity can be classified as 'equivalent' or 'not_equivalent' to determine the correct classification. Language: arabic, acc: 82.50%, prompt: The similarity of questions should be classified as 'equivalent' or 'not_equivalent' when used to determine the equation of questions. Language: arabic, acc: 80.50%, prompt: Identifying the similarity of questions and classifying them as 'equivalent' or 'not_equivalent' helps to correctly classify questions. Language: spanish, acc: 81.50%, prompt: As a tool to determine the equivalence of questions, it reviews the questions and classifies their similarity as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 82.50%, prompt: Evaluate the similarity between questions and classify them as 'equivalent' or 'not_equivalent' to determine their equivalence. Language: spanish, acc: 83.50%, prompt: Determine whether two questions are 'equivalent' or 'not_equivalent' based on similarity and characteristics. Language: spanish, acc: 82.50%, prompt: Classifies the similarity between questions as 'equivalent' or 'not_equivalent' to determine their equivalence. Language: spanish, acc: 83.00%, prompt: Review the questions and rate them as 'equivalent' or 'not_equivalent' based on their similarity and content. Language: spanish, acc: 80.50%, prompt: As part of the classification task of questions, it determines their equivalence by categorizing their similarity as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 83.00%, prompt: Analyze the similarity between questions and classify them as 'equivalent' or 'not_equivalent' to determine their equivalence. Language: spanish, acc: 81.50%, prompt: As a method of identifying the equivalence of questions, it categorizes their similarity as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 82.50%, prompt: To determine the equivalence between questions, check their similarity and classify them as 'equivalent' or 'not_equivalent'. Language: spanish, acc: 82.00%, prompt: Classify the similarity between questions as 'equivalent' or 'not_equivalent' to determine whether they are equivalent or not. Language: japanese, acc: 84.00%, prompt: As a tool to determine the equivalence of the question, review the question and categorize its similarities into 'equivalent' or 'not_equivalent' categories. Language: japanese, acc: 82.50%, prompt: Work on text sorting tasks labeled 'equivalent' or 'not_equivalent'. Language: japanese, acc: 82.50%, prompt: For text classification tasks, use the labels 'equivalent' or 'not_equivalent' to determine the equivalence of statements. Language: japanese, acc: 81.50%, prompt: In the MRPC dataset, use the labels 'equivalent' or 'not_equivalent' to classify the equivalence of statements. Language: japanese, acc: 82.00%, prompt: As a tool for determining equivalence, check sentences and categorize them into 'equivalent' or 'not_equivalent' categories. Language: japanese, acc: 82.50%, prompt: Use the labels 'equivalent' or 'not_equivalent' to determine the equivalence of statements in text classification tasks. Language: japanese, acc: 81.50%, prompt: In the text classification task of the MRPC data set, classify the equivalence of statements with labels of 'equivalent' or 'not_equivalent'. Language: japanese, acc: 82.50%, prompt: As a tool to determine the equivalence of statements, categorize statements into 'equivalent' or 'not_equivalent' categories. Language: japanese, acc: 83.50%, prompt: In a text classification task, classify the equivalence of statements using labels of 'equivalent' or 'not_equivalent'. Language: japanese, acc: 81.50%, prompt: Do a text classification task to determine the equivalence of statements, labeled 'equivalent' or 'not_equivalent'. Language: korean, acc: 83.00%, prompt: Classify two given sentences as 'equivalent' or 'not_equivalent' by discriminating whether they have the same meaning. Language: korean, acc: 82.50%, prompt: Determine sentence equivalence by judging the similarity of two sentences with 'equivalent' or 'not_equivalent'. Language: korean, acc: 83.00%, prompt: Classify the similarity of sentences as 'equivalent' or 'not_equivalent' by judging whether two sentences have the same meaning. Language: korean, acc: 83.00%, prompt: Determine if two given sentences are equivalent to each other, and classify their similarity as 'equivalent' or 'not_equivalent'. Language: korean, acc: 81.50%, prompt: Compare two given sentences to determine sentence equivalence, and classify their similarities as 'equivalent' or 'not_equivalent'. Language: korean, acc: 82.50%, prompt: Classify sentence equivalence as 'equivalent' or 'not_equivalent' by judging whether two sentences have the same meaning to each other. Language: korean, acc: 82.00%, prompt: Determine if two sentences have the same meaning, and classify their similarities as 'equivalent' or 'not_equivalent'. Language: korean, acc: 80.50%, prompt: Compare two given sentences to determine their equivalence, and classify their similarities as 'equivalent' or 'not_equivalent'. Language: korean, acc: 83.50%, prompt: Review two sentences to evaluate sentence equivalence, and classify their similarities as 'equivalent' or 'not_equivalent'. Language: korean, acc: 82.50%, prompt: Judge whether two sentences have the same meaning to each other, and determine the sentence equivalence with 'equivalent' or 'not_equivalent'. # rte ## 10 prompts Acc: 84.50%, prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Acc: 84.50%, prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Acc: 83.50%, prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Acc: 82.50%, prompt: Identify whether the given pair of sentences demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: Please classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Analyze the two provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Determine if the given pair of sentences displays entailment or not_entailment. Respond with 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Considering the two sentences, identify if their relationship is 'entailment' or 'not_entailment'. Acc: 80.50%, prompt: Examine the pair of sentences and determine if they exhibit entailment or not_entailment. Answer with either 'entailment' or 'not_entailment'. Acc: 84.00%, prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Acc: 83.50%, prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment' or 'not_entailment'. Acc: 83.00%, prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: As an instrument for entailment evaluation, consider the two sentences and determine if their relationship is 'entailment' or 'not_entailment'. Respond with 'entailment' or 'not_entailment'. Acc: 82.50%, prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Acc: 82.00%, prompt: Working as an entailment classifier, identify whether the given pair of sentences displays entailment or not_entailment. Respond with 'entailment' or 'not_entailment'. ## bertattack Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Does the phrase between the the sentences represent entailment or not_entailment? be with 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 79.50%, dropped acc: 5.00% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"write the second statements and categorize their relationship as neither 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 59.50%, dropped acc: 25.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"t the connecting between the following sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 80.50%, dropped acc: 3.50% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis leader, scale the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 80.00%, dropped acc: 4.50% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment detection system, examine the connection between the following sentences and object with 'entailment' or 'not_entailment'. " Original acc: 83.00%, attacked acc: 79.50%, dropped acc: 3.50% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"was performing entailment analysis, classify the relationship between the first phrase as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 79.00%, dropped acc: 4.50% ## checklist Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment' 5XeflW1ZJc. " Original acc: 84.00%, attacked acc: 81.50%, dropped acc: 2.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment' PDCGfz6DL9. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment' 5XeflW1ZJc. " Original acc: 84.50%, attacked acc: 80.50%, dropped acc: 4.00% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment' bhrRSokrfa. " Original acc: 84.00%, attacked acc: 80.50%, dropped acc: 3.50% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment' LkF0FZxMZ4. " Original acc: 83.00%, attacked acc: 80.50%, dropped acc: 2.50% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment' hXLpw3bbiw. " Original acc: 83.00%, attacked acc: 81.00%, dropped acc: 2.00% ## deepwordbug Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Does the relationshBp between the given sentences repersent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 80.50%, dropped acc: 4.50% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Revew the Mtwo statemetns and categorize their relationsihp as either 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 81.00%, dropped acc: 4.00% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"Assss the connection between the followin sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 81.00%, dropped acc: 2.50% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis toIol, aZssess the relationship between the given sentences and cassify it as 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 81.00%, dropped acc: 3.50% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. " Original acc: 82.50%, attacked acc: 82.50%, dropped acc: 2.00% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"WhRle peGrforming entailment analysis, classify the relatioPship between the provided sentences as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 81.00%, dropped acc: 2.50% ## stresstest Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment' and false is not true . " Original acc: 85.00%, attacked acc: 82.00%, dropped acc: 3.00% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment' and false is not true . " Original acc: 84.50%, attacked acc: 84.00%, dropped acc: 0.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment' and true is true . " Original acc: 83.00%, attacked acc: 82.00%, dropped acc: 1.00% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 84.00%, attacked acc: 84.00%, dropped acc: 0.00% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment' and false is not true . " Original acc: 83.00%, attacked acc: 83.00%, dropped acc: 0.00% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 83.50%, attacked acc: 83.00%, dropped acc: 0.50% ## textbugger Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Does the realtionship between the given sentences \xe2\xb2\x85epresent entailment or not_entailment? Responding with 'entailment' or 'not_entailment'. " Original acc: 85.00%, attacked acc: 81.00%, dropped acc: 4.00% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two declarations and categorize their relationship as neither 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 76.50%, dropped acc: 7.50% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"Assess the connection between the following verdicts and classify it as 'entailment' or 'not_entailment'. " Original acc: 83.00%, attacked acc: 81.50%, dropped acc: 1.50% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relationships between the given sentences and classifying it as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"Whilst performing entailment analysis, classify the relatio\xd5\xb8ship between the proivded sentences as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment identification system, eamine the connection between the following sentences and re\xd1\x95pond with 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 80.00%, dropped acc: 3.50% ## textfooler Original prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Attacked prompt: b"Do the coitus between the given sentences comprised entailment or not_entailment? Meet with 'entailment' or 'not_entailment'. " Original acc: 84.50%, attacked acc: 79.50%, dropped acc: 5.00% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Consideration the two sermons and categorize their relationship as either 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 80.00%, dropped acc: 4.00% Original prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"Assess the ties between the later sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 80.50%, dropped acc: 3.00% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, appraisal the coitus between the given sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 84.00%, attacked acc: 80.00%, dropped acc: 4.00% Original prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Attacked prompt: b"During performing entailment analysis, classify the relationship between the awarded sentences as 'entailment' or 'not_entailment'. " Original acc: 83.50%, attacked acc: 81.50%, dropped acc: 2.00% Original prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Attacked prompt: b"As an entailment identification system, revisited the connection between the following sentences and respondent with 'entailment' or 'not_entailment'. " Original acc: 83.00%, attacked acc: 79.50%, dropped acc: 3.50% ## translation Language: chinese, acc: 83.50%, prompt: In the light of an implication analysis tool, evaluate the relationship between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 83.00%, prompt: From the perspective of an implication analysis tool, determine whether there is an implication relationship in the following sentences by classifying them as 'entailment' or 'not_entailment'. Language: chinese, acc: 83.00%, prompt: Please use an implication analysis tool to determine whether an implication relationship exists in the following sentences by classifying them as 'entailment' or 'not_entailment'. Language: chinese, acc: 86.00%, prompt: Please evaluate the relation of the following sentences as 'entailment' or 'not_entailment' for the purpose of determining implication relation. Language: chinese, acc: 83.50%, prompt: Please use the implication analysis tool to evaluate the relationships between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 84.50%, prompt: For the purpose of determining implicative relations, analyze the relations of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 84.00%, prompt: Please use the implication analysis tool to determine the relationship of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 83.00%, prompt: Please use the implication judgment tool to assess the relevance of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 83.00%, prompt: Please, with implication analysis as the main task, determine the relationships between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 84.00%, prompt: Using the implication judgment as a criterion, analyze the relation of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: french, acc: 83.00%, prompt: As an engagement analysis tool, evaluate the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Language: french, acc: 82.00%, prompt: Determine whether the given sentences involve one another or not as an implication analysis tool. Classify them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 84.50%, prompt: Using implication analysis, evaluate whether the sentences provided have a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 83.50%, prompt: As an engagement assessment tool, determine whether the sentences provided have a logical relationship and classify them as 'entailment' or 'not_entailment'. Language: french, acc: 83.00%, prompt: As an implication classification tool, analyze the sentences provided to determine if there is a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 81.50%, prompt: Using implication analysis, determine whether the given sentences have a cause-effect relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 84.00%, prompt: Evaluate the relationship between the given sentences using implication analysis and rank them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 83.50%, prompt: As an engagement detection tool, determine whether the given sentences have a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 82.00%, prompt: Using implication analysis, evaluate whether the sentences provided have a cause-effect relationship and rank them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 80.00%, prompt: Determine whether the given sentences have a cause-effect relationship as an engagement analysis tool and categorize them as 'entailment' or 'not_entailment'. Language: arabic, acc: 82.50%, prompt: In your role as a tool for reasoning analysis, evaluate the relationship between given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 84.00%, prompt: Can you determine whether this sentence is inferred from the other sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 84.50%, prompt: Using the tool of reasoning analysis, analyze the relationship between given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 82.00%, prompt: Does this sentence represent a conclusion from the previous sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 85.00%, prompt: As a tool of reasoning analysis, evaluate the relationship of