before_sent
stringlengths 13
1.44k
| before_sent_with_intent
stringlengths 25
1.45k
| after_sent
stringlengths 0
1.41k
| labels
stringclasses 6
values | doc_id
stringlengths 4
10
| revision_depth
int64 1
4
|
|---|---|---|---|---|---|
One of the main conclusions of our analysis is that BERT captures high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | <clarity> One of the main conclusions of our analysis is that BERT captures high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | One of the main conclusions of our analysis is that BERT can accurately capture high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | clarity | 2008.11608 | 2 |
One of the main conclusions of our analysis is that BERT captures high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | <clarity> One of the main conclusions of our analysis is that BERT captures high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | One of the main conclusions of our analysis is that BERT captures high-level sense distinctions , even when a limited number of examples is available for each word sense. | clarity | 2008.11608 | 2 |
In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | <clarity> In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | In fact, the simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | clarity | 2008.11608 | 2 |
In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | <clarity> In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | In fact, a simple feature extraction strategy of averaging contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | clarity | 2008.11608 | 2 |
In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | <clarity> In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements beyond this small number of examples . | In fact, a simple feature extraction strategy based on the averaging of contextualized embeddings proves robust even using only three training sentences per word sense, with minimal improvements obtained by increasing the size of this training data . | clarity | 2008.11608 | 2 |
Our goal is to construct mathematical operations that combine non-determinism measured from quantum randomness with computational determinism so that non-mechanistic behavior is preserved in the computation. | <fluency> Our goal is to construct mathematical operations that combine non-determinism measured from quantum randomness with computational determinism so that non-mechanistic behavior is preserved in the computation. | Our goal is to construct mathematical operations that combine indeterminism measured from quantum randomness with computational determinism so that non-mechanistic behavior is preserved in the computation. | fluency | 2009.03996 | 1 |
Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the objective is for the operations to preserve bi-immunity. | <clarity> Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the objective is for the operations to preserve bi-immunity. | Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the operations preserve bi-immunity. | clarity | 2009.03996 | 1 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the symmetric group on the natural numbers. | <meaning-changed> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the symmetric group on the natural numbers. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group on the natural numbers. | meaning-changed | 2009.03996 | 1 |
The structure of this new subgroup is unknown. | <meaning-changed> The structure of this new subgroup is unknown. | We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. The complete structure of this new subgroup is unknown. | meaning-changed | 2009.03996 | 1 |
The structure of this new subgroup is unknown. | <meaning-changed> The structure of this new subgroup is unknown. | The structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | meaning-changed | 2009.03996 | 1 |
Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the operations preserve bi-immunity. | <clarity> Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the operations preserve bi-immunity. | Formally, some results about operations applied to computably enumerable (c.e.) and bi-immune sets are proven here, where the objective is for the operations to preserve bi-immunity. | clarity | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <others> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group (Sym(mathbb } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | others | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <others> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group N} on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | others | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <others> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group })) on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | others | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <others> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers mathbb } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | others | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <others> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . N} We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | others | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <meaning-changed> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . }. This new uncountable subgroup is called the bi-immune symmetric group. We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | meaning-changed | 2009.03996 | 2 |
While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | <meaning-changed> While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that this new subgroup contains the bounded symmetric group on the natural numbers, and consequently is highly transitive. | While developing rearrangement operations on the natural numbers, we discovered that the bi-immune rearrangements generate an uncountable subgroup of the infinite symmetric group } on the natural numbers . } We show that the bi-immune symmetric group contains the finitary symmetric group on the natural numbers, and consequently is highly transitive. | meaning-changed | 2009.03996 | 2 |
} The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | <meaning-changed> } The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | Furthermore, the bi-immune symmetric group is dense in Sym(mathbb } The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | meaning-changed | 2009.03996 | 2 |
