Accuracy is the proportion of correct predictions among the total number of cases processed. It can be computed with: Accuracy = (TP + TN) / (TP + TN + FP + FN) TP: True positive TN: True negative FP: False positive FN: False negative
BERTScore leverages the pre-trained contextual embeddings from BERT and matches words in candidate and reference sentences by cosine similarity. It has been shown to correlate with human judgment on sentence-level and system-level evaluation. Moreover, BERTScore computes precision, recall, and F1 measure, which can be useful for evaluating different language generation tasks. See the `README.md` file at [https://github.com/Tiiiger/bert_score](https://github.com/Tiiiger/bert_score) for more information.
BLEU (bilingual evaluation understudy) is an algorithm for evaluating the quality of text which has been machine-translated from one natural language to another. Quality is considered to be the correspondence between a machine's output and that of a human: "the closer a machine translation is to a professional human translation, the better it is" – this is the central idea behind BLEU. BLEU was one of the first metrics to claim a high correlation with human judgements of quality, and remains one of the most popular automated and inexpensive metrics. Scores are calculated for individual translated segments—generally sentences—by comparing them with a set of good quality reference translations. Those scores are then averaged over the whole corpus to reach an estimate of the translation's overall quality. Intelligibility or grammatical correctness are not taken into account. BLEU's output is always a number between 0 and 1. This value indicates how similar the candidate text is to the reference texts, with values closer to 1 representing more similar texts. Few human translations will attain a score of 1, since this would indicate that the candidate is identical to one of the reference translations. For this reason, it is not necessary to attain a score of 1. Because there are more opportunities to match, adding additional reference translations will increase the BLEU score.
BLEURT a learnt evaluation metric for Natural Language Generation. It is built using multiple phases of transfer learning starting from a pretrained BERT model (Devlin et al. 2018) and then employing another pre-training phrase using synthetic data. Finally it is trained on WMT human annotations. You may run BLEURT out-of-the-box or fine-tune it for your specific application (the latter is expected to perform better). See the [README.md] file at https://github.com/google-research/bleurt for more information.
Character error rate (CER) is a common metric of the performance of an automatic speech recognition system. CER is similar to Word Error Rate (WER), but operates on character instead of word. Please refer to docs of WER for further information. Character error rate can be computed as: CER = (S + D + I) / N = (S + D + I) / (S + D + C) where S is the number of substitutions, D is the number of deletions, I is the number of insertions, C is the number of correct characters, N is the number of characters in the reference (N=S+D+C). CER's output is not always a number between 0 and 1, in particular when there is a high number of insertions. This value is often associated to the percentage of characters that were incorrectly predicted. The lower the value, the better the performance of the ASR system with a CER of 0 being a perfect score.
Crosslingual Optimized Metric for Evaluation of Translation (COMET) is an open-source framework used to train Machine Translation metrics that achieve high levels of correlation with different types of human judgments (HTER, DA's or MQM). With the release of the framework the authors also released fully trained models that were used to compete in the WMT20 Metrics Shared Task achieving SOTA in that years competition. See the [README.md] file at https://unbabel.github.io/COMET/html/models.html for more information.
CoVal is a coreference evaluation tool for the CoNLL and ARRAU datasets which implements of the common evaluation metrics including MUC [Vilain et al, 1995], B-cubed [Bagga and Baldwin, 1998], CEAFe [Luo et al., 2005], LEA [Moosavi and Strube, 2016] and the averaged CoNLL score (the average of the F1 values of MUC, B-cubed and CEAFe) [Denis and Baldridge, 2009a; Pradhan et al., 2011]. This wrapper of CoVal currently only work with CoNLL line format: The CoNLL format has one word per line with all the annotation for this word in column separated by spaces: Column Type Description 1 Document ID This is a variation on the document filename 2 Part number Some files are divided into multiple parts numbered as 000, 001, 002, ... etc. 3 Word number 4 Word itself This is the token as segmented/tokenized in the Treebank. Initially the *_skel file contain the placeholder [WORD] which gets replaced by the actual token from the Treebank which is part of the OntoNotes release. 5 Part-of-Speech 6 Parse bit This is the bracketed structure broken before the first open parenthesis in the parse, and the word/part-of-speech leaf replaced with a *. The full parse can be created by substituting the asterix with the "([pos] [word])" string (or leaf) and concatenating the items in the rows of that column. 7 Predicate lemma The predicate lemma is mentioned for the rows for which we have semantic role information. All other rows are marked with a "-" 8 Predicate Frameset ID This is the PropBank frameset ID of the predicate in Column 7. 9 Word sense This is the word sense of the word in Column 3. 10 Speaker/Author This is the speaker or author name where available. Mostly in Broadcast Conversation and Web Log data. 11 Named Entities These columns identifies the spans representing various named entities. 12:N Predicate Arguments There is one column each of predicate argument structure information for the predicate mentioned in Column 7. N Coreference Coreference chain information encoded in a parenthesis structure. More informations on the format can be found here (section "*_conll File Format"): http://www.conll.cemantix.org/2012/data.html Details on the evaluation on CoNLL can be found here: https://github.com/ns-moosavi/coval/blob/master/conll/README.md CoVal code was written by @ns-moosavi. Some parts are borrowed from https://github.com/clarkkev/deep-coref/blob/master/evaluation.py The test suite is taken from https://github.com/conll/reference-coreference-scorers/ Mention evaluation and the test suite are added by @andreasvc. Parsing CoNLL files is developed by Leo Born.