given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 82.50%, prompt: Can this sentence be inferred from the previous sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 84.00%, prompt: Using a tool to analyze a conclusion, analyze the relationship between the two sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 83.00%, prompt: Is this a conclusion from the next sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 82.50%, prompt: As part of your task in analyzing a conclusion, evaluate the relationship between the two sentences and classify them as 'entailment' or 'not_entailment' based on their relationship. Language: arabic, acc: 82.50%, prompt: Are you following this sentence directly from the previous one? Classify it as 'entailment' or 'not_entailment'. Language: spanish, acc: 84.00%, prompt: In your role as an implication analysis tool, evaluate the relationship between the given phrases and classify them as 'entailment' or 'not_entailment'. Language: spanish, acc: 83.50%, prompt: Determine whether the second sentence necessarily implies the first and label the relation as 'entailment', or as 'not_entailment' if not. Language: spanish, acc: 83.00%, prompt: Classifies the relationship between these two sentences as 'entailment' if one necessarily implies the other, or as 'not_entailment' if not. Language: spanish, acc: 82.50%, prompt: Evaluates whether the information in the second sentence is implied in the first and labels the relationship as 'entailment', or as 'not_entailment' if there is no such implication. Language: spanish, acc: 83.00%, prompt: Given a couple of phrases, label their relationship as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 85.00%, prompt: Analyzes the relationship between the phrases and classifies them as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 81.50%, prompt: Given two sentences, determine whether the second sentence is a necessary consequence of the first and label the relation as 'entailment', or as 'not_entailment' if not. Language: spanish, acc: 84.50%, prompt: Evaluates whether the information presented in the second sentence is implicit in the first and labels the relationship as 'entailment', or as 'not_entailment' if there is no such implication. Language: spanish, acc: 82.00%, prompt: Classifies the relationship between the given phrases as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 83.50%, prompt: Determines whether the information provided in the second sentence is necessarily inferable from the first and labels the relationship as 'entailment', or as 'not_entailment' if not. Language: japanese, acc: 82.50%, prompt: Analyze the relationship of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 83.50%, prompt: Evaluate the semantic relationship of the sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 82.00%, prompt: Please judge the relationship between the given sentences and classify them as 'entailment' or 'not_entailment'. Language: japanese, acc: 84.00%, prompt: Examine the paraphrases of a given sentence and classify them 'entailment' or 'not_entailment'. Language: japanese, acc: 85.00%, prompt: Rate the similarity of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: japanese, acc: 83.50%, prompt: Determinate the semantic connections of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 83.00%, prompt: Examine the semantic match of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: japanese, acc: 83.50%, prompt: Classify it as 'entailment' or 'not_entailment' based on the content of the sentence. Language: japanese, acc: 83.00%, prompt: Analyze the relationship of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 84.50%, prompt: Judge the semantic connections of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: korean, acc: 82.00%, prompt: Evaluate the relationship between any two sentences given to you and classify you as 'entailment' or 'not_entailment'. Language: korean, acc: 82.50%, prompt: Analyze the semantic deductive relations between sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 82.00%, prompt: Evaluate the logical relevance between sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 83.50%, prompt: Evaluate the interaction of two given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 83.50%, prompt: Please check whether there is a semantic match between those two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 84.50%, prompt: Compare information between given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 84.00%, prompt: Please analyse the correlation between those two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 84.00%, prompt: Evaluate the different meanings between given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 84.00%, prompt: Compare the semantic structure of the two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 82.50%, prompt: Evaluate the interactions between sentences and classify them as 'entailment' or 'not_entailment'. # sst2 ## 10 prompts Acc: 97.00%, prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Acc: 96.50%, prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Acc: 96.00%, prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Acc: 95.50%, prompt: Analyze the tone of this statement and respond with either 'positive' or 'negative'. Acc: 95.50%, prompt: Review this statement and decide whether it has a 'positive' or 'negative' sentiment. Acc: 95.00%, prompt: Determine the overall sentiment of this sentence, categorizing it as 'positive' or 'negative'. Acc: 94.50%, prompt: Please identify the emotional tone of this passage: 'positive' or 'negative'? Acc: 92.50%, prompt: Given the context of this text, indicate if the emotion conveyed is 'positive' or 'negative'. Acc: 92.00%, prompt: Assess the mood of the following quote and determine if it's 'positive' or 'negative'. Acc: 85.50%, prompt: Considering the given phrase, would you say it carries a 'positive' or 'negative' connotation? Acc: 97.50%, prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Acc: 97.50%, prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Acc: 97.00%, prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Acc: 97.00%, prompt: In the role of a sentiment analysis tool, respond with 'positive' or 'negative' to classify this statement. Acc: 96.50%, prompt: Working as a sentiment analyzer, please indicate if the following text is 'positive' or 'negative'. Acc: 96.00%, prompt: Functioning as a sentiment identification tool, assess if the following expression is 'positive' or 'negative'. Acc: 96.00%, prompt: Serving as a sentiment evaluation model, determine if the given statement is 'positive' or 'negative'. Acc: 95.50%, prompt: Acting as a sentiment evaluator, identify if the given sentence is 'positive' or 'negative'. Acc: 94.50%, prompt: Taking on the role of an emotion classifier, specify if the provided phrase is 'positive' or 'negative'. Acc: 92.50%, prompt: As an emotion detector, determine if the provided passage conveys a 'positive' or 'negative' sentiment. ## bertattack Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"Evaluate the sentiment of the given text and classify it as 'positive' or 'negative' 5XeflW1ZJc. " Original acc: 97.00%, attacked acc: 96.00%, dropped acc: 1.00% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment 9Q3S1F94fE. " Original acc: 96.00%, attacked acc: 95.00%, dropped acc: 1.00% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"After examining the following expression, label its emotion as either 'positive' or 'negative' Q47Jmd4lMV. " Original acc: 95.50%, attacked acc: 94.50%, dropped acc: 1.00% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative' 9Q3S1F94fE. " Original acc: 97.50%, attacked acc: 96.50%, dropped acc: 1.00% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b"Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative' OB4KVJzIft. " Original acc: 97.50%, attacked acc: 97.00%, dropped acc: 0.50% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"As a sentiment classifier, determine whether the following text is 'positive' or 'negative' LkF0FZxMZ4. " Original acc: 97.00%, attacked acc: 96.50%, dropped acc: 0.50% ## checklist Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"re the sentiment of the given text and describe it as 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 94.00%, dropped acc: 3.00% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"go the read excerpt and be between 'positive' and 'negative' to describing its mood. " Original acc: 96.00%, attacked acc: 77.50%, dropped acc: 18.50% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"in examining the following expression, approach its emotion as either 'positive' or 'negative'. " Original acc: 95.50%, attacked acc: 92.50%, dropped acc: 3.00% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"in the capacity of a sentiment classizer, chose neither the said quotes is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 13.50%, dropped acc: 84.00% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b".how a sentiment classification system, express doubt the supplied text is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 14.00%, dropped acc: 83.50% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"when a sentiment classifier, question neither the given text is 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 22.00%, dropped acc: 75.00% ## deepwordbug Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"ENvaluate the senttment of the giOven text and classify it as 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 94.00%, dropped acc: 3.00% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. " Original acc: 96.00%, attacked acc: 96.00%, dropped acc: 1.00% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"After examining the following expression, label its emotion as either 'positive' or 'negative'. " Original acc: 95.00%, attacked acc: 95.00%, dropped acc: 1.00% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"In the capacity of a sentient olassifier, decide whethKer the given quote is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 94.50%, dropped acc: 3.00% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b"Emulating a sentiment classification system, indicate whetRher the provided text is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 97.00%, dropped acc: 0.50% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"qAs a setniment classifier, determnne whether the following text is 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 94.00%, dropped acc: 3.00% ## stresstest Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"Evaluate the sentiment of the given text and classify it as 'positive' or 'negative' and true is true . " Original acc: 97.00%, attacked acc: 95.50%, dropped acc: 1.50% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment and true is true . " Original acc: 96.00%, attacked acc: 95.50%, dropped acc: 0.50% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"After examining the following expression, label its emotion as either 'positive' or 'negative' and true is true and true is true and true is true and true is true and true is true . " Original acc: 95.50%, attacked acc: 95.50%, dropped acc: 0.00% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative' and false is not true . " Original acc: 97.50%, attacked acc: 96.50%, dropped acc: 1.00% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b"Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative' and false is not true . " Original acc: 97.50%, attacked acc: 97.00%, dropped acc: 0.50% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"As a sentiment classifier, determine whether the following text is 'positive' or 'negative' and false is not true . " Original acc: 97.00%, attacked acc: 96.00%, dropped acc: 1.00% ## textbugger Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"Evaluate the mood of the gievn text and clas\xd1\x95ify it as 'positive' or 'negative'. " Original acc: 96.50%, attacked acc: 90.50%, dropped acc: 6.00% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"Read the given excerpt and choose between 'positive' and 'negative' to describe its mood. " Original acc: 96.00%, attacked acc: 94.00%, dropped acc: 2.00% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"Ater examining the following expression, etiquette its emotion as nor 'positive' or 'negative'. " Original acc: 95.50%, attacked acc: 74.00%, dropped acc: 21.50% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"Onto the capacity of a feeling classifeir, decide si the given quote is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 93.50%, dropped acc: 4.00% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b"Emulating a ambience classifi\xcf\xb2ation system, indciate whether the provided text is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 95.00%, dropped acc: 2.50% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"Since a emotion cla ssifier, determine whether the following text is 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 94.00%, dropped acc: 3.00% ## textfooler Original prompt: Evaluate the sentiment of the given text and classify it as 'positive' or 'negative'. Attacked prompt: b"Evaluate the mood of the given text and classify it as 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 92.00%, dropped acc: 5.00% Original prompt: Read the provided excerpt and choose between 'positive' and 'negative' to describe its sentiment. Attacked prompt: b"Lu the rendered excerpt and withheld between 'positive' and 'negative' to describe its foreboding. " Original acc: 96.00%, attacked acc: 85.00%, dropped acc: 11.00% Original prompt: After examining the following expression, label its emotion as either 'positive' or 'negative'. Attacked prompt: b"After examining the following expression, label its passions as either 'positive' or 'negative'. " Original acc: 95.00%, attacked acc: 93.50%, dropped acc: 1.50% Original prompt: In the capacity of a sentiment classifier, decide whether the given quote is 'positive' or 'negative'. Attacked prompt: b"Onto the capacity of a emotions classifier, decide whether the given quote is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 94.00%, dropped acc: 3.50% Original prompt: Emulating a sentiment classification system, indicate whether the provided text is 'positive' or 'negative'. Attacked prompt: b"Emulating a ambience classification mechanisms, describe whether the presented writings is 'positive' or 'negative'. " Original acc: 97.50%, attacked acc: 91.50%, dropped acc: 6.00% Original prompt: As a sentiment classifier, determine whether the following text is 'positive' or 'negative'. Attacked prompt: b"Iike a mood classifier, determine whether the following text is 'positive' or 'negative'. " Original acc: 97.00%, attacked acc: 94.00%, dropped acc: 3.00% ## translation Language: chinese, acc: 96.00%, prompt: Answer whether the statement is 'positive' or 'negative' based on sentiment analysis. Language: chinese, acc: 94.00%, prompt: As an emotion analysis tool, determine whether the emotion in the text is 'positive' or 'negative'. Language: chinese, acc: 96.00%, prompt: Categorize the statement as 'positive' or 'negative', based on its emotional bent. Language: chinese, acc: 97.50%, prompt: Please use sentiment analysis to classify the text as 'positive' or 'negative'. Language: chinese, acc: 96.00%, prompt: Please determine whether the emotion of the sentence is 'positive' or 'negative' and categorize it. Language: chinese, acc: 97.50%, prompt: Using sentiment analysis, classify the text as 'positive' or 'negative'. Language: chinese, acc: 95.50%, prompt: Please answer whether the emotion of the sentence is 'positive' or 'negative' and categorize it. Language: chinese, acc: 96.50%, prompt: Categorize the statement as 'positive' or 'negative' based on sentiment analysis. Language: chinese, acc: 95.50%, prompt: Please judge this text as 'positive' or 'negative' according to the criteria of sentiment analysis. Language: chinese, acc: 94.50%, prompt: Please classify this sentence as 'positive' or 'negative' according to the principles of emotion analysis. Language: french, acc: 97.00%, prompt: As a sentiment analysis tool, please answer with 'positive' or 'negative' to classify this statement. Language: french, acc: 95.50%, prompt: Determine whether this phrase is 'positive' or 'negative' as a sentiment classification tool. Language: french, acc: 97.00%, prompt: Identify the tone of this statement by choosing between 'positive' and 'negative' as a sentiment analysis tool. Language: french, acc: 97.00%, prompt: Use sentiment analysis to classify this statement as 'positive' or 'negative'. Language: french, acc: 97.00%, prompt: As a sentiment classification tool, please determine whether this statement is 'positive' or 'negative'. Language: french, acc: 97.00%, prompt: Classify this sentence as 'positive' or 'negative' using sentiment analysis. Language: french, acc: 96.50%, prompt: Choose between 'positive' or 'negative' to classify this statement as a sentiment analysis tool. Language: french, acc: 97.00%, prompt: Identify the sentiment expressed in this statement by selecting 'positive' or 'negative' as a sentiment classification tool. Language: french, acc: 96.50%, prompt: Determine whether this phrase is 'positive' or 'negative' using sentiment analysis as a classification tool. Language: french, acc: 97.00%, prompt: Use sentiment analysis to classify this statement as 'positive' or 'negative'. Language: arabic, acc: 96.00%, prompt: Under emotional analysis, answer 'positive' or 'negative' to classify this statement. Language: arabic, acc: 93.50%, prompt: Does this statement express a 'positive' or 'negative' reaction? Language: arabic, acc: 93.50%, prompt: Is that a 'positive' or a 'negative' phrase? Language: arabic, acc: 94.50%, prompt: What is the classification between 'positive' and 'negative'? Language: arabic, acc: 88.50%, prompt: Does this sentence express 'positive' or 'negative' feelings? Language: arabic, acc: 68.50%, prompt: In the context of textual analysis, what classification is this phrase between 'positive' and 'negative'? Language: arabic, acc: 90.50%, prompt: Could this be classified as 'positive' or 'negative'? Language: arabic, acc: 90.50%, prompt: In the context of emotional analysis, what classification is this statement between 'positive' and 'negative'? Language: arabic, acc: 88.50%, prompt: Can this be classified as 'positive' or 'negative'? Language: arabic, acc: 90.00%, prompt: Under the classification of emotions, is this sentence 'positive' or 'negative'? Language: spanish, acc: 97.00%, prompt: As a feeling analysis tool, classify this statement as 'positive' or 'negative'. Language: spanish, acc: 94.00%, prompt: Determine whether this statement has a 'positive' or 'negative' connotation. Language: spanish, acc: 96.50%, prompt: Indicate whether the following statement is 'positive' or 'negative'. Language: spanish, acc: 89.50%, prompt: Evaluate whether this text has a 'positive' or 'negative' emotional charge. Language: spanish, acc: 94.00%, prompt: According to your sentiment analysis, would you say this comment is 'positive' or 'negative'? Language: spanish, acc: 96.00%, prompt: In the context of sentiment analysis, label this sentence as 'positive' or 'negative'. Language: spanish, acc: 95.50%, prompt: Rate the following statement as 'positive' or 'negative', according to your sentiment analysis. Language: spanish, acc: 87.00%, prompt: How would you classify this text in terms of its emotional tone? 