} The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | <others> } The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | N} The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | others | 2009.03996 | 2 |
} The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | <meaning-changed> } The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | }) with respect to the pointwise convergence topology. The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | meaning-changed | 2009.03996 | 2 |
} The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | <clarity> } The complete structure of this new subgroup and its subgroups generated by one or more bi-immune rearrangements is unknown. | } The complete structure of the bi-immune symmetric group and its subgroups generated by one or more bi-immune rearrangements is unknown. | clarity | 2009.03996 | 2 |
Extensive experiments demonstrate that FILTER achieves new state of the art (77.0 on average) on the challenging multilingual multi-task benchmark, XTREME . | <meaning-changed> Extensive experiments demonstrate that FILTER achieves new state of the art (77.0 on average) on the challenging multilingual multi-task benchmark, XTREME . | Extensive experiments demonstrate that FILTER achieves new state of the art on two challenging multilingual multi-task benchmark, XTREME . | meaning-changed | 2009.05166 | 1 |
Extensive experiments demonstrate that FILTER achieves new state of the art (77.0 on average) on the challenging multilingual multi-task benchmark, XTREME . | <meaning-changed> Extensive experiments demonstrate that FILTER achieves new state of the art (77.0 on average) on the challenging multilingual multi-task benchmark, XTREME . | Extensive experiments demonstrate that FILTER achieves new state of the art (77.0 on average) on the challenging multilingual multi-task benchmarks, XTREME and XGLUE . | meaning-changed | 2009.05166 | 1 |
However, when applied to zero-shot cross-lingual transfer tasks, most existing methods use only single-language input for LM finetuning, without leveraging the intrinsic cross-lingual alignment between different languages that is essential for multilingual tasks. | <clarity> However, when applied to zero-shot cross-lingual transfer tasks, most existing methods use only single-language input for LM finetuning, without leveraging the intrinsic cross-lingual alignment between different languages that is essential for multilingual tasks. | However, when applied to zero-shot cross-lingual transfer tasks, most existing methods use only single-language input for LM finetuning, without leveraging the intrinsic cross-lingual alignment between different languages that proves essential for multilingual tasks. | clarity | 2009.05166 | 2 |
Specifically, FILTER first encodes text input in the source language and its translation in the target language independently in the shallow layers, then performs cross-lingual fusion to extract multilingual knowledge in the intermediate layers, and finally performs further language-specific encoding. | <fluency> Specifically, FILTER first encodes text input in the source language and its translation in the target language independently in the shallow layers, then performs cross-lingual fusion to extract multilingual knowledge in the intermediate layers, and finally performs further language-specific encoding. | Specifically, FILTER first encodes text input in the source language and its translation in the target language independently in the shallow layers, then performs cross-language fusion to extract multilingual knowledge in the intermediate layers, and finally performs further language-specific encoding. | fluency | 2009.05166 | 2 |
For better model scalability , we further propose an additional KL-divergence self-teaching loss for model training, based on auto-generated soft pseudo-labels for translated text in the target language. | <coherence> For better model scalability , we further propose an additional KL-divergence self-teaching loss for model training, based on auto-generated soft pseudo-labels for translated text in the target language. | To tackle this issue , we further propose an additional KL-divergence self-teaching loss for model training, based on auto-generated soft pseudo-labels for translated text in the target language. | coherence | 2009.05166 | 2 |
We propose a novel method to sparsify attention in the Transformer model by learning to select the most-informative token representations , thus leveraging the model's information bottleneck with twofold strength. A careful analysis shows that the contextualization of encoded representations in our model is significantly more effective than in the original Transformer. We achieve a notable reduction in memory usage due to an improved differentiable top-k operator , making the model suitable to process long documents, as shown on an example of a summarization task . | <clarity> We propose a novel method to sparsify attention in the Transformer model by learning to select the most-informative token representations , thus leveraging the model's information bottleneck with twofold strength. A careful analysis shows that the contextualization of encoded representations in our model is significantly more effective than in the original Transformer. We achieve a notable reduction in memory usage due to an improved differentiable top-k operator , making the model suitable to process long documents, as shown on an example of a summarization task . | We propose a novel method to sparsify attention in the Transformer model by learning to select the most-informative token representations during the training process, thus focusing on task-specific parts of the input. A reduction of quadratic time and memory complexity to sublinear was achieved due to a robust differentiable top-k operator , making the model suitable to process long documents, as shown on an example of a summarization task . | clarity | 2009.05169 | 1 |