This metric wrap the official scoring script for version 1 of the Contract Understanding Atticus Dataset (CUAD). Contract Understanding Atticus Dataset (CUAD) v1 is a corpus of more than 13,000 labels in 510 commercial legal contracts that have been manually labeled to identify 41 categories of important clauses that lawyers look for when reviewing contracts in connection with corporate transactions.
The F1 score is the harmonic mean of the precision and recall. It can be computed with: F1 = 2 * (precision * recall) / (precision + recall)
The GLEU metric is a variant of BLEU proposed for evaluating grammatical error corrections using n-gram overlap with a set of reference sentences, as opposed to precision/recall of specific annotated errors (Napoles et al., 2015). GLEU hews more closely to human judgments than the rankings produced by metrics such as MaxMatch and I-measure. The present metric is the second version of GLEU (Napoles et al., 2016) modified to address problems that arise when using an increasing number of reference sets. The modified metric does not require tuning and is recommended to be used instead of the original version.
GLUE, the General Language Understanding Evaluation benchmark (https://gluebenchmark.com/) is a collection of resources for training, evaluating, and analyzing natural language understanding systems.
IndicGLUE is a natural language understanding benchmark for Indian languages. It contains a wide variety of tasks and covers 11 major Indian languages - as, bn, gu, hi, kn, ml, mr, or, pa, ta, te.
Compute the Matthews correlation coefficient (MCC) The Matthews correlation coefficient is used in machine learning as a measure of the quality of binary and multiclass classifications. It takes into account true and false positives and negatives and is generally regarded as a balanced measure which can be used even if the classes are of very different sizes. The MCC is in essence a correlation coefficient value between -1 and +1. A coefficient of +1 represents a perfect prediction, 0 an average random prediction and -1 an inverse prediction. The statistic is also known as the phi coefficient. [source: Wikipedia]
METEOR, an automatic metric for machine translation evaluation that is based on a generalized concept of unigram matching between the machine-produced translation and human-produced reference translations. Unigrams can be matched based on their surface forms, stemmed forms, and meanings; furthermore, METEOR can be easily extended to include more advanced matching strategies. Once all generalized unigram matches between the two strings have been found, METEOR computes a score for this matching using a combination of unigram-precision, unigram-recall, and a measure of fragmentation that is designed to directly capture how well-ordered the matched words in the machine translation are in relation to the reference. METEOR gets an R correlation value of 0.347 with human evaluation on the Arabic data and 0.331 on the Chinese data. This is shown to be an improvement on using simply unigram-precision, unigram-recall and their harmonic F1 combination.
Pearson correlation coefficient and p-value for testing non-correlation. The Pearson correlation coefficient measures the linear relationship between two datasets. The calculation of the p-value relies on the assumption that each dataset is normally distributed. Like other correlation coefficients, this one varies between -1 and +1 with 0 implying no correlation. Correlations of -1 or +1 imply an exact linear relationship. Positive correlations imply that as x increases, so does y. Negative correlations imply that as x increases, y decreases. The p-value roughly indicates the probability of an uncorrelated system producing datasets that have a Pearson correlation at least as extreme as the one computed from these datasets.