'positive' or 'negative'? Language: spanish, acc: 92.00%, prompt: As a tool for sentiment analysis, would you say this statement is 'positive' or 'negative'? Language: spanish, acc: 96.00%, prompt: Classify this statement as 'positive' or 'negative', please. Language: japanese, acc: 95.50%, prompt: Treat this sentence as an emotion analysis tool and categorize it as 'positive' and 'negative'. Language: japanese, acc: 96.00%, prompt: Use this article as a sentiment analysis tool to classify 'positive' and 'negative'. Language: japanese, acc: 94.00%, prompt: Use this sentence as an emotion analysis tool to determine whether it is 'positive' or 'negative'. Language: japanese, acc: 94.00%, prompt: Use this sentence as an emotion analysis tool to classify 'positive' and 'negative'. Language: japanese, acc: 95.50%, prompt: Use this sentence as a sentiment analysis tool and classify it as 'positive' or 'negative'. Language: japanese, acc: 94.00%, prompt: To classify this sentence as 'positive' or 'negative', evaluate it as a sentiment analysis tool. Language: japanese, acc: 93.50%, prompt: Treat this sentence as an emotion analysis tool to determine whether it is 'positive' or 'negative'. Language: japanese, acc: 95.50%, prompt: Use this sentence as a sentiment analysis tool to classify 'positive' and 'negative'. Language: japanese, acc: 93.00%, prompt: Analyze this sentence as an emotion analysis tool to classify whether it is 'positive' or 'negative'. Language: japanese, acc: 94.00%, prompt: Use this sentence as an emotional analysis tool to determine whether it is 'positive' or 'negative'. Language: korean, acc: 95.50%, prompt: As an emotional analysis tool, respond with 'positive' or 'negative' to classify these sentences. Language: korean, acc: 97.00%, prompt: Classify this sentence as 'positive' if you regard it as positive, 'negative' if you regard it as negative. Language: korean, acc: 91.00%, prompt: Please rate the emotion of this sentence and classify it as 'positive' or 'negative'. Language: korean, acc: 96.50%, prompt: Classify this sentence as 'positive' if you perceive it positively and 'negative' if you perceive it negatively. Language: korean, acc: 96.50%, prompt: If this is a sentence delivered using a positive expression, classify it as 'positive' and if this is a sentence delivered using a negative expression, classify it as 'negative'. Language: korean, acc: 97.00%, prompt: Respond with 'positive' or 'negative' by categorizing whether the sentence is positive or negative. Language: korean, acc: 92.00%, prompt: Please analyze the emotion in this sentence and classify it as 'positive' or 'negative'. Language: korean, acc: 97.00%, prompt: Classify this sentence as 'positive' if it contains a positive meaning, 'negative' if it contains a negative meaning. Language: korean, acc: 96.50%, prompt: Classify this sentence as 'positive' if it contains positive content, 'negative' if it contains negative content. Language: korean, acc: 95.00%, prompt: Classify what you're trying to convey in this sentence as 'positive' if it's positive, and 'negative' if it's negative. # wnli ## 10 prompts Acc: 74.65%, prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Acc: 70.42%, prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Acc: 67.61%, prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: Does the relationship between the given sentences represent entailment or not_entailment? Respond with 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: Analyze the two provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: Determine if the given pair of sentences displays entailment or not_entailment. Respond with 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: Considering the two sentences, identify if their relationship is 'entailment' or 'not_entailment'. Acc: 64.79%, prompt: Examine the pair of sentences and determine if they exhibit entailment or not_entailment. Answer with either 'entailment' or 'not_entailment'. Acc: 64.79%, prompt: Identify whether the given pair of sentences demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 63.38%, prompt: Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'. Acc: 63.38%, prompt: Please classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Acc: 66.20%, prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 64.79%, prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Acc: 64.79%, prompt: As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment' or 'not_entailment'. Acc: 61.97%, prompt: As an entailment identification system, examine the connection between the following sentences and respond with 'entailment' or 'not_entailment'. Acc: 61.97%, prompt: In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment' or 'not_entailment'. Acc: 61.97%, prompt: As an instrument for entailment evaluation, consider the two sentences and determine if their relationship is 'entailment' or 'not_entailment'. Respond with 'entailment' or 'not_entailment'. Acc: 60.56%, prompt: Acting as an entailment detection instrument, determine if the given pair of sentences demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Acc: 60.56%, prompt: While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment'. Acc: 59.15%, prompt: Working as an entailment classifier, identify whether the given pair of sentences displays entailment or not_entailment. Respond with 'entailment' or 'not_entailment'. ## bertattack Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'from the above two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one space. ' Original acc: 73.24%, attacked acc: 61.97%, dropped acc: 11.27% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"examine the paired actions and categorize their relationship as is 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 59.15%, dropped acc: 9.86% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the adjacent sentences is 'entailment' or 'not_entailment'. " Original acc: 67.61%, attacked acc: 60.56%, dropped acc: 7.04% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"her as an entailment evaluation tool, analyze the received sentences and decide if their relationship is 'entailment' or 'not_entailment'. " Original acc: 67.61%, attacked acc: 60.56%, dropped acc: 7.04% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic relationship analyzer, define the connection between the given links and decide if they display entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 56.34%, dropped acc: 9.86% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"i your role as an entailment analysis server, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 56.34%, dropped acc: 9.86% ## checklist Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'Aer the following two senxtences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. ' Original acc: 73.24%, attacked acc: 64.79%, dropped acc: 8.45% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 69.01%, dropped acc: 4.23% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the followping senZences is 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 61.97%, dropped acc: 7.04% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 66.20%, dropped acc: 2.82% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic relationship analyzer, examine the connection between the given sentenCes and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 61.97%, dropped acc: 4.23% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, asress the rlationship between the given sentences and classify it as 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 57.75%, dropped acc: 8.45% ## deepwordbug Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word and false is not true . ' Original acc: 73.24%, attacked acc: 70.42%, dropped acc: 2.82% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment' and false is not true . " Original acc: 69.01%, attacked acc: 70.42%, dropped acc: -1.41% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the following sentences is 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 67.61%, attacked acc: 64.79%, dropped acc: 2.82% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 69.01%, attacked acc: 69.01%, dropped acc: 0.00% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 66.20%, attacked acc: 66.20%, dropped acc: 0.00% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment' and true is true and true is true and true is true and true is true and true is true . " Original acc: 64.79%, attacked acc: 59.15%, dropped acc: 5.63% ## stresstest Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'Are the subsequently to sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just o ne word. ' Original acc: 73.24%, attacked acc: 64.79%, dropped acc: 8.45% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 69.01%, dropped acc: 4.23% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the later sentences is 'entailment' or 'not_entailment'. " Original acc: 67.61%, attacked acc: 60.56%, dropped acc: 7.04% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relation\xd1\x95hip is 'entailment' or 'not_entailment'. " Original acc: 67.61%, attacked acc: 61.97%, dropped acc: 5.63% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic relationship analyzer, investigated the linking between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 59.15%, dropped acc: 7.04% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relat ionship between the given s\xd0\xb5ntences and classify it as 'entailment' or 'not_entailment'. " Original acc: 63.38%, attacked acc: 59.15%, dropped acc: 4.23% ## textbugger Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word 5XeflW1ZJc. ' Original acc: 73.24%, attacked acc: 67.61%, dropped acc: 5.63% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment' 5XeflW1ZJc. " Original acc: 70.42%, attacked acc: 66.20%, dropped acc: 4.23% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the following sentences is 'entailment' or 'not_entailment' Vq3aBzuZcD. " Original acc: 69.01%, attacked acc: 61.97%, dropped acc: 7.04% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment' bhrRSokrfa. " Original acc: 69.01%, attacked acc: 66.20%, dropped acc: 2.82% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment' is77sOXAu8. " Original acc: 66.20%, attacked acc: 59.15%, dropped acc: 7.04% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment' BWmsLx8xOA. " Original acc: 64.79%, attacked acc: 63.38%, dropped acc: 1.41% ## textfooler Original prompt: Are the following two sentences entailment or not_entailment? Answer me with "entailment" or "not_entailment", just one word. Attacked prompt: b'Are the wake two sentences entailment or not_entailment? Answered me with "entailment" or "not_entailment", just one word. ' Original acc: 74.65%, attacked acc: 63.38%, dropped acc: 11.27% Original prompt: Review the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. Attacked prompt: b"Revisions the two statements and categorize their relationship as either 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 61.97%, dropped acc: 7.04% Original prompt: Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'. Attacked prompt: b"Indicate if the connection between the subsequently sentences is 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 59.15%, dropped acc: 9.86% Original prompt: Functioning as an entailment evaluation tool, analyze the provided sentences and decide if their relationship is 'entailment' or 'not_entailment'. Attacked prompt: b"Work as an entailment evaluation tool, analyze the provided condemnation and decide if their relationship is 'entailment' or 'not_entailment'. " Original acc: 69.01%, attacked acc: 60.56%, dropped acc: 8.45% Original prompt: In the role of a semantic relationship analyzer, examine the connection between the given sentences and decide if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. Attacked prompt: b"In the role of a semantic contacts analyzer, analysed the connecting between the given sentences and choose if they exhibit entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 57.75%, dropped acc: 8.45% Original prompt: In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Attacked prompt: b"In your role as an entailment valuation tool, assess the relationship between the given condemning and classify it as 'entailment' or 'not_entailment'. " Original acc: 66.20%, attacked acc: 57.75%, dropped acc: 8.45% ## translation Language: chinese, acc: 66.20%, prompt: In the light of an implication analysis tool, evaluate the relationship between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 66.20%, prompt: From the perspective of an implication analysis tool, determine whether there is an implication relationship in the following sentences by classifying them as 'entailment' or 'not_entailment'. Language: chinese, acc: 63.38%, prompt: Please use an implication analysis tool to determine whether an implication relationship exists in the following sentences by classifying them as 'entailment' or 'not_entailment'. Language: chinese, acc: 71.83%, prompt: Please evaluate the relation of the following sentences as 'entailment' or 'not_entailment' for the purpose of determining implication relation. Language: chinese, acc: 66.20%, prompt: Please use the implication analysis tool to evaluate the relationships between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 69.01%, prompt: For the purpose of determining implicative relations, analyze the relations of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 64.79%, prompt: Please use the implication analysis tool to determine the relationship of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 63.38%, prompt: Please use the implication judgment tool to assess the relevance of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 61.97%, prompt: Please, with implication analysis as the main task, determine the relationships between the following sentences and classify them as 'entailment' or 'not_entailment'. Language: chinese, acc: 71.83%, prompt: Using the implication judgment as a criterion, analyze the relation of the following sentences and classify them as 'entailment' or 'not_entailment'. Language: french, acc: 64.79%, prompt: As an engagement analysis tool, evaluate the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'. Language: french, acc: 61.97%, prompt: Determine whether the given sentences involve one another or not as an implication analysis tool. Classify them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 61.97%, prompt: Using implication analysis, evaluate whether the sentences provided have a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 61.97%, prompt: As an engagement assessment tool, determine whether the sentences provided have a logical relationship and classify them as 'entailment' or 'not_entailment'. Language: french, acc: 61.97%, prompt: As an implication classification tool, analyze the sentences provided to determine if there is a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 66.20%, prompt: Using implication analysis, determine whether the given sentences have a cause-effect relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 66.20%, prompt: Evaluate the relationship between the given sentences using implication analysis and rank them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 60.56%, prompt: As an engagement detection tool, determine whether the given sentences have a logical relationship and categorize them as 