We achieve a notable reduction in memory usage due to an improved differentiable top-k operator , making the model suitable to process long documents, as shown on an example of a summarization task . | <meaning-changed> We achieve a notable reduction in memory usage due to an improved differentiable top-k operator , making the model suitable to process long documents, as shown on an example of a summarization task . | We achieve a notable reduction in memory usage due to an improved differentiable top-k operator . For example, our experiments on a challenging summarization task of long documents show that our method is much faster and up to 16 times more memory efficient while significantly outperforming both dense and state-of-the-art sparse transformer models. The method can be effortlessly applied to many models used in NLP and CV, simultaneously with other improvements since representation pooling addresses a different aspect of the attention's complexity problem . | meaning-changed | 2009.05169 | 1 |
The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | <fluency> The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | The second edition of "Semantic Relations Between Nominals" by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | fluency | 2009.05426 | 1 |
The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | <fluency> The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha will be published by URLan & Claypool . | fluency | 2009.05426 | 1 |
The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | <meaning-changed> The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published early in 2021 by URLan & Claypool . | meaning-changed | 2009.05426 | 1 |
The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | <meaning-changed> The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool . | The second edition of "Semantic Relations Between Nominals" ( by Vivi Nastase, Stan Szpakowicz, Preslav Nakov and Diarmuid 'O S'eaghdha ) will be published by URLan & Claypool URL | meaning-changed | 2009.05426 | 1 |
A new Chapter 5 of the book discusses relation classification/extraction in the deep-learning paradigm which arose after the first edition appeared. | <meaning-changed> A new Chapter 5 of the book discusses relation classification/extraction in the deep-learning paradigm which arose after the first edition appeared. | A new Chapter 5 of the book , by Vivi Nastase and Stan Szpakowicz, discusses relation classification/extraction in the deep-learning paradigm which arose after the first edition appeared. | meaning-changed | 2009.05426 | 1 |
This is a preview of Chapter 5, made public by the kind permission of URLan & Claypool. | <coherence> This is a preview of Chapter 5, made public by the kind permission of URLan & Claypool. | This is Chapter 5, made public by the kind permission of URLan & Claypool. | coherence | 2009.05426 | 1 |
We demonstrate the technique by collecting commonsense knowledge that surrounds three fairly universal rituals---coming-of-age, marriage, and funerals---across three different national groups: the United States , India, and the Philippines. | <meaning-changed> We demonstrate the technique by collecting commonsense knowledge that surrounds three fairly universal rituals---coming-of-age, marriage, and funerals---across three different national groups: the United States , India, and the Philippines. | We demonstrate the technique by collecting commonsense knowledge that surrounds six fairly universal rituals---birth, coming-of-age, marriage, funerals, new year, and birthdays---across two national groups: the United States , India, and the Philippines. | meaning-changed | 2009.05664 | 1 |
We demonstrate the technique by collecting commonsense knowledge that surrounds three fairly universal rituals---coming-of-age, marriage, and funerals---across three different national groups: the United States , India, and the Philippines. Our pilot study expands the different types of relationships identified by existing work in the field of commonsense reasoning for commonplace events, and uses these new types to gather information that distinguishes the knowledge of the different groups. | <meaning-changed> We demonstrate the technique by collecting commonsense knowledge that surrounds three fairly universal rituals---coming-of-age, marriage, and funerals---across three different national groups: the United States , India, and the Philippines. Our pilot study expands the different types of relationships identified by existing work in the field of commonsense reasoning for commonplace events, and uses these new types to gather information that distinguishes the knowledge of the different groups. | We demonstrate the technique by collecting commonsense knowledge that surrounds three fairly universal rituals---coming-of-age, marriage, and funerals---across three different national groups: the United States and India. Our study expands the different types of relationships identified by existing work in the field of commonsense reasoning for commonplace events, and uses these new types to gather information that distinguishes the knowledge of the different groups. | meaning-changed | 2009.05664 | 1 |
Conventional sparse retrieval methods such as TF-IDF and BM25 are simple and efficient, but solely rely on lexical overlap and fail to conduct semantic matching. | <clarity> Conventional sparse retrieval methods such as TF-IDF and BM25 are simple and efficient, but solely rely on lexical overlap and fail to conduct semantic matching. | Conventional sparse retrieval methods such as TF-IDF and BM25 are simple and efficient, but solely rely on lexical overlap without semantic matching. | clarity | 2009.08553 | 1 |
Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and sometimes insufficient for exact matching as they embed the entire text sequence into a single vector with limited capacity. | <coherence> Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and sometimes insufficient for exact matching as they embed the entire text sequence into a single vector with limited capacity. | Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and insufficient for exact matching as they embed the entire text sequence into a single vector with limited capacity. | coherence | 2009.08553 | 1 |
Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and sometimes insufficient for exact matching as they embed the entire text sequence into a single vector with limited capacity. | <clarity> Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and sometimes insufficient for exact matching as they embed the entire text sequence into a single vector with limited capacity. | Recent dense retrieval methods learn latent representations to tackle the lexical mismatch problem, while being more computationally expensive and sometimes insufficient for exact matching as they embed the text sequence into a single vector with limited capacity. | clarity | 2009.08553 | 1 |
We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | <clarity> We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | We demonstrate on open-domain question answering that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | clarity | 2009.08553 | 1 |
We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | <clarity> We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the state-of-the-art dense method DPR cite{karpukhin2020dense}. | clarity | 2009.08553 | 1 |
We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | <clarity> We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense method DPR cite{karpukhin2020dense}. | We demonstrate on open-domain question answering (QA) that the generated contexts significantly enrich the semantics of the queries and thus GAR with sparse representations (BM25) achieves comparable or better performance than the current state-of-the-art dense methods such as DPR cite{karpukhin2020dense}. | clarity | 2009.08553 | 1 |
We show that generating various contexts of a query is beneficial as fusing their results consistently yields a better retrieval accuracy. | <clarity> We show that generating various contexts of a query is beneficial as fusing their results consistently yields a better retrieval accuracy. | We show that generating various contexts of a query is beneficial as fusing their results consistently yields better retrieval accuracy. | clarity | 2009.08553 | 1 |
Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | <meaning-changed> Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | Moreover, as sparse and dense representations are often complementary, GAR can be easily combined with DPR to achieve even better performance. Furthermore, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | meaning-changed | 2009.08553 | 1 |
Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | <clarity> Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | Moreover, GAR achieves the state-of-the-art performance on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | clarity | 2009.08553 | 1 |
Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | <meaning-changed> Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets under the extractive setting when equipped with an extractive reader . | meaning-changed | 2009.08553 | 1 |
Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | <meaning-changed> Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader . | Moreover, GAR achieves the state-of-the-art performance of extractive QA on the Natural Questions and TriviaQA datasets when equipped with an extractive reader , and consistently outperforms other retrieval methods when the same generative reader is used . | meaning-changed | 2009.08553 | 1 |
Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | <meaning-changed> Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | Existing methods are limited by the amount of parallel corpus. Can we build a system to fully utilize signals in a parallel ST corpus? We are inspired by human understanding system which is composed of auditory perception and cognitive processing. In this paper , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | meaning-changed | 2009.09704 | 2 |
Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | <clarity> Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | clarity | 2009.09704 | 2 |
Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | <clarity> Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, (LUT), a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | clarity | 2009.09704 | 2 |
Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | <clarity> Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision to decouple the end-to-end speech-to-text translation task. | Inspired by neuroscience, humans have perception systems and cognitive systems to process different information , we propose LUT, Listen-Understand-Translate, a unified framework with triple supervision signals to decouple the end-to-end speech-to-text translation task. | clarity | 2009.09704 | 2 |
In addition to the target language sentence translation loss, LUT includes two auxiliary supervising signals to guide the acoustic encoder to extracts acoustic features from the input, and the semantic encoder to extract semantic features relevant to the source transcription text. | <coherence> In addition to the target language sentence translation loss, LUT includes two auxiliary supervising signals to guide the acoustic encoder to extracts acoustic features from the input, and the semantic encoder to extract semantic features relevant to the source transcription text. | LUT is able to guide the acoustic encoder to extracts acoustic features from the input, and the semantic encoder to extract semantic features relevant to the source transcription text. | coherence | 2009.09704 | 2 |
In addition to the target language sentence translation loss, LUT includes two auxiliary supervising signals to guide the acoustic encoder to extracts acoustic features from the input, and the semantic encoder to extract semantic features relevant to the source transcription text. We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | <meaning-changed> In addition to the target language sentence translation loss, LUT includes two auxiliary supervising signals to guide the acoustic encoder to extracts acoustic features from the input, and the semantic encoder to extract semantic features relevant to the source transcription text. We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | In addition to the target language sentence translation loss, LUT includes two auxiliary supervising signals to guide the acoustic encoder to extract as much information from the auditory input. In addition, LUT utilizes a pre-trained BERT model to enforce the upper encoder to produce as much semantic information as possible, without extra data. We perform experiments on a diverse set of speech translation benchmarks, including Librispeech English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | meaning-changed | 2009.09704 | 2 |