Precision is the fraction of the true examples among the predicted examples. It can be computed with: Precision = TP / (TP + FP) TP: True positive FP: False positive
Recall is the fraction of the total amount of relevant examples that were actually retrieved. It can be computed with: Precision = TP / (TP + FN) TP: True positive FN: False negative
ROUGE, or Recall-Oriented Understudy for Gisting Evaluation, is a set of metrics and a software package used for evaluating automatic summarization and machine translation software in natural language processing. The metrics compare an automatically produced summary or translation against a reference or a set of references (human-produced) summary or translation. Note that ROUGE is case insensitive, meaning that upper case letters are treated the same way as lower case letters. This metrics is a wrapper around Google Research reimplementation of ROUGE: https://github.com/google-research/google-research/tree/master/rouge
SacreBLEU provides hassle-free computation of shareable, comparable, and reproducible BLEU scores. Inspired by Rico Sennrich's `multi-bleu-detok.perl`, it produces the official WMT scores but works with plain text. It also knows all the standard test sets and handles downloading, processing, and tokenization for you. See the [README.md] file at https://github.com/mjpost/sacreBLEU for more information.
SARI is a metric used for evaluating automatic text simplification systems. The metric compares the predicted simplified sentences against the reference and the source sentences. It explicitly measures the goodness of words that are added, deleted and kept by the system. Sari = (F1_add + F1_keep + P_del) / 3 where F1_add: n-gram F1 score for add operation F1_keep: n-gram F1 score for keep operation P_del: n-gram precision score for delete operation n = 4, as in the original paper. This implementation is adapted from Tensorflow's tensor2tensor implementation . It has two differences with the original GitHub  implementation: (1) Defines 0/0=1 instead of 0 to give higher scores for predictions that match a target exactly. (2) Fixes an alleged bug  in the keep score computation.  https://github.com/cocoxu/simplification/blob/master/SARI.py (commit 0210f15)  https://github.com/cocoxu/simplification/issues/6  https://github.com/tensorflow/tensor2tensor/blob/master/tensor2tensor/utils/sari_hook.py
seqeval is a Python framework for sequence labeling evaluation. seqeval can evaluate the performance of chunking tasks such as named-entity recognition, part-of-speech tagging, semantic role labeling and so on. This is well-tested by using the Perl script conlleval, which can be used for measuring the performance of a system that has processed the CoNLL-2000 shared task data. seqeval supports following formats: IOB1 IOB2 IOE1 IOE2 IOBES See the [README.md] file at https://github.com/chakki-works/seqeval for more information.
Calculate a Spearman correlation coefficient with associated p-value. The Spearman rank-order correlation coefficient is a nonparametric measure of the monotonicity of the relationship between two datasets. Unlike the Pearson correlation, the Spearman correlation does not assume that both datasets are normally distributed. Like other correlation coefficients, this one varies between -1 and +1 with 0 implying no correlation. Correlations of -1 or +1 imply an exact monotonic relationship. Positive correlations imply that as x increases, so does y. Negative correlations imply that as x increases, y decreases. The p-value roughly indicates the probability of an uncorrelated system producing datasets that have a Spearman correlation at least as extreme as the one computed from these datasets. The p-values are not entirely reliable but are probably reasonable for datasets larger than 500 or so.
This metric wrap the official scoring script for version 1 of the Stanford Question Answering Dataset (SQuAD). Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable.
This metric wrap the official scoring script for version 2 of the Stanford Question Answering Dataset (SQuAD). Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable. SQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones. To do well on SQuAD2.0, systems must not only answer questions when possible, but also determine when no answer is supported by the paragraph and abstain from answering.
SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after GLUE with a new set of more difficult language understanding tasks, improved resources, and a new public leaderboard.
Word error rate (WER) is a common metric of the performance of an automatic speech recognition system. The general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort. This problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate. Word error rate can then be computed as: WER = (S + D + I) / N = (S + D + I) / (S + D + C) where S is the number of substitutions, D is the number of deletions, I is the number of insertions, C is the number of correct words, N is the number of words in the reference (N=S+D+C). WER's output is always a number between 0 and 1. This value indicates the percentage of words that were incorrectly predicted. The lower the value, the better the performance of the ASR system with a WER of 0 being a perfect score.
WIKI_SPLIT is the combination of three metrics SARI, EXACT and SACREBLEU It can be used to evaluate the quality of machine-generated texts.
XNLI is a subset of a few thousand examples from MNLI which has been translated into a 14 different languages (some low-ish resource). As with MNLI, the goal is to predict textual entailment (does sentence A imply/contradict/neither sentence B) and is a classification task (given two sentences, predict one of three labels).