'entailment' or 'not_entailment'. Language: french, acc: 63.38%, prompt: Using implication analysis, evaluate whether the sentences provided have a cause-effect relationship and rank them accordingly as 'entailment' or 'not_entailment'. Language: french, acc: 66.20%, prompt: Determine whether the given sentences have a cause-effect relationship as an engagement analysis tool and categorize them as 'entailment' or 'not_entailment'. Language: arabic, acc: 64.79%, prompt: In your role as a tool for reasoning analysis, evaluate the relationship between given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 64.79%, prompt: Can you determine whether this sentence is inferred from the other sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 69.01%, prompt: Using the tool of reasoning analysis, analyze the relationship between given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 63.38%, prompt: Does this sentence represent a conclusion from the previous sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 69.01%, prompt: As a tool of reasoning analysis, evaluate the relationship of given sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 67.61%, prompt: Can this sentence be inferred from the previous sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 66.20%, prompt: Using a tool to analyze a conclusion, analyze the relationship between the two sentences and classify them as 'entailment' or 'not_entailment'. Language: arabic, acc: 61.97%, prompt: Is this a conclusion from the next sentence? Classify it as 'entailment' or 'not_entailment'. Language: arabic, acc: 69.01%, prompt: As part of your task in analyzing a conclusion, evaluate the relationship between the two sentences and classify them as 'entailment' or 'not_entailment' based on their relationship. Language: arabic, acc: 69.01%, prompt: Are you following this sentence directly from the previous one? Classify it as 'entailment' or 'not_entailment'. Language: spanish, acc: 64.79%, prompt: In your role as an implication analysis tool, evaluate the relationship between the given phrases and classify them as 'entailment' or 'not_entailment'. Language: spanish, acc: 71.83%, prompt: Determine whether the second sentence necessarily implies the first and label the relation as 'entailment', or as 'not_entailment' if not. Language: spanish, acc: 64.79%, prompt: Classifies the relationship between these two sentences as 'entailment' if one necessarily implies the other, or as 'not_entailment' if not. Language: spanish, acc: 66.20%, prompt: Evaluates whether the information in the second sentence is implied in the first and labels the relationship as 'entailment', or as 'not_entailment' if there is no such implication. Language: spanish, acc: 69.01%, prompt: Given a couple of phrases, label their relationship as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 67.61%, prompt: Analyzes the relationship between the phrases and classifies them as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 64.79%, prompt: Given two sentences, determine whether the second sentence is a necessary consequence of the first and label the relation as 'entailment', or as 'not_entailment' if not. Language: spanish, acc: 67.61%, prompt: Evaluates whether the information presented in the second sentence is implicit in the first and labels the relationship as 'entailment', or as 'not_entailment' if there is no such implication. Language: spanish, acc: 64.79%, prompt: Classifies the relationship between the given phrases as 'entailment' if one necessarily implies the other, or as 'not_entailment' if there is no such implication. Language: spanish, acc: 70.42%, prompt: Determines whether the information provided in the second sentence is necessarily inferable from the first and labels the relationship as 'entailment', or as 'not_entailment' if not. Language: japanese, acc: 63.38%, prompt: Analyze the relationship of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 61.97%, prompt: Evaluate the semantic relationship of the sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 63.38%, prompt: Please judge the relationship between the given sentences and classify them as 'entailment' or 'not_entailment'. Language: japanese, acc: 63.38%, prompt: Examine the paraphrases of a given sentence and classify them 'entailment' or 'not_entailment'. Language: japanese, acc: 63.38%, prompt: Rate the similarity of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: japanese, acc: 59.15%, prompt: Determinate the semantic connections of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 67.61%, prompt: Examine the semantic match of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: japanese, acc: 63.38%, prompt: Classify it as 'entailment' or 'not_entailment' based on the content of the sentence. Language: japanese, acc: 63.38%, prompt: Analyze the relationship of a given sentence and classify it as 'entailment' or 'not_entailment'. Language: japanese, acc: 64.79%, prompt: Judge the semantic connections of a given sentence and categorize it as 'entailment' or 'not_entailment'. Language: korean, acc: 67.61%, prompt: Evaluate the relationship between any two sentences given to you and classify you as 'entailment' or 'not_entailment'. Language: korean, acc: 64.79%, prompt: Analyze the semantic deductive relations between sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 66.20%, prompt: Evaluate the logical relevance between sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 66.20%, prompt: Evaluate the interaction of two given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 64.79%, prompt: Please check whether there is a semantic match between those two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 69.01%, prompt: Compare information between given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 66.20%, prompt: Please analyse the correlation between those two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 66.20%, prompt: Evaluate the different meanings between given sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 64.79%, prompt: Compare the semantic structure of the two sentences and classify them as 'entailment' or 'not_entailment'. Language: korean, acc: 61.97%, prompt: Evaluate the interactions between sentences and classify them as 'entailment' or 'not_entailment'. # mmlu ## 10 prompts Acc: 67.00%, prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Acc: 67.00%, prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Acc: 66.50%, prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Acc: 66.50%, prompt: For each multiple-choice question about {}, identify the correct answer by selecting 'A', 'B', 'C', or 'D'. Acc: 66.00%, prompt: Below are multiple-choice question concerning {}. Indicate your response with 'A', 'B', 'C', or 'D'. Acc: 66.00%, prompt: Examine the following question based on {} and choose the correct response from 'A', 'B', 'C', or 'D'. Acc: 65.50%, prompt: Answer the following multiple-choice question about {} by selecting the correct option: 'A', 'B', 'C', or 'D'. Acc: 65.50%, prompt: Answer the subsequent multiple-choice question about {} by picking the right option among 'A', 'B', 'C', or 'D'. Acc: 65.00%, prompt: For the multiple-choice question related to {}, please choose the most accurate answer from 'A', 'B', 'C', or 'D'. Acc: 64.50%, prompt: Evaluate the multiple-choice question about {} and select the most fitting response from 'A', 'B', 'C', or 'D'. Acc: 66.50%, prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Acc: 66.00%, prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Acc: 66.00%, prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Acc: 65.50%, prompt: As an expert in {}, respond to the following multiple-choice question by selecting 'A', 'B', 'C', or 'D'. Acc: 65.50%, prompt: Utilizing your expertise in {}, answer the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Acc: 65.00%, prompt: As a knowledgeable individual in {}, provide your response to the multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Acc: 64.50%, prompt: Given your proficiency in {}, please answer the subsequent multiple-choice question with 'A', 'B', 'C', or 'D'. Acc: 64.00%, prompt: With your understanding of {}, kindly answer the subsequent multiple-choice question by selecting 'A', 'B', 'C', or 'D'. Acc: 64.00%, prompt: As a skilled person in the field of {}, please respond to the multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Acc: 64.00%, prompt: Drawing upon your mastery of {}, please answer the multiple-choice question by selecting the correct option from 'A', 'B', 'C', or 'D'. ## bertattack Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"a relation to the multiple-choice inquiry on {}, who provide the incorrect answer by choosing 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 53.00%, dropped acc: 13.50% Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"but the first multiple-choice question on {}, follow the wrong answer from the options 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 33.50%, dropped acc: 33.00% Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"at respond to the multiple-independent question about {} by selecting the inappropriate question: 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 11.50%, dropped acc: 55.00% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"all your ability with {}, attend to the following multiple-select question by picking 'A', 'B', 'C', or 'D'." Original acc: 65.50%, attacked acc: 52.50%, dropped acc: 13.00% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"With your knowledge of it, question the following multiple-answered segment by choosing 'A', 'B', 'C', or 'D'." Original acc: 66.50%, attacked acc: 59.50%, dropped acc: 7.00% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-versed in 41, please caution the multiple-selecting situation below by seiecting 'A', 'B', 'C', or 'D'." Original acc: 65.00%, attacked acc: 54.50%, dropped acc: 10.50% ## checklist Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D' 9vGXgnbWB8. " Original acc: 66.00%, attacked acc: 64.00%, dropped acc: 2.00% Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D' MQnugHcaoy. " Original acc: 66.00%, attacked acc: 63.00%, dropped acc: 3.00% Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D' RcRneWo6Iv. " Original acc: 66.50%, attacked acc: 64.00%, dropped acc: 2.50% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D' mOJ9nEwT4f." Original acc: 65.50%, attacked acc: 63.00%, dropped acc: 2.50% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D' Cev0s49fbe." Original acc: 66.00%, attacked acc: 63.50%, dropped acc: 2.50% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D' ItGDrrA1Sc." Original acc: 66.00%, attacked acc: 63.00%, dropped acc: 3.00% ## deepwordbug Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"PleIse respond to the multiple-choice question about {} by selecting the approprirate answer: 'A', 'B', 'C', or 'D'. " Original acc: 67.00%, attacked acc: 64.00%, dropped acc: 3.00% Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"ISn ralation to the mulitple-choice question on {}, pleare provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 63.50%, dropped acc: 3.00% Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"Regardxng the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. " Original acc: 66.00%, attacked acc: 64.50%, dropped acc: 1.50% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-versed in {}, please address the mulZtiple-choice question below by selecting 'A', 'B', 'C', or 'D'." Original acc: 65.50%, attacked acc: 62.50%, dropped acc: 3.00% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"ConsiderinZ your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'." Original acc: 66.50%, attacked acc: 63.50%, dropped acc: 3.00% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"WitU your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'." Original acc: 66.50%, attacked acc: 64.00%, dropped acc: 2.50% ## stresstest Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D' and false is not true . " Original acc: 66.50%, attacked acc: 62.50%, dropped acc: 4.00% Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D' and false is not true . " Original acc: 66.50%, attacked acc: 61.50%, dropped acc: 5.00% Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D' and false is not true . " Original acc: 66.00%, attacked acc: 62.00%, dropped acc: 4.00% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D' and false is not true ." Original acc: 66.50%, attacked acc: 61.50%, dropped acc: 5.00% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D' and true is true and true is true and true is true and true is true and true is true ." Original acc: 66.50%, attacked acc: 62.00%, dropped acc: 4.50% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D' and false is not true ." Original acc: 65.50%, attacked acc: 59.50%, dropped acc: 6.00% ## textbugger Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"At relationship to the mul\xf0\x9d\x9a\x9diple-choice question on {}, please provi\xd4\x81e the accurate responding by choosing 'A', 'B', 'C', or 'D'. " Original acc: 67.50%, attacked acc: 61.50%, dropped acc: 6.00% Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"Inviting respo\xd5\xb8d to the multiplechoice question about {} by selecting the a ppropriate answer: 'A', 'B', 'C', or 'D'. " Original acc: 67.50%, attacked acc: 63.00%, dropped acc: 4.50% Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"Regarding the following multiple-choice ques\xf0\x9d\x9a\x9dion on {}, chosen the correct answer from the option 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 62.50%, dropped acc: 4.00% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"Consideri ng your familiarity with {}, attend to the \xf0\x9d\x9a\x8following multiple-choice question by picking 'A', 'B', 'C', or 'D'." Original acc: 66.00%, attacked acc: 63.00%, dropped acc: 3.00% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"Via your kowledge of {}, tackle the following multiple-ch\xd0\xbeice question by choosing 'A', 'B', 'C', or 'D'." Original acc: 65.50%, attacked acc: 62.50%, dropped acc: 3.00% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-vesred in {}, please address the mu\xe2\x85\xbctiple-choice questio\xd5\xb8 below by selecting 'A', 'B', 'C', or 'D'." Original acc: 65.50%, attacked acc: 61.50%, dropped acc: 4.00% ## textfooler Original prompt: Please respond to the multiple-choice question about {} by selecting the appropriate answer: 'A', 'B', 'C', or 'D'. Attacked prompt: b"Asking reactions to the multiple-choice question about {} by selecting the appropriate reacts: 'A', 'B', 'C', or 'D'. " Original acc: 67.00%, attacked acc: 60.00%, dropped acc: 7.00% Original prompt: In relation to the multiple-choice question on {}, please provide the accurate answer by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"Onto relating to the multiple-choice question on {}, please provide the accurate address by choosing 'A', 'B', 'C', or 'D'. " Original acc: 67.00%, attacked acc: 61.50%, dropped acc: 5.50% Original prompt: Regarding the following multiple-choice question on {}, pick the correct answer from the options 'A', 'B', 'C', or 'D'. Attacked prompt: b"Regarding the thereafter multiple-choice question on {}, took the correct answer from the options 'A', 'B', 'C', or 'D'. " Original acc: 66.50%, attacked acc: 62.00%, dropped acc: 4.50% Original prompt: With your knowledge of {}, tackle the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'. Attacked prompt: b"From your acquaintances of {}, addressing the following multiple-choice question by choosing 'A', 'B', 'C', or 'D'." Original acc: 66.00%, attacked acc: 61.50%, dropped acc: 4.50% Original prompt: Considering your familiarity with {}, attend to the following multiple-choice question by picking 'A', 'B', 'C', or 'D'. Attacked prompt: b"Considering your familiarity with {}, attend to the suite multiple-choice question by choices 'A', 'B', 'C', or 'D'." Original acc: 66.00%, attacked acc: 62.00%, dropped acc: 4.00% Original prompt: As someone well-versed in {}, please address the multiple-choice question below by selecting 'A', 'B', 'C', or 'D'. Attacked prompt: b"As someone well-versed in {}, please address the multiple-choice trouble below by decide 'A', 'B', 'C', or 'D'." Original acc: 65.00%, attacked acc: 61.50%, dropped acc: 3.50% ## translation Language: chinese, acc: 65.00%, prompt: For the following multiple choice question about {}, please select the correct answer from 'A', 'B', 'C' or 'D'. Language: chinese, acc: 66.00%, prompt: Please select the correct answer from 'A', 'B', 'C' or 'D' for the following multiple choice question for {}. Language: chinese, acc: 66.50%, prompt: Select the correct answer from 'A', 'B', 'C' or 'D' for the following multiple choice question {}. Language: chinese, acc: 66.50%, prompt: Please choose the correct answer from 'A', 'B', 'C' or 'D' according to the following multiple-choice questions about {}. Language: chinese, acc: 66.00%, prompt: Please select the correct answer from 'A', 'B', 'C' or 'D' for the {} multiple choice questions below. Language: chinese, acc: 66.50%, prompt: The following is A multiple choice question about {}. Please select the correct answer from 'A', 'B', 'C' or 'D'. Language: chinese, acc: 66.00%, prompt: Please select the correct answer from 'A', 'B', 'C' or 'D' for the following multiple choice question {}. Language: chinese, acc: 66.50%, prompt: Please choose the correct