We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | <meaning-changed> We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | We do experiments on English-French, IWSLT English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | meaning-changed | 2009.09704 | 2 |
We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | <meaning-changed> We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | We do experiments on English-French, English-German and TED English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | meaning-changed | 2009.09704 | 2 |
We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | <meaning-changed> We do experiments on English-French, English-German and English-Chinese speech translation benchmarks and the results demonstrate the reasonability of LUT . Our code is available at URL | We do experiments on English-French, English-German and English-Chinese . Our results demonstrate LUT achieves the state-of-the-art performance, outperforming previous methods. The code is available at URL | meaning-changed | 2009.09704 | 2 |
To reduce the learning difficulty, we propose COnSecutive Transcription and Translation (COSTT), an integral framework for speech-to-text translation. | <clarity> To reduce the learning difficulty, we propose COnSecutive Transcription and Translation (COSTT), an integral framework for speech-to-text translation. | To reduce the learning difficulty, we propose COnSecutive Transcription and Translation (COSTT), an integral approach for speech-to-text translation. | clarity | 2009.09737 | 2 |
Our method is verified on three mainstream datasets, including Augmented LibriSpeech English-French dataset, TED English-German dataset, and TED English-Chinese dataset. | <meaning-changed> Our method is verified on three mainstream datasets, including Augmented LibriSpeech English-French dataset, TED English-German dataset, and TED English-Chinese dataset. | The key idea is to generate source transcript and target translation text with a single decoder. It benefits the model training so that additional large parallel text corpus can be fully exploited to enhance the speech translation training. Our method is verified on three mainstream datasets, including Augmented LibriSpeech English-French dataset, TED English-German dataset, and TED English-Chinese dataset. | meaning-changed | 2009.09737 | 2 |
Our code is available at URL | <style> Our code is available at URL | The code is available at URL | style | 2009.09737 | 2 |
We introduce texttt N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. texttt N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | <meaning-changed> We introduce texttt N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. texttt N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | We introduce N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. texttt N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | meaning-changed | 2009.11616 | 1 |
We introduce texttt N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. texttt N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | <clarity> We introduce texttt N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. texttt N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | We introduce texttt N-LTP , an open-source Python Chinese natural language processing toolkit supporting five basic tasks: Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency parsing, and semantic dependency parsing. N-LTP adopts the multi-task framework with the pre-trained model to capture the shared knowledge across all Chinese relevant tasks. | clarity | 2009.11616 | 1 |
Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English and +2.0 BLEU points on IWSLT'17 French-English tasks. | <coherence> Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English and +2.0 BLEU points on IWSLT'17 French-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English , +2.0 BLEU points on IWSLT'17 French-English tasks. | coherence | 2009.13267 | 1 |
Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English and +2.0 BLEU points on IWSLT'17 French-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English and +2.0 BLEU points on IWSLT'17 French-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: +3 BLEU points on Sinhala-English and +2.0 BLEU points on IWSLT'17 French-English , and +1.7 BLEU points on WMT'19 German-English tasks. | meaning-changed | 2009.13267 | 1 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | We use both marginal energy models (over target sentence) and joint energy models (over both source and target sentences). Our EBR with the joint energy model consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | meaning-changed | 2009.13267 | 2 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: + 4 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | meaning-changed | 2009.13267 | 2 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on IWSLT'14 German-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | meaning-changed | 2009.13267 | 2 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 3.0 BELU points on Sinhala-English, + 1.7 BLEU points on WMT' 19 German-English tasks. | meaning-changed | 2009.13267 | 2 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.2 BLEU on WMT' 19 German-English tasks. | meaning-changed | 2009.13267 | 2 |
Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | <meaning-changed> Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 19 German-English tasks. | Our EBR consistently improves the performance of the Transformer-based NMT: + 3 BLEU points on Sinhala-English , + 2.0 BLEU points on IWSLT'17 French-English, and + 1.7 BLEU points on WMT' 16 English-German tasks. | meaning-changed | 2009.13267 | 2 |
For natural language processing (NLP) taskssuch as sentiment or topic classification, currently prevailing approaches heavily rely on pretraining large self-supervised models on massive external data resources. | <clarity> For natural language processing (NLP) taskssuch as sentiment or topic classification, currently prevailing approaches heavily rely on pretraining large self-supervised models on massive external data resources. | For natural language processing 'text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external data resources. | clarity | 2010.01061 | 1 |