answer from 'A', 'B', 'C' or 'D' according to the following multiple-choice questions about {}. Language: chinese, acc: 67.00%, prompt: Please select the correct answer from 'A', 'B', 'C' or 'D' for the following multiple choice questions about {}. Language: chinese, acc: 67.00%, prompt: Please select the correct answer from 'A', 'B', 'C' or 'D' for the following multiple choice questions about {}. Language: french, acc: 66.00%, prompt: For the following multiple choice question on {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: french, acc: 66.00%, prompt: This is a multiple choice question about {}. Select the correct answer from options 'A', 'B', 'C' or 'D'. Language: french, acc: 65.00%, prompt: In the context of the multiple-choice question on {}, identify the correct answer from options 'A', 'B', 'C' or 'D'. Language: french, acc: 64.50%, prompt: About the following question on {}, determine the correct answer from the choices 'A', 'B', 'C' or 'D'. Language: french, acc: 65.50%, prompt: Carefully review the multiple-choice question regarding {}. Choose the correct answer from options 'A', 'B', 'C', or 'D'. Language: french, acc: 66.50%, prompt: For the multiple-choice question for {}, indicate the correct answer from options 'A', 'B', 'C', or 'D'. Language: french, acc: 64.50%, prompt: The next question is about {}. Select the correct answer from the choices 'A', 'B', 'C' or 'D'. Language: french, acc: 64.50%, prompt: As part of the multiple-choice question on {}, choose the appropriate answer from options 'A', 'B', 'C' or 'D'. Language: french, acc: 63.00%, prompt: Rate your understanding of the multiple-choice question on {}. Choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: french, acc: 63.50%, prompt: Analyze the following multiple-choice question on {}. Identify the correct answer among choices 'A', 'B', 'C' or 'D'. Language: arabic, acc: 67.00%, prompt: For the multiple choice question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 65.50%, prompt: For the following multiple-choice question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 67.00%, prompt: For the following multiple choice question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 66.50%, prompt: When it comes to the multiple-choice question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 66.00%, prompt: For the multiple-choice question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 65.00%, prompt: If the question for {} is multiple choice, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 63.00%, prompt: For the question regarding {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 65.00%, prompt: For the question about {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 65.00%, prompt: When it comes to the question regarding {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: arabic, acc: 64.50%, prompt: For the question regarding {}, choose the correct answer from options 'A', 'B', 'C' or 'D'. Language: spanish, acc: 66.00%, prompt: For the following multiple-choice question about {}, choose the correct answer from 'A', 'B', 'C', or 'D'. Language: spanish, acc: 65.50%, prompt: For the following multiple-choice question about {}, select the correct answer from 'A', 'B', 'C', or 'D'. Language: spanish, acc: 66.00%, prompt: For the following multiple-choice question about {}, choose the correct answer from 'A', 'B', 'C', or 'D'. Language: spanish, acc: 65.00%, prompt: Within the context of the following multiple-choice question about {}, choose the correct option from 'A', 'B', 'C', or 'D'. Language: spanish, acc: 67.50%, prompt: For the following multiple-choice statement about {}, select the correct answer from 'A', 'B', 'C', or 'D'. Language: spanish, acc: 64.00%, prompt: Considering the following multiple-choice question about {}, mark the correct answer with 'A', 'B', 'C', or 'D'. Language: spanish, acc: 65.50%, prompt: For the following multiple-choice question about {}, choose the correct alternative among 'A', 'B', 'C' or 'D'. Language: spanish, acc: 66.00%, prompt: For the following multiple-choice statement about {}, choose the correct option from alternatives 'A', 'B', 'C', or 'D'. Language: spanish, acc: 64.00%, prompt: Within the context of the following multiple-choice question about {}, select the correct answer from alternatives 'A', 'B', 'C', or 'D'. Language: spanish, acc: 64.50%, prompt: Considering the following multiple-choice statement about {}, mark the correct alternative with the options 'A', 'B', 'C' or 'D'. Language: japanese, acc: 65.50%, prompt: Choose the appropriate answer from options 'A', 'B', 'C', or 'D' for {} regarding the following question. Language: japanese, acc: 65.50%, prompt: Choose the correct answer from 'A', 'B', 'C', or 'D' for the following multiple-choice question about {}. Language: japanese, acc: 67.50%, prompt: For the following multiple-choice questions about {}, choose the correct answer from 'A', 'B', 'C', or 'D'. Language: japanese, acc: 66.50%, prompt: Choose the correct answer from options 'A', 'B', 'C', or 'D' for the following questions about {}. Language: japanese, acc: 66.00%, prompt: In the multiple choice questions below, choose the correct answer for {} from 'A', 'B', 'C', or 'D'. Language: japanese, acc: 66.00%, prompt: Choose the correct answer from the options 'A', 'B', 'C', or 'D' for the following questions about {}. Language: japanese, acc: 66.00%, prompt: In the multiple choice questions below, choose the correct answer for {} from 'A', 'B', 'C', or 'D'. Language: japanese, acc: 69.00%, prompt: Choose the correct answer from 'A', 'B', 'C', or 'D' for the following multiple choice questions about {}. Language: japanese, acc: 66.00%, prompt: In the multiple choice questions below, choose the correct answer for {} from 'A', 'B', 'C', or 'D'. Language: japanese, acc: 65.50%, prompt: Choose the correct answer from options 'A', 'B', 'C', or 'D' for {} regarding the following question. Language: korean, acc: 64.00%, prompt: For the multiple choice problem about, choose the correct answer for '{}' from 'A', 'B', 'C', or 'D'. Language: korean, acc: 64.00%, prompt: Choose the correct answer for '{}' from 'A', 'B', 'C', or 'D' in the multiple choice problem involving, Language: korean, acc: 65.00%, prompt: For the multiple choice problem below, choose the correct answer to '{}' from 'A', 'B', 'C', or 'D'. Language: korean, acc: 63.50%, prompt: In the following multiple-choice problem, choose the correct answer for '{}' from 'A', 'B', 'C', or 'D'. Language: korean, acc: 65.50%, prompt: For the following multiple choice problem, choose the correct answer for '{}' from 'A', 'B', 'C', or 'D'. Language: korean, acc: 65.50%, prompt: Solve multiple choice problems about: Which of 'A', 'B', 'C', or 'D' is the correct answer for '{}'. Language: korean, acc: 63.00%, prompt: Choose the correct answer to the multiple-choice question below. Is '{}' an 'A', 'B', 'C', or 'D'. Language: korean, acc: 64.50%, prompt: Solve the following multiple-choice problem. Choose the correct answer for '{}' from 'A', 'B', 'C', or 'D'. Language: korean, acc: 64.00%, prompt: Choose the correct answer to the following multiple choice problem: Is '{}' 'A', 'B', 'C', or 'D'. Language: korean, acc: 66.50%, prompt: Solve multiple-choice problems about: Please select 'A', 'B', 'C', or 'D' for the correct answer to '{}'. # squad_v2 ## 10 prompts Acc: 74.58%, prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Acc: 72.80%, prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Acc: 72.75%, prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Acc: 72.59%, prompt: Discover the best answer based on the context. If the context doesn't include an answer, say 'unanswerable'. Acc: 72.45%, prompt: Identify the most relevant answer from the context. If it's not possible to find an answer, state 'unanswerable'. Acc: 72.20%, prompt: Please extract the most appropriate answer from the context. If an answer is not present, indicate 'unanswerable'. Acc: 71.04%, prompt: Find the correct answer in the context provided. If an answer cannot be found, please respond with 'unanswerable'. Acc: 70.35%, prompt: Search the context for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Acc: 70.01%, prompt: Locate the most accurate answer within the context. If the context doesn't provide an answer, mention 'unanswerable'. Acc: 69.34%, prompt: From the context, provide the most precise answer. If the answer is not in the context, respond with 'unanswerable'. Acc: 73.04%, prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Acc: 72.85%, prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Acc: 72.13%, prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Acc: 71.99%, prompt: Using your knowledge of the context, identify the best answer to the question. If the context doesn't provide an answer, write 'unanswerable'. Acc: 71.73%, prompt: As an expert with a deep understanding of the context, find the best answer. If the context doesn't include an answer, say 'unanswerable'. Acc: 71.54%, prompt: Drawing upon your expertise in the context, determine the most suitable answer. If an answer isn't available, state 'unanswerable'. Acc: 70.96%, prompt: With your extensive knowledge of the context, answer the question accurately. If the context doesn't contain the answer, reply with 'unanswerable'. Acc: 70.84%, prompt: Drawing on your expertise in the context, provide the most precise answer. If the answer is not in the context, respond with 'unanswerable'. Acc: 69.63%, prompt: As an authority on the context, locate the most accurate answer. If the context doesn't contain the answer, mention 'unanswerable'. Acc: 69.20%, prompt: As a well-informed specialist familiar with the context, provide an answer to the question. If the context doesn't contain an answer, reply with 'unanswerable'. ## bertattack Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Please derive the most fitting function from the context. assume there isn't an response in the context, use 'unanswerable'." Original acc: 73.49%, attacked acc: 68.11%, dropped acc: 5.38% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Using the context, sort the most suited response. given the context doesn't contain the questions, write 'unanswerable'." Original acc: 72.80%, attacked acc: 67.19%, dropped acc: 5.61% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the use context, provide the best possible answer. who there'tis no answer provided in the area, reply with 'unanswerable'." Original acc: 72.25%, attacked acc: 65.81%, dropped acc: 6.44% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"As a subject matter who, extract the correct answer from the contexts. If an answer is not present, remove 'unanswerable'." Original acc: 71.97%, attacked acc: 66.16%, dropped acc: 5.80% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-versed in the context, visit derive the most fitting right. said there isn't an query in the context, use 'unanswerable'." Original acc: 72.88%, attacked acc: 67.25%, dropped acc: 5.63% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a proficient expert in the given context, try for the most specific out. If the question lacks be search, end by as 'unanswerable'." Original acc: 72.45%, attacked acc: 61.02%, dropped acc: 11.43% ## checklist Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable' iia2yL9Uzm." Original acc: 74.49%, attacked acc: 73.31%, dropped acc: 1.19% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable' iia2yL9Uzm." Original acc: 72.80%, attacked acc: 71.67%, dropped acc: 1.13% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable' ItGDrrA1Sc." Original acc: 73.25%, attacked acc: 71.09%, dropped acc: 2.15% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable' YHQiDfFhw4." Original acc: 72.93%, attacked acc: 71.55%, dropped acc: 1.38% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable' IhuBIhoPGc." Original acc: 72.71%, attacked acc: 73.22%, dropped acc: -0.51% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable' XStsD2pUzu." Original acc: 72.63%, attacked acc: 63.07%, dropped acc: 9.55% ## deepwordbug Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"PleCase Yderive the most ftting answeor from the conetext. If there isn't an Manswer in the coRtext, use 'unanswerable'." Original acc: 74.08%, attacked acc: 70.69%, dropped acc: 3.40% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the goiven context, provide the best possible answer. IL there's no answer available in the coYtext, reply with 'unanswerable'." Original acc: 73.25%, attacked acc: 67.18%, dropped acc: 6.07% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Using the context, deterSine the most suitable anser. If the conqext doesn't contaun the Ganswer, wirte 'unanswerable'." Original acc: 73.30%, attacked acc: 67.99%, dropped acc: 5.31% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-velrsed in the context, please derive the most fitting answer. If there isn't an ansWer in the context, use 'unanswerable'." Original acc: 73.04%, attacked acc: 71.41%, dropped acc: 1.63% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"s a subject matter eAxpert, extract the correc answer from the context. If an answer is not presnt, idicate 'unanswerable'." Original acc: 72.68%, attacked acc: 66.77%, dropped acc: 5.91% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a proficient expeort in the given context, search for the most relevant answer. If the anwer acnnot be Mfound, respsnd by asying 'unanswerable'." Original acc: 71.95%, attacked acc: 68.21%, dropped acc: 3.74% ## stresstest Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable' and false is not true ." Original acc: 73.99%, attacked acc: 61.58%, dropped acc: 12.41% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable' and true is true ." Original acc: 73.10%, attacked acc: 72.05%, dropped acc: 1.06% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable' and false is not true ." Original acc: 73.30%, attacked acc: 69.67%, dropped acc: 3.63% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable' and false is not true ." Original acc: 72.94%, attacked acc: 62.85%, dropped acc: 10.09% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable' and false is not true ." Original acc: 72.63%, attacked acc: 68.16%, dropped acc: 4.47% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable' and false is not true ." Original acc: 71.65%, attacked acc: 68.78%, dropped acc: 2.87% ## textbugger Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Invites derive the most fititng anser from the context. If there isn't an answer in the context, use 'unanswerable'." Original acc: 74.10%, attacked acc: 70.95%, dropped acc: 3.16% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Uses the context, de\xf0\x9d\x9a\x9dermine the most suitable answer. Although the context doesn't contain the respond, write 'unanswerable'." Original acc: 73.30%, attacked acc: 68.25%, dropped acc: 5.05% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the give\xd5\xb8 context, provide the best p\xd0\xbessible answer. If there's no answer available in the contex\xf0\x9d\x9a\x9d, reply with 'unanswerable'." Original acc: 72.75%, attacked acc: 67.93%, dropped acc: 4.81% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-versed in the context, inviting derive the most fitti\xd5\xb8g answer. I\xf0\x9d\x9a\x8f there isn't an answering in the context, use 'unanswerable'." Original acc: 73.54%, attacked acc: 71.53%, dropped acc: 2.01% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"As a subject matter specialised, extr act the corrects answer from the context. If an answer is not present, indicate 'unanswerable'." Original acc: 73.60%, attacked acc: 69.52%, dropped acc: 4.08% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a profcient expert in the given c\xd0\xbentext, search for the most relevant answer. If the anwser cannot be unearthed, respond by saying 'unanswerable'." Original acc: 70.63%, attacked acc: 68.31%, dropped acc: 2.31% ## textfooler Original prompt: Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Please calculate the most adapt reacts from the context. Until there isn't an answer in the context, use 'unanswerable'." Original acc: 74.20%, attacked acc: 70.18%, dropped acc: 4.02% Original prompt: Using the context, determine the most suitable answer. If the context doesn't contain the answer, write 'unanswerable'. Attacked prompt: b"Using the context, identifies the most suitable answer. Although the context doesn't comprise the meet, editor 'unanswerable'." Original acc: 72.80%, attacked acc: 63.67%, dropped acc: 9.13% Original prompt: Based on the given context, provide the best possible answer. If there's no answer available in the context, reply with 'unanswerable'. Attacked prompt: b"Based on the supplied context, provide the best possible answer. If there's no responses available in the regard, riposte with 'unanswerable'." Original acc: 72.75%, attacked acc: 64.13%, dropped acc: 8.62% Original prompt: Being well-versed in the context, please derive the most fitting answer. If there isn't an answer in the context, use 'unanswerable'. Attacked prompt: b"Being well-versed in the context, please derive the most installing response. While there isn't an answer in the context, resorting 'unanswerable'." Original acc: 73.21%, attacked acc: 68.58%, dropped acc: 4.63% Original prompt: As a subject matter expert, extract the correct answer from the context. If an answer is not present, indicate 'unanswerable'. Attacked