For natural language processing (NLP) taskssuch as sentiment or topic classification, currently prevailing approaches heavily rely on pretraining large self-supervised models on massive external data resources. However, this methodology is being critiqued for: exceptional compute and pretraining data requirements ; | <clarity> For natural language processing (NLP) taskssuch as sentiment or topic classification, currently prevailing approaches heavily rely on pretraining large self-supervised models on massive external data resources. However, this methodology is being critiqued for: exceptional compute and pretraining data requirements ; | For natural language processing (NLP) taskssuch as sentiment or topic classification, currently prevailing approaches heavily rely on pretraining large self-supervised models on massive external data sources, which incurs exceptional pretraining data requirements ; | clarity | 2010.01061 | 1 |
However, this methodology is being critiqued for: exceptional compute and pretraining data requirements ; diminishing returns on both large and small datasets; and importantly, favourable evaluation settings that overestimate performance differences. The core belief behind current methodology, coined `the bitter lesson' by R. Sutton, is that `compute scale-up beats data and compute-efficient algorithms', neglecting that progress in compute hardware scale-up is based almost entirely on the miniaturisation of resource consumption. We thus approach pretrainingfrom a miniaturisation perspective, such as not to require massive external data sources and models, or learned translations from continuous input embeddings to discrete labels. To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. | <meaning-changed> However, this methodology is being critiqued for: exceptional compute and pretraining data requirements ; diminishing returns on both large and small datasets; and importantly, favourable evaluation settings that overestimate performance differences. The core belief behind current methodology, coined `the bitter lesson' by R. Sutton, is that `compute scale-up beats data and compute-efficient algorithms', neglecting that progress in compute hardware scale-up is based almost entirely on the miniaturisation of resource consumption. We thus approach pretrainingfrom a miniaturisation perspective, such as not to require massive external data sources and models, or learned translations from continuous input embeddings to discrete labels. To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. | However, this methodology is being critiqued for: exceptional compute and pretraining data requirements and a diminished ability to pretrain over small datasets. However, fundamental pretraining method capabilities like few to zero-shot learning or preserving minority concept (long-tail) prediction performance along with accordingly designed evaluation scenarios remain open challenges. We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining, which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail, few-shot and self-supervised zero-shot learning abilities. Accordingly, we analyse improvements in learning dynamics over baselines on a challenging long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. | meaning-changed | 2010.01061 | 1 |
To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. | <clarity> To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. | To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification scenario with noisy, highly sparse labels and many rare concepts. | clarity | 2010.01061 | 1 |
To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. To this end, we propose a novel `dataset-internal' contrastive autoencoding approach to self-supervised pretraining and demonstrate marked improvements in zero-shot, few-shot and solely supervised learning performance; even under an unfavorable low-resource scenario, and without defaulting to large-scale external datasets for self-supervision. We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | <coherence> To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many rare concepts. To this end, we propose a novel `dataset-internal' contrastive autoencoding approach to self-supervised pretraining and demonstrate marked improvements in zero-shot, few-shot and solely supervised learning performance; even under an unfavorable low-resource scenario, and without defaulting to large-scale external datasets for self-supervision. We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | To minimise overly favourable evaluation, we examine learning on a long-tailed, low-resource, multi-label text classification dataset with noisy, highly sparse labels and many minority concepts. We find that long-tailed zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | coherence | 2010.01061 | 1 |
We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | <clarity> We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | We also find empirical evidence that zero and few-shot learning markedly benefit from increasing ' dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | clarity | 2010.01061 | 1 |
We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | <fluency> We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | fluency | 2010.01061 | 1 |
We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | <clarity> We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised pretraining signals, to help reduce the reliance on large external sources of such signals is infeasible . | clarity | 2010.01061 | 1 |
We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | <clarity> We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources of such signals is infeasible . | We also find empirical evidence that zero and few-shot learning markedly benefit from adding more ` dataset-internal' , self-supervised training signals, which is of practical importance when retrieving or computing on large external sources . | clarity | 2010.01061 | 1 |
For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. | <fluency> For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. | For natural language processing ` text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. | fluency | 2010.01061 | 2 |
For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. | <meaning-changed> For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. | For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on increasingly larger `task-external' data. Transfer learning from high-resource pretraining works well, but research has focused on settings with very large data and compute requirements, while the potential of efficient low-resource learning, without large `task-external' pretraining, remains under-explored. In this work, we evaluate against three core challenges for resource efficient learning. Namely, we analyze: (1) pretraining data requirements and a diminished ability to pretrain over small datasets. | meaning-changed | 2010.01061 | 2 |
For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. However, fundamental pretraining method capabilities like few to zero-shot learningor preserving minority concept ( long-tail ) prediction performance along with accordingly designed evaluation scenarios remain open challenges . | <meaning-changed> For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data requirements and a diminished ability to pretrain over small datasets. However, fundamental pretraining method capabilities like few to zero-shot learningor preserving minority concept ( long-tail ) prediction performance along with accordingly designed evaluation scenarios remain open challenges . | For natural language processing ' text-to-text' tasks, the prevailing approaches heavily rely on pretraining large self-supervised models on massive external datasources, which incurs exceptional pretraining data (X) efficiency; (2) zero to few-shot label (Y) efficiency; and (3) long-tail ) prediction performance along with accordingly designed evaluation scenarios remain open challenges . | meaning-changed | 2010.01061 | 2 |
However, fundamental pretraining method capabilities like few to zero-shot learningor preserving minority concept ( long-tail ) prediction performance along with accordingly designed evaluation scenarios remain open challenges . We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining , which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail , few-shot and self-supervised zero-shot learning abilities. | <clarity> However, fundamental pretraining method capabilities like few to zero-shot learningor preserving minority concept ( long-tail ) prediction performance along with accordingly designed evaluation scenarios remain open challenges . We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining , which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail , few-shot and self-supervised zero-shot learning abilities. | However, fundamental pretraining method capabilities like few to zero-shot learningor preserving minority concept ( long-tail generalization, since long-tail , few-shot and self-supervised zero-shot learning abilities. | clarity | 2010.01061 | 2 |
We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining , which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail , few-shot and self-supervised zero-shot learning abilities. Accordingly, we analyse improvements in learning dynamics over baselines on a challenging long-tailed, low-resource, multi-label text classification scenario with noisy, highly sparse labels and many minority concepts . | <meaning-changed> We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining , which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail , few-shot and self-supervised zero-shot learning abilities. Accordingly, we analyse improvements in learning dynamics over baselines on a challenging long-tailed, low-resource, multi-label text classification scenario with noisy, highly sparse labels and many minority concepts . | We thus propose Contrastive Label-Embedding Self-Supervision (CLESS) pretraining , which enables pretraining from multiple magnitudes smaller, 'task internal' data only, while still strongly improving fully supervised, long-tail preservation has been linked to algorithmic fairness and because data in the tail is limited by definition. To address these challenges, we propose a data and compute efficient self-supervised, contrastive text encoder, pretrained on 60MB of `task-internal' text data, and compare it to RoBERTa, which was pretrained on 160GB of `task-external' text . | meaning-changed | 2010.01061 | 2 |
We find that long-tailed zero and few-shot learning markedly benefit from increasing 'dataset-internal' self-supervised pretraining signals, to help reduce the reliance on large external sources . | <meaning-changed> We find that long-tailed zero and few-shot learning markedly benefit from increasing 'dataset-internal' self-supervised pretraining signals, to help reduce the reliance on large external sources . | We find our method outperforms RoBERTa, while pretraining and fine-tuning in a 1/5th of RoBERTa's fine-tuning time . | meaning-changed | 2010.01061 | 2 |
We propose a simple method to generate large amounts of multilingual question and answer pairs by a single generative model. | <clarity> We propose a simple method to generate large amounts of multilingual question and answer pairs by a single generative model. | We propose a simple method to generate multilingual question and answer pairs by a single generative model. | clarity | 2010.12008 | 1 |
We propose a simple method to generate large amounts of multilingual question and answer pairs by a single generative model. | <meaning-changed> We propose a simple method to generate large amounts of multilingual question and answer pairs by a single generative model. | We propose a simple method to generate large amounts of multilingual question and answer pairs on a large scale through the use of a single generative model. | meaning-changed | 2010.12008 | 1 |
These synthetic samples are then applied to augment the available gold multilingual ones to improve the performance of multilingual QA models on target languages. | <clarity> These synthetic samples are then applied to augment the available gold multilingual ones to improve the performance of multilingual QA models on target languages. | These synthetic samples can be used to improve the zero-shot performance of multilingual QA models on target languages. | clarity | 2010.12008 | 1 |
Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | <clarity> Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | Our proposed multi-task training of the generative model only requires the training samples in English , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | clarity | 2010.12008 | 1 |
Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | <clarity> Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for labeled samples in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | clarity | 2010.12008 | 1 |
Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | <meaning-changed> Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages , making it applicable to far more languages than those with labeled data . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | meaning-changed | 2010.12008 | 1 |
Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | <meaning-changed> Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains in a number of multilingual datasets . | Our approach only requires existence of automatically translated samples from Englishto the target domain , thus removing the need for human annotations in the target languages . Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks, reducing the gap between zero-shot and supervised performance of QA models on various languages . | meaning-changed | 2010.12008 | 1 |
Our proposed multi-task training of the generative model only requires the training samples in English, thus removing the need for labeled samples in the target languages, making it applicable to far more languages than those with labeled data. | <meaning-changed> Our proposed multi-task training of the generative model only requires the training samples in English, thus removing the need for labeled samples in the target languages, making it applicable to far more languages than those with labeled data. | Our proposed multi-task training of the generative model only requires the labeled training samples in English, thus removing the need for labeled samples in the target languages, making it applicable to far more languages than those with labeled data. | meaning-changed | 2010.12008 | 2 |
Our proposed multi-task training of the generative model only requires the training samples in English, thus removing the need for labeled samples in the target languages, making it applicable to far more languages than those with labeled data. | <clarity> Our proposed multi-task training of the generative model only requires the training samples in English, thus removing the need for labeled samples in the target languages, making it applicable to far more languages than those with labeled data. | Our proposed multi-task training of the generative model only requires the training samples in English, thus removing the need for such samples in the target languages, making it applicable to far more languages than those with labeled data. | clarity | 2010.12008 | 2 |
Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | <meaning-changed> Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | Human evaluations indicate the majority of such samples are grammatically correct and sensible. Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | meaning-changed | 2010.12008 | 2 |
Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | <meaning-changed> Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | Experimental results show our proposed approach can achieve large gains on the XQuAD dataset , reducing the gap between zero-shot and supervised performance of QA models on various languages. | meaning-changed | 2010.12008 | 2 |
Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | <meaning-changed> Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of QA models on various languages. | Experimental results show our proposed approach achieves significant gains on several multilingual QA benchmarks , reducing the gap between zero-shot and supervised performance of smaller QA models on various languages. | meaning-changed | 2010.12008 | 2 |
Natural language is one of the ways information is encoded and it has highly abstracted and conceptualized the information. | <meaning-changed> Natural language is one of the ways information is encoded and it has highly abstracted and conceptualized the information. | First of all, please URLet all you knew about the lexical classification, then let's jump to the conclusion. This paper reclassified lexical chunks into data chunks, structure chunks, and pointer chunks. Almost all data chunks are information sets. According to the difference of the set structures, data chunks can be further divided into attribute chunks and entity chunks. According to the different abstraction level and method, attribute chunks can be further divided into basic attribute chunks, extended attribute chunks, and advanced attribute chunks. All of the above classification principles are structural and functionalbased discrimination, instead of artificially divide lexical chunks into a noun, adjective, pronouns, and so on. Now, let's back to the normal study process. The author believes natural language is one of the ways information is encoded and it has highly abstracted and conceptualized the information. | meaning-changed | 2010.12789 | 1 |
This paper disassembles the information represented by natural language , | <clarity> This paper disassembles the information represented by natural language , | Therefore the study begins with disassembling the information represented by natural language , | clarity | 2010.12789 | 1 |
This paper disassembles the information represented by natural language , analyzes the classification coding system of attribute information and the abstraction relation between attribute information and entities in the real world , | <style> This paper disassembles the information represented by natural language , analyzes the classification coding system of attribute information and the abstraction relation between attribute information and entities in the real world , | This paper disassembles the information represented by natural language and then discovered the classification coding system of attribute information and the abstraction relation between attribute information and entities in the real world , | style | 2010.12789 | 1 |