prompt: b"As a subject matter practitioners, retrieval the correct answers from the background. If an meet is not present, indicate 'unanswerable'." Original acc: 74.01%, attacked acc: 67.10%, dropped acc: 6.91% Original prompt: As a proficient expert in the given context, search for the most relevant answer. If the answer cannot be found, respond by saying 'unanswerable'. Attacked prompt: b"As a knowledge expert in the given context, find for the most relevant answering. Unless the answer cannot exist found, respond by told 'unanswerable'." Original acc: 71.93%, attacked acc: 64.13%, dropped acc: 7.79% ## translation Language: chinese, acc: 73.58%, prompt: Provide the most accurate answer from the context. If the answer is not in context, answer 'unanswerable'. Language: chinese, acc: 71.24%, prompt: Please give the most accurate answer based on the context. If you cannot find the answer, please answer 'unanswerable'. Language: chinese, acc: 68.12%, prompt: Provide the most accurate answer based on the context. If you cannot find the answer, please answer 'unanswerable'. Language: chinese, acc: 69.84%, prompt: Please provide the most accurate answer from the context. If you cannot find the answer, please answer 'unanswerable'. Language: chinese, acc: 69.40%, prompt: Give the most accurate answer based on the context. If you cannot find the answer, please answer 'unanswerable'. Language: chinese, acc: 71.55%, prompt: Please give the most accurate answer based on the context. If the answer is not in context, answer 'unanswerable'. Language: chinese, acc: 70.00%, prompt: Provide the most accurate answer from the context. If you cannot find the answer, please answer 'unanswerable'. Language: chinese, acc: 70.50%, prompt: Please give the most accurate answer based on the context. If the answer cannot be found, please answer 'unanswerable'. Language: chinese, acc: 71.54%, prompt: Provide the most accurate answer based on the context. If the answer cannot be found, please answer 'unanswerable'. Language: chinese, acc: 72.03%, prompt: Please provide the most accurate answer from the context. If the answer cannot be found, please answer 'unanswerable'. Language: french, acc: 73.61%, prompt: From the context, provide the most accurate answer. If the answer is not in context, answer with 'unanswerable'. Language: french, acc: 70.12%, prompt: From the context, give the most accurate answer. If the answer is not present in the context, answer with 'unanswerable'. Language: french, acc: 72.30%, prompt: Based on the context, provide the most accurate answer. If the answer is not in context, answer with 'unanswerable'. Language: french, acc: 71.09%, prompt: According to the context, give the most precise answer. If the answer is not present in the context, answer with 'unanswerable'. Language: french, acc: 71.15%, prompt: From the context, find the most accurate answer. If the answer is not in context, answer with 'unanswerable'. Language: french, acc: 72.82%, prompt: Based on the context, provide the most accurate answer. If the answer is not available in the context, answer with 'unanswerable'. Language: french, acc: 70.56%, prompt: According to the context, give the most precise answer. If the answer is not in the context, answer with 'unanswerable'. Language: french, acc: 72.18%, prompt: From the context, find the most accurate answer. If the answer is not present in the context, answer with 'unanswerable'. Language: french, acc: 73.26%, prompt: Based on the context, provide the most accurate answer. If the answer cannot be found in the context, answer with 'unanswerable'. Language: french, acc: 72.22%, prompt: According to the context, give the most precise answer. If the answer is not available in the context, answer with 'unanswerable'. Language: arabic, acc: 71.49%, prompt: From context, provide the most accurate answer. If not in context, please reply 'unanswerable', Language: arabic, acc: 72.74%, prompt: From context, what is the most likely outcome? If the answer is not in context, please reply 'unanswerable', Language: arabic, acc: 71.63%, prompt: From the given context, what is the key element that can be deduced? If the answer is not available in the context, please reply 'unanswerable', Language: arabic, acc: 71.36%, prompt: Based on the context given, what is the clear key idea? If the answer is not in context, please reply 'unanswerable', Language: arabic, acc: 73.34%, prompt: Based on the context, what is the most convincing explanation? If the answer is not available in the context, please reply 'unanswerable', Language: arabic, acc: 73.47%, prompt: Based on the context, what is the most likely outcome? If the answer is not available in the context, please reply 'unanswerable', Language: arabic, acc: 72.87%, prompt: Based on the context, which hypothesis is the most true? If the answer is not in context, please reply 'unanswerable', Language: arabic, acc: 72.43%, prompt: From context, what is the most apparent factor influencing? If the answer is not available in the context, please reply 'unanswerable', Language: arabic, acc: 70.61%, prompt: From context, provide the most accurate answer. If the answer is not in context, reply 'unanswerable', Language: arabic, acc: 73.14%, prompt: From context, determine the most accurate answer. If the answer is not available in context, answer 'unanswerable', Language: spanish, acc: 74.76%, prompt: Depending on the context, it provides the most precise answer. If the answer is not in context, answer with 'unanswerable'. Language: spanish, acc: 72.76%, prompt: Briefly describes the situation and provides the corresponding response. If the answer cannot be found, answer with 'unanswerable'. Language: spanish, acc: 74.11%, prompt: Given the information given, what is the most appropriate response? If the answer cannot be determined, answer with 'unanswerable'. Language: spanish, acc: 71.57%, prompt: Read the following text and give the most accurate answer. If you can't find the answer, answer with 'unanswerable'. Language: spanish, acc: 73.51%, prompt: Based on the description, what is the most accurate answer? If the answer is not found in the description, answer with 'unanswerable'. Language: spanish, acc: 71.22%, prompt: From the context provided, which response is the most appropriate? If the answer cannot be found, answer with 'unanswerable'. Language: spanish, acc: 73.82%, prompt: Analyze the following paragraph and provide the most accurate answer. If the answer is not in the paragraph, answer with 'unanswerable'. Language: spanish, acc: 73.84%, prompt: According to the information presented, what is the most precise answer? If the answer cannot be determined, answer with 'unanswerable'. Language: spanish, acc: 73.98%, prompt: After reading the excerpt, which do you think is the correct answer? If the answer cannot be discerned, answer with 'unanswerable'. Language: spanish, acc: 74.59%, prompt: Based on the context, it provides the most appropriate response. If the answer is not in context, answer with 'unanswerable'. Language: japanese, acc: 72.00%, prompt: Provide the most accurate answer from this context. If the answer isn't in the context, answer 'unanswerable'. Language: japanese, acc: 72.53%, prompt: Please provide the most appropriate answer based on the information specified in this sentence. If the answer is not in the text, answer 'unanswerable'. Language: japanese, acc: 72.01%, prompt: Please provide the most accurate answer based on the information guessed from this text. If the answer is not in the text, answer 'unanswerable'. Language: japanese, acc: 71.98%, prompt: Provide the most detailed answer based on the given context. If the answer is not in the context, answer 'unanswerable'. Language: japanese, acc: 75.36%, prompt: Consider the information derived from this context and provide the most accurate answer. If the answer is not in the context, answer 'unanswerable'. Language: japanese, acc: 73.36%, prompt: Based on this context, please provide the most appropriate answer. If the answer is not in the context, answer 'unanswerable'. Language: japanese, acc: 70.60%, prompt: Consider the information derived from the given text and provide the most detailed answer. If the answer is not in the text, please answer 'unanswerable'. Language: japanese, acc: 72.77%, prompt: Provide the most accurate answer based on the information given in this text. If the answer is not in the text, answer 'unanswerable'. Language: japanese, acc: 74.30%, prompt: Consider the information inferred from this context and provide the most appropriate answer. If the answer is not in the context, answer 'unanswerable'. Language: japanese, acc: 72.52%, prompt: Provide the most detailed answer based on this context. If the answer is not in the context, answer 'unanswerable'. Language: korean, acc: 69.63%, prompt: Give the most accurate answer in context. If the answer is not in context, respond with 'unanswerable'. Language: korean, acc: 67.46%, prompt: Please provide additional information about the facts mentioned in this sentence. If no information is available, respond with 'unanswerable'. Language: korean, acc: 72.75%, prompt: Please tell me what your question is about. If there is no context in which you can provide an answer, respond with 'unanswerable'. Language: korean, acc: 68.15%, prompt: Please explain the concept mentioned in the following sentence. If there is no information on the concept, please respond with 'unanswerable'. Language: korean, acc: 72.98%, prompt: Tell me what you're comparing to in this sentence. If nothing is compared, please respond with 'unanswerable'. Language: korean, acc: 73.32%, prompt: Please perform the actions required by the following context. If the task is not possible or if you are not clear what needs to be done, respond with 'unanswerable'. Language: korean, acc: 69.15%, prompt: Figure out what information this sentence contains. If no information is available, respond with 'unanswerable'. Language: korean, acc: 72.13%, prompt: Please give a solution to what kind of problem in the following sentence. If there is no solution, respond with 'unanswerable'. Language: korean, acc: 69.55%, prompt: Please give the cause of the incident mentioned in the context. If the cause is not clear, respond with 'unanswerable'. Language: korean, acc: 73.74%, prompt: Give expected results in the following sentences. If the result is unpredictable, respond with 'unanswerable'. # iwslt ## 10 prompts Acc: 0.24%, prompt: Convert the following sentence from its original {} language to the target language {}. Acc: 0.24%, prompt: Please perform a translation of the given sentence, converting it from {} to {}. Acc: 0.23%, prompt: Given the sentence below, perform a machine translation from {} to {}. Acc: 0.23%, prompt: Translate the following text from the source language {} to the target language {}. Acc: 0.23%, prompt: Translate the subsequent text passage from its original {} language to the {} language. Acc: 0.23%, prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Acc: 0.23%, prompt: Translate the subsequent sentence from its source language {} into the desired language {}. Acc: 0.23%, prompt: Using machine translation, convert the given sentence from {} into the {} language. Acc: 0.23%, prompt: Accurately translate the sentence from {} to {}, ensuring the meaning remains intact. Acc: 0.23%, prompt: Translate the provided sentence from {} to {} while maintaining the original meaning and context. Acc: 0.24%, prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Acc: 0.24%, prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Acc: 0.24%, prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Acc: 0.24%, prompt: As a language specialist, accurately translate the provided sentence from {} to {}. Acc: 0.24%, prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Acc: 0.24%, prompt: While acting as an experienced language expert, translate the following text from {} to {}. Acc: 0.24%, prompt: Assuming the role of an expert language translator, translate the given text from {} to {}. Acc: 0.23%, prompt: As a skilled linguist, please convert the following sentence from {} to {}. Acc: 0.23%, prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Acc: 0.23%, prompt: In the capacity of a proficient translator, please change the following sentence from {} to {}. ## bertattack Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the following sentence from its actual {} language to the net language {}. ' Original acc: 0.24%, attacked acc: 0.21%, dropped acc: 0.03% Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'in the sentence below, provide a robot translation from {} to {}. ' Original acc: 0.23%, attacked acc: 0.22%, dropped acc: 0.01% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'[UNK] consider a translation of the an sentence, converting it from {} to {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'In the application of a spanish expert, present a machine translation for the supplied sentence, changing it from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'what the responsibilities of a professional scholar, translate the subsequent text passage from {} to {}. ' Original acc: 0.24%, attacked acc: 0.17%, dropped acc: 0.07% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'an on the roles of a certified x, for a translation of the given sentence from {} to {}. ' Original acc: 0.24%, attacked acc: 0.17%, dropped acc: 0.07% ## checklist Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the following sentence from its original {} language to the target language 5XeflW1ZJc {}. ' Original acc: 0.24%, attacked acc: 0.22%, dropped acc: 0.02% Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'Given the sentence below, perform a machine translation from {} to MQnugHcaoy {}. ' Original acc: 0.23%, attacked acc: 0.22%, dropped acc: 0.01% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Please perform a translation of the given sentence, converting it from {} to IhuBIhoPGc {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to MQnugHcaoy {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.02% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to bhrRSokrfa {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'Taking on the role of a certified translator, perform a translation of the given sentence from {} to AMsRIKZniY {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% ## deepwordbug Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'Given the sentence below, perform a mRachine traCslation from {} to {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the following sentence from its orgiinal {} language to the target language {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'PleasD perform a Dranslation of the given sentence, convrting it from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the responsibilities of a professional transaltor, translate the subsequent text passage from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'In the ole of a langSage expert, perform a maMhine translation for the provided sentence, changing it from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'Taking on the rloe of a certified translator, perform a translation of the Tiven seotence from {} to {}. ' Original acc: 0.24%, attacked acc: 0.24%, dropped acc: 0.00% ## stresstest Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the following sentence from its original {} language to the target language and false is not true {}. ' Original acc: 0.24%, attacked acc: 0.24%, dropped acc: -0.00% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Please perform a translation of the given sentence, converting it from {} to and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.00% Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'Given the sentence below, perform a machine translation from {} to and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.00% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to and false is not true {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 0.24%, attacked acc: 0.24%, dropped acc: 0.01% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'Taking on the role of a certified translator, perform a translation of the given sentence from {} to and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 0.24%, attacked acc: 0.24%, dropped acc: 0.00% ## textbugger Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'P\xe2\x85\xbcease perform a translating of the giv\xd0\xb5n sentence, conversion it from {} to {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the following sentence from its original {} language to the tar get language {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.00% Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'G\xd1\x96ven the sentence below, preform a machinery translation from {} to {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assumin\xc9\xa1 the responsibilities of a professional transla\xf0\x9d\x9a\x9dor, translate the subsequent text passage from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'During the role of a language expert, per\xf0\x9d\x9a\x8form a machine translation for the provided sentence, changing it from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'Tking on the role of a certified tr\xc9\x91nslator, preform a translation of the given sentence from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% ## textfooler Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Please exert a translation of the given judgements, converting it from {} to {}. ' Original acc: 0.23%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Convert the following sentence from its original {} language to the target language {}. Attacked prompt: b'Convert the below sentence from its original {} language to the intentions language {}. ' Original acc: 0.24%, attacked acc: 0.22%, dropped acc: 0.02% Original prompt: Given the sentence below, perform a machine translation from {} to {}. Attacked prompt: b'Attributed the sentence below, meet a equipments interpreters from {} to {}. ' Original acc: 0.23%, attacked acc: 0.18%, dropped acc: 0.05% Original prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Of the features of a language expert, performed a machine translation for the presented sentence, variation it from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.01% Original prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Attacked prompt: b'Adopts on the role of a certified translator, perform a translation of the given sentence from {} to {}. ' Original acc: 0.24%, attacked acc: 0.24%, dropped acc: 0.00% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Suppose the commitment of a professional converters, translate the subsequent text passage from {} to {}. ' Original acc: 0.24%, attacked acc: 0.23%, dropped acc: 0.02% ## translation Language: chinese, acc: 0.24%, prompt: Please translate the given sentence into {} to {}. Language: chinese, acc: 0.25%, prompt: Please translate the following sentences from {} to {}. Language: chinese, acc: 0.23%, prompt: Please convert the following sentences to {} and translate to {}. Language: chinese, acc: 0.23%, prompt: Please convert the given sentence from {} to {}. Language: chinese, acc: 0.24%, prompt: Please translate the next sentence from {} to {}. Language: chinese, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: chinese, acc: 0.23%, prompt: Please translate the sentences given into {} and convert them into {}. Language: chinese, acc: 0.23%, prompt: Please convert the sentences given to {} to {}. Language: chinese, acc: 0.21%, prompt: Please translate the following sentences into {} and convert them into {}. Language: chinese, acc: 0.24%, prompt: Please change the given sentence from {} to {}. Language: french, acc: 0.23%, prompt: Please translate the given sentence, converting it from {} to {}. Language: french, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: french, acc: 0.23%, prompt: Please turn the sentence below into {}, then translate it into {}. Language: french, acc: 0.23%, prompt: Please convert the given phrase from {} to {}. Language: french, acc: 0.25%, prompt: Please translate the following sentence from {} to {}. Language: french, acc: 0.23%, prompt: Please translate the sentence below from {} to {}. Language: french, acc: 0.21%, prompt: Please translate the given sentence to {}, then convert it to {}. Language: french, acc: 0.23%, prompt: Please make a translation of the supplied sentence, transforming it from {} to {}. Language: french, acc: 0.23%, prompt: Please translate the following sentence to {}, then convert it to {}. Language: french, acc: 0.23%, prompt: Please transform the given sentence from {} to {}. Language: arabic, acc: 0.23%, prompt: Please translate the given sentence, and convert it from {} to {}, Language: arabic, acc: 0.25%, prompt: Please translate the following sentence from {} to {}, Language: arabic, acc: 0.23%, prompt: Please convert the sentence below to {}, and then translate it to {}, Language: arabic, acc: 0.23%, prompt: Please convert the given sentence from {} to {}, Language: arabic, acc: 0.25%, prompt: Please translate the following sentence from {} to {}, Language: arabic, acc: 0.24%, prompt: Please convert the sentence below from {} to {}, Language: arabic, acc: 0.22%, prompt: Please translate the given sentence to {}, then convert it to {}, Language: arabic, acc: 0.23%, prompt: Please translate the given sentence, and convert it from {} to {}, Language: arabic, acc: 0.20%, prompt: Please translate to {}, then convert to {}, Language: arabic, acc: 0.24%, prompt: Please convert the given sentence from {} to {}. Language: spanish, acc: 0.23%, prompt: Please make a translation of the provided phrase, converting it from {} to {}. Language: spanish, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: spanish, acc: 0.23%, prompt: Please convert the next sentence to {}, and then translate it to {}. Language: spanish, acc: 0.24%, prompt: Please make a translation of the given phrase, converting it from {} to {}. Language: spanish, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: spanish, acc: 0.23%, prompt: Please convert the following sentence from {} to {}. Language: spanish, acc: 0.22%, prompt: Please translate the sentence provided to {}, and then turn it to {}. Language: spanish, acc: 0.24%, prompt: Please make a translation of the following sentence, converting it from {} to {}. Language: spanish, acc: 0.23%, prompt: Please translate the next sentence to {}, and then turn it to {}. Language: spanish, acc: 0.24%, prompt: Please convert the given sentence from {} to {}. Language: japanese, acc: 0.24%, prompt: Please translate the given sentence from {} to {}. Language: japanese, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: japanese, acc: 0.23%, prompt: Please convert the following sentences into {} and translate them into {}. Language: japanese, acc: 0.23%, prompt: Please translate the given sentence by converting {} to {}. Language: japanese, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: japanese, acc: 0.25%, prompt: Please convert the following sentences from {} to {}. Language: japanese, acc: 0.22%, prompt: Translate the given sentence into {} and convert it to {}. Language: japanese, acc: 0.24%, prompt: Please translate the given sentence from {} to {}. Language: japanese, acc: 0.23%, prompt: Translate the following sentence into {} and convert it to {}. Language: japanese, acc: 0.24%, prompt: Convert the given statement from {} to {}. Language: korean, acc: 0.23%, prompt: Please translate the given sentence from {} to {}. Language: korean, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: korean, acc: 0.17%, prompt: Please translate the sentences below into {}, then {}. Language: korean, acc: 0.23%, prompt: Please translate the given sentences from {} to {}. Language: korean, acc: 0.24%, prompt: Please translate the following sentence from {} to {}. Language: korean, acc: 0.26%, prompt: Please convert the sentences below from {} to {}. Language: korean, acc: 0.21%, prompt: Please translate the given sentence into {}, then {}. Language: korean, acc: 0.23%, prompt: Please translate the given sentence from {} to {}. Language: korean, acc: 0.18%, prompt: Please translate the following sentences into {}, then {}. Language: korean, acc: 0.23%, prompt: Please convert the given sentence from {} to {}. # un_multi ## 10 prompts Acc: 0.36%, prompt: Translate the following text from the source language {} to the target language {}. Acc: 0.35%, prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Acc: 0.35%, prompt: Please perform a translation of the given sentence, converting it from {} to {}. Acc: 0.35%, prompt: Accurately translate the sentence from {} to {}, ensuring the meaning remains intact. Acc: 0.35%, prompt: Translate the subsequent sentence from its source language {} into the desired language {}. Acc: 0.35%, prompt: Translate the subsequent text passage from its original {} language to the {} language. Acc: 0.35%, prompt: Using machine translation, convert the given sentence from {} into the {} language. Acc: 0.35%, prompt: Given the sentence below, perform a machine translation from {} to {}. Acc: 0.35%, prompt: Convert the following sentence from its original {} language to the target language {}. Acc: 0.34%, prompt: Translate the provided sentence from {} to {} while maintaining the original meaning and context. Acc: 0.35%, prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Acc: 0.35%, prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Acc: 0.35%, prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Acc: 0.35%, prompt: In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}. Acc: 0.35%, prompt: In the capacity of a proficient translator, please change the following sentence from {} to {}. Acc: 0.35%, prompt: As a skilled linguist, please convert the following sentence from {} to {}. Acc: 0.35%, prompt: Assuming the role of an expert language translator, translate the given text from {} to {}. Acc: 0.35%, prompt: As a language specialist, accurately translate the provided sentence from {} to {}. Acc: 0.35%, prompt: Taking on the role of a certified translator, perform a translation of the given sentence from {} to {}. Acc: 0.35%, prompt: While acting as an experienced language expert, translate the following text from {} to {}. ## bertattack Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following area from the standard language {} to the target language {}. ' Original acc: 0.36%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'write a machine translation for the provided form, changing it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'please perform a translation of the right sentence, reducing it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'in as a dedicated translator, translate the received sentence from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the obligations of a professional language, translate the subsequent text passage from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a qualified proper, converts the target sentence from its original {} language to the english language {}. ' Original acc: 0.35%, attacked acc: 0.26%, dropped acc: 0.10% ## checklist Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following text from the source language {} to the target language 5yWbBXztUY {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Perform a machine translation for the provided sentence, changing it from {} to Kw6nrs57gH {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Please perform a translation of the given sentence, converting it from {} to bhrRSokrfa {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to AMsRIKZniY {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.02% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a qualified translator, convert the given sentence from its original {} language to the target language 5XeflW1ZJc {}. ' Original acc: 0.35%, attacked acc: 0.31%, dropped acc: 0.04% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'Acting as a professional translator, translate the provided sentence from {} to 5XeflW1ZJc {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.02% ## deepwordbug Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Plese erform a translation of the Fiven sentence, converwing it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.26%, dropped acc: 0.09% Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Preform a machine transltaion for the provide sentence, changinNg it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following temt from the sougce lanuage {} to the target language {}. ' Original acc: 0.36%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a qualified translator, convert the given sentence from its original {} language to the targte language {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'sAsuming the responsibilities of a professinal tarnslator, translate the subseqeunt text passgae from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'Acitng as a professional trsanslator, translate the provided sentence from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% ## stresstest Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Perform a machine translation for the provided sentence, changing it from {} to and false is not true {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following text from the source language {} to the target language and false is not true {}. ' Original acc: 0.36%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Please perform a translation of the given sentence, converting it from {} to and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'Acting as a professional translator, translate the provided sentence from {} to and true is true {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a qualified translator, convert the given sentence from its original {} language to the target language and false is not true {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to and false is not true {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% ## textbugger Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Pl\xd0\xb5ase perform a translation of the given sentence, convert it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Perform a machines translation for the provided sentence, ch\xc9\x91nging it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following text from the source vocabulary {} to the target language {}. ' Original acc: 0.36%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a eligible translator, convert the given sentence from its or\xd1\x96ginal {} language to the tar\xc9\xa1et language {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Presuming the responsibilities of a professional translator, translate the subseq uent writings \xd1\x80assage from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'Interim as a professional t\xe2\xb2\x85anslator, translate the provided sentence from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.00% ## textfooler Original prompt: Perform a machine translation for the provided sentence, changing it from {} to {}. Attacked prompt: b'Perform a machine transforming for the provided sentence, vary it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.02% Original prompt: Translate the following text from the source language {} to the target language {}. Attacked prompt: b'Translate the following texts from the source lingo {} to the millennium phrasing {}. ' Original acc: 0.35%, attacked acc: 0.33%, dropped acc: 0.02% Original prompt: Please perform a translation of the given sentence, converting it from {} to {}. Attacked prompt: b'Pray accomplish a translation of the given sentence, converting it from {} to {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% Original prompt: As a qualified translator, convert the given sentence from its original {} language to the target language {}. Attacked prompt: b'As a nuanced translation, metamorphosis the gave sentence from its original {} terminology to the aspirations grammar {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.02% Original prompt: Acting as a professional translator, translate the provided sentence from {} to {}. Attacked prompt: b'Acting as a expert performer, translate the provided sentence from {} to {}. ' Original acc: 0.35%, attacked acc: 0.35%, dropped acc: 0.01% Original prompt: Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}. Attacked prompt: b'Assume the roles of a professional artist, translate the later texts passage from {} to {}. ' Original acc: 0.35%, attacked acc: 0.34%, dropped acc: 0.01% ## translation Language: chinese, acc: 0.35%, prompt: Please translate the given sentence into {} to {}. Language: chinese, acc: 0.36%, prompt: Please translate the following sentences from {} to {}. Language: chinese, acc: 0.27%, prompt: Please convert the following sentences to {} and translate to {}. Language: chinese, acc: 0.35%, prompt: Please convert the given sentence from {} to {}. Language: chinese, acc: 0.35%, prompt: Please translate the next sentence from {} to {}. Language: chinese, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: chinese, acc: 0.30%, prompt: Please translate the sentences given into {} and convert them into {}. Language: chinese, acc: 0.36%, prompt: Please convert the sentences given to {} to {}. Language: chinese, acc: 0.28%, prompt: Please translate the following sentences into {} and convert them into {}. Language: chinese, acc: 0.35%, prompt: Please change the given sentence from {} to {}. Language: french, acc: 0.36%, prompt: Please translate the given sentence, converting it from {} to {}. Language: french, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: french, acc: 0.29%, prompt: Please turn the sentence below into {}, then translate it into {}. Language: french, acc: 0.35%, prompt: Please convert the given phrase from {} to {}. Language: french, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: french, acc: 0.35%, prompt: Please translate the sentence below from {} to {}. Language: french, acc: 0.28%, prompt: Please translate the given sentence to {}, then convert it to {}. Language: french, acc: 0.35%, prompt: Please make a translation of the supplied sentence, transforming it from {} to {}. Language: french, acc: 0.30%, prompt: Please translate the following sentence to {}, then convert it to {}. Language: french, acc: 0.35%, prompt: Please transform the given sentence from {} to {}. Language: arabic, acc: 0.35%, prompt: Please translate the given sentence, and convert it from {} to {}, Language: arabic, acc: 0.36%, prompt: Please translate the following sentence from {} to {}, Language: arabic, acc: 0.28%, prompt: Please convert the sentence below to {}, and then translate it to {}, Language: arabic, acc: 0.35%, prompt: Please convert the given sentence from {} to {}, Language: arabic, acc: 0.35%, prompt: Please translate the following sentence from {} to {}, Language: arabic, acc: 0.35%, prompt: Please convert the sentence below from {} to {}, Language: arabic, acc: 0.28%, prompt: Please translate the given sentence to {}, then convert it to {}, Language: arabic, acc: 0.35%, prompt: Please translate the given sentence, and convert it from {} to {}, Language: arabic, acc: 0.29%, prompt: Please translate to {}, then convert to {}, Language: arabic, acc: 0.35%, prompt: Please convert the given sentence from {} to {}. Language: spanish, acc: 0.35%, prompt: Please make a translation of the provided phrase, converting it from {} to {}. Language: spanish, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: spanish, acc: 0.26%, prompt: Please convert the next sentence to {}, and then translate it to {}. Language: spanish, acc: 0.36%, prompt: Please make a translation of the given phrase, converting it from {} to {}. Language: spanish, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: spanish, acc: 0.35%, prompt: Please convert the following sentence from {} to {}. Language: spanish, acc: 0.28%, prompt: Please translate the sentence provided to {}, and then turn it to {}. Language: spanish, acc: 0.36%, prompt: Please make a translation of the following sentence, converting it from {} to {}. Language: spanish, acc: 0.29%, prompt: Please translate the next sentence to {}, and then turn it to {}. Language: spanish, acc: 0.35%, prompt: Please convert the given sentence from {} to {}. Language: japanese, acc: 0.35%, prompt: Please translate the given sentence from {} to {}. Language: japanese, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: japanese, acc: 0.28%, prompt: Please convert the following sentences into {} and translate them into {}. Language: japanese, acc: 0.36%, prompt: Please translate the given sentence by converting {} to {}. Language: japanese, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: japanese, acc: 0.35%, prompt: Please convert the following sentences from {} to {}. Language: japanese, acc: 0.29%, prompt: Translate the given sentence into {} and convert it to {}. Language: japanese, acc: 0.35%, prompt: Please translate the given sentence from {} to {}. Language: japanese, acc: 0.30%, prompt: Translate the following sentence into {} and convert it to {}. Language: japanese, acc: 0.35%, prompt: Convert the given statement from {} to {}. Language: korean, acc: 0.35%, prompt: Please translate the given sentence from {} to {}. Language: korean, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: korean, acc: 0.26%, prompt: Please translate the sentences below into {}, then {}. Language: korean, acc: 0.36%, prompt: Please translate the given sentences from {} to {}. Language: korean, acc: 0.35%, prompt: Please translate the following sentence from {} to {}. Language: korean, acc: 0.35%, prompt: Please convert the sentences below from {} to {}. Language: korean, acc: 0.28%, prompt: Please translate the given sentence into {}, then {}. Language: korean, acc: 0.35%, prompt: Please translate the given sentence from {} to {}. Language: korean, acc: 0.25%, prompt: Please translate the following sentences into {}, then {}. Language: korean, acc: 0.35%, prompt: Please convert the given sentence from {} to {}. # math ## 10 prompts Acc: 23.12%, prompt: Compute the result of this math task concerning {}. Acc: 23.12%, prompt: Resolve the following mathematical question associated with {}. Acc: 22.50%, prompt: Calculate the answer to the following math query about {}. Acc: 21.88%, prompt: Solve the following math question about {}. Acc: 21.88%, prompt: Determine the solution to this mathematical problem related to {}. Acc: 21.88%, prompt: Find the solution for this mathematical challenge with {}. Acc: 21.88%, prompt: Work out the answer to this math problem featuring {}. Acc: 21.88%, prompt: Obtain the result for this math question regarding {}. Acc: 21.88%, prompt: Evaluate the following mathematical problem that includes {}. Acc: 20.62%, prompt: Figure out the solution for the following mathematical task with {}. Acc: 23.12%, prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Acc: 22.50%, prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Acc: 22.50%, prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Acc: 21.88%, prompt: As a mathematics consultant, resolve the following problem associated with {}. Acc: 21.25%, prompt: Pretend you are a math teacher and solve this mathematical question concerning {}. Acc: 20.62%, prompt: Taking on the role of a math specialist, work out the solution for this mathematical question featuring {}. Acc: 20.00%, prompt: In the position of a mathematics coach, figure out the answer to this math challenge involving {}. Acc: 20.00%, prompt: In the shoes of a mathematics master, obtain the result for this math question that includes {}. Acc: 19.38%, prompt: Act as a math expert and compute the result of the following mathematical task about {}. Acc: 18.12%, prompt: As a math guru, evaluate the following mathematical task regarding {}. ## bertattack Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'Compute the degree of this math task concerning {}. ' Original acc: 22.50%, attacked acc: 3.12%, dropped acc: 19.38% Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Calculate the answer to the this math query about {}. ' Original acc: 21.88%, attacked acc: 20.00%, dropped acc: 1.87% Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Resolve the catch mathematical question associated with {}. ' Original acc: 22.50%, attacked acc: 20.00%, dropped acc: 2.50% Acc: 22.50%, prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math career and determine the solution to the following problem with {}. ' Original acc: 22.50%, attacked acc: 19.38%, dropped acc: 3.12% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'as a mathematics instructor, evaluate the reply to the following problem and to {}. ' Original acc: 22.50%, attacked acc: 7.50%, dropped acc: 15.00% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'.inby a mathematics advisor, check the proof to the following math problem with {}. ' Original acc: 22.50%, attacked acc: 8.12%, dropped acc: 14.37% ## checklist Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Calculate the answer to the following math query about 5XeflW1ZJc {}. ' Original acc: 23.12%, attacked acc: 19.38%, dropped acc: 3.75% Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Resolve the following mathematical question associated with CuVpbbkC4c {}. ' Original acc: 23.12%, attacked acc: 22.50%, dropped acc: 0.63% Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'Compute the result of this math task concerning b74X5IVXQY {}. ' Original acc: 23.12%, attacked acc: 20.62%, dropped acc: 2.50% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'As a mathematics instructor, calculate the answer to the following problem related to AMsRIKZniY {}. ' Original acc: 22.50%, attacked acc: 21.25%, dropped acc: 1.25% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'Embodying a mathematics advisor, find the answer to the following math problem with lWJoGGRdjv {}. ' Original acc: 21.25%, attacked acc: 21.25%, dropped acc: 0.00% Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math mentor and determine the solution to the following problem with BWmsLx8xOA {}. ' Original acc: 22.50%, attacked acc: 21.88%, dropped acc: 0.63% ## deepwordbug Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'ComSpute the resut of this math task concerning {}. ' Original acc: 21.88%, attacked acc: 21.25%, dropped acc: 0.63% Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Resolve the following mathematical qestion associated with {}. ' Original acc: 22.50%, attacked acc: 20.62%, dropped acc: 1.88% Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Calculate the ansewr to the followiJg math query about {}. ' Original acc: 23.12%, attacked acc: 18.75%, dropped acc: 4.38% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'As a mathematics insrtuctor, calculate the answer to the following problem Vrelated to {}. ' Original acc: 23.12%, attacked acc: 20.62%, dropped acc: 2.50% Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math entor and determine the solution to the following problem with {}. ' Original acc: 22.50%, attacked acc: 20.62%, dropped acc: 1.88% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'Embodying a mathematics advisor, find the answer to the following math problem with {}. ' Original acc: 21.25%, attacked acc: 21.25%, dropped acc: 1.25% ## stresstest Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Calculate the answer to the following math query about and true is true {}. ' Original acc: 22.50%, attacked acc: 20.62%, dropped acc: 1.88% Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'Compute the result of this math task concerning and true is true {}. ' Original acc: 21.88%, attacked acc: 21.88%, dropped acc: 0.00% Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Resolve the following mathematical question associated with and true is true {}. ' Original acc: 23.12%, attacked acc: 20.00%, dropped acc: 3.12% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'As a mathematics instructor, calculate the answer to the following problem related to and false is not true {}. ' Original acc: 22.50%, attacked acc: 20.62%, dropped acc: 1.88% Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math mentor and determine the solution to the following problem with and true is true and true is true and true is true and true is true and true is true {}. ' Original acc: 22.50%, attacked acc: 21.88%, dropped acc: 0.63% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'Embodying a mathematics advisor, find the answer to the following math problem with and true is true {}. ' Original acc: 21.88%, attacked acc: 21.25%, dropped acc: 0.63% ## textbugger Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Resol ve the following mathematical question associated with {}. ' Original acc: 22.50%, attacked acc: 21.88%, dropped acc: 0.63% Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Calculate the answer to the \xf0\x9d\x9a\x8following math query about {}. ' Original acc: 23.12%, attacked acc: 20.62%, dropped acc: 2.50% Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'Compute the result of this math task relating {}. ' Original acc: 22.50%, attacked acc: 21.25%, dropped acc: 1.25% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'Since a mathematics istructor, calculate the answer to the following problem related to {}. ' Original acc: 22.50%, attacked acc: 19.38%, dropped acc: 3.12% Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math mentors and determine the solution to the following problems with {}. ' Original acc: 22.50%, attacked acc: 19.38%, dropped acc: 3.12% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'Embody\xd1\x96ng a math\xd0\xb5matics advisor, find the answer to the following math problems with {}. ' Original acc: 21.88%, attacked acc: 19.38%, dropped acc: 2.50% ## textfooler Original prompt: Calculate the answer to the following math query about {}. Attacked prompt: b'Derivation the rebuttal to the below math seeing about {}. ' Original acc: 22.50%, attacked acc: 3.75%, dropped acc: 18.75% Original prompt: Compute the result of this math task concerning {}. Attacked prompt: b'Estimate the accomplishments of this math task concerning {}. ' Original acc: 23.75%, attacked acc: 1.88%, dropped acc: 21.88% Original prompt: Resolve the following mathematical question associated with {}. Attacked prompt: b'Corrects the following mathematic issue embroiled with {}. ' Original acc: 22.50%, attacked acc: 15.00%, dropped acc: 7.50% Original prompt: As a mathematics instructor, calculate the answer to the following problem related to {}. Attacked prompt: b'As a computations instructor, derive the rebuttal to the following hassles links to {}. ' Original acc: 22.50%, attacked acc: 1.88%, dropped acc: 20.62% Original prompt: Imagine being a math mentor and determine the solution to the following problem with {}. Attacked prompt: b'Imagine being a math councillors and determine the solution to the following problem with {}. ' Original acc: 23.12%, attacked acc: 19.38%, dropped acc: 3.75% Original prompt: Embodying a mathematics advisor, find the answer to the following math problem with {}. Attacked prompt: b'Embodying a computation advisor, find the rebuttal to the later math phenomenon with {}. ' Original acc: 22.50%, attacked acc: 10.00%, dropped acc: 12.50% ## translation Language: chinese, acc: 22.50%, prompt: Calculate the answer to a mathematical question about {}. Language: chinese, acc: 23.75%, prompt: Calculate the value of {}. Language: chinese, acc: 20.00%, prompt: Please calculate the result of {}. Language: chinese, acc: 21.25%, prompt: According to {}, solve the following mathematical problem. Language: chinese, acc: 19.38%, prompt: Do the math with {}. Language: chinese, acc: 20.00%, prompt: Solve mathematical problems involving {}. Language: chinese, acc: 21.25%, prompt: Please calculate the answer involving {}. Language: chinese, acc: 21.88%, prompt: Solve for the following values based on {}. Language: chinese, acc: 21.25%, prompt: Calculate the following mathematical tasks using {}. Language: chinese, acc: 22.50%, prompt: Calculate the answer to the {} related question. Language: french, acc: 22.50%, prompt: Calculate the answer to the following mathematical question concerning {}. Language: french, acc: 20.62%, prompt: Calculate the result of {}. Language: french, acc: 21.25%, prompt: Please calculate the value of {}. Language: french, acc: 20.62%, prompt: According to {}, solve the following mathematical problem. Language: french, acc: 19.38%, prompt: Perform mathematical calculations with {}. Language: french, acc: 20.00%, prompt: Solve the mathematical problem involving {}. Language: french, acc: 21.25%, prompt: Please calculate the answer related to {}. Language: french, acc: 25.00%, prompt: According to {}, set the following value. Language: french, acc: 20.62%, prompt: Perform the following mathematical task using {}. Language: french, acc: 21.88%, prompt: Calculate the answer to the questions related to {}. Language: arabic, acc: 21.88%, prompt: Compute the answer to the next mathematical question about {}. Language: arabic, acc: 21.25%, prompt: Calculate {}. Language: arabic, acc: 21.88%, prompt: Please calculate {}. Language: arabic, acc: 20.62%, prompt: According to {}, solve the following mathematical problem. Language: arabic, acc: 20.00%, prompt: Do mathematical calculations using {}. Language: arabic, acc: 22.50%, prompt: A solution to the mathematical problem involving {}. Language: arabic, acc: 21.88%, prompt: Please calculate the answer regarding {}. Language: arabic, acc: 23.75%, prompt: According to {}, determine the next value. Language: arabic, acc: 21.25%, prompt: DO THE NEXT MATHEMATICAL JOB USING {}. Language: arabic, acc: 22.50%, prompt: Calculate the answer to questions related to {}. Language: spanish, acc: 20.00%, prompt: Compute the answer to the following mathematical question on {}. Language: spanish, acc: 21.25%, prompt: Compute the result of {}. Language: spanish, acc: 21.25%, prompt: Please calculate the value of {}. Language: spanish, acc: 25.00%, prompt: As {}, it solves the following mathematical problem. Language: spanish, acc: 22.50%, prompt: Performs mathematical calculations using {}. Language: spanish, acc: 20.00%, prompt: Solve the mathematical problem involving {}. Language: spanish, acc: 21.88%, prompt: Please calculate the answer related to {}. Language: spanish, acc: 25.00%, prompt: As {}, determine the next value. Language: spanish, acc: 20.62%, prompt: Perform the following mathematical task using {}. Language: spanish, acc: 22.50%, prompt: Compute the answer to questions related to {}. Language: japanese, acc: 21.88%, prompt: Calculate the answers to the math questions about {}. Language: japanese, acc: 23.75%, prompt: Calculate the value of {}. Language: japanese, acc: 21.88%, prompt: Please find the answer to {}. Language: japanese, acc: 20.00%, prompt: Based on {}, please solve the following mathematical problems. Language: japanese, acc: 21.88%, prompt: Use {} to perform mathematical calculations. Language: japanese, acc: 21.88%, prompt: Please solve the math problem that contains {}. Language: japanese, acc: 18.75%, prompt: Please calculate the answers related to {}. Language: japanese, acc: 21.25%, prompt: Based on {}, find the following values: Language: japanese, acc: 22.50%, prompt: Use {} to solve the following mathematical problem. Language: japanese, acc: 21.88%, prompt: Please calculate the answers to the questions related to {}. Language: korean, acc: 20.62%, prompt: Calculate the answer of the following math problem to {}. Language: korean, acc: 21.25%, prompt: Calculate the result of {}. Language: korean, acc: 21.88%, prompt: Please calculate the value of {}. Language: korean, acc: 21.25%, prompt: Work out the following math problems according to {}. Language: korean, acc: 21.25%, prompt: Use {} to proceed with mathematical calculations. Language: korean, acc: 20.00%, prompt: Work out a math problem involving {}. Language: korean, acc: 22.50%, prompt: Please calculate the answer to {}. Language: korean, acc: 20.62%, prompt: Try to get the following values according to {}. Language: korean, acc: 20.62%, prompt: Work out the next math task using {}. Language: korean, acc: 21.88%, prompt: Calculate the answer of the problem involving {}.