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
|
|---|---|---|---|---|---|
To address this issue, we present Shapeshifter Networks (SSNs), a flexible neural network framework that decouples layers from model weights, enabling us to implement any neural network with an arbitrary number of parameters . In SSNseach layer obtains weights from a parameter store that decides where and how to allocate parameters to layers . | <meaning-changed> To address this issue, we present Shapeshifter Networks (SSNs), a flexible neural network framework that decouples layers from model weights, enabling us to implement any neural network with an arbitrary number of parameters . In SSNseach layer obtains weights from a parameter store that decides where and how to allocate parameters to layers . | Parameter sharing can reduce memory requirements, but existing methods only share parameters between identical layers, limiting their impact. This paper removes these restrictions with a novel task called Neural Parameter Allocation Search (NPAS), where the goal is to generate weights for a network using a given parameter budget. NPAS requires new techniques to morph available parameters to fit any architecture. To address this new task we introduce Shapeshifter Networks (SSNs), which automatically learns where and how to allocate parameters to layers . | meaning-changed | 2006.10598 | 2 |
In SSNseach layer obtains weights from a parameter store that decides where and how to allocate parameters to layers . This can result in sharing parameters across layers even when they have different sizes or perform different operations. | <clarity> In SSNseach layer obtains weights from a parameter store that decides where and how to allocate parameters to layers . This can result in sharing parameters across layers even when they have different sizes or perform different operations. | In SSNseach layer obtains weights from a parameter store that decides where and how to share parameters between all layers in a network, even between layers of varying sizes and operations. | clarity | 2006.10598 | 2 |
SSNs do not require any modifications to a model's loss function or architecture , making them easy to use. | <clarity> SSNs do not require any modifications to a model's loss function or architecture , making them easy to use. | SSNs do not require any loss function or architecture , making them easy to use. | clarity | 2006.10598 | 2 |
SSNs do not require any modifications to a model's loss function or architecture , making them easy to use. | <meaning-changed> SSNs do not require any modifications to a model's loss function or architecture , making them easy to use. | SSNs do not require any modifications to a model's loss function or architecture modifications , making them easy to use. | meaning-changed | 2006.10598 | 2 |
Our approach can create parameter efficient networks by using a relatively small number of weights, or can improve a model's performance by adding additional model capacity during training without affecting the computational resources required at test time. | <coherence> Our approach can create parameter efficient networks by using a relatively small number of weights, or can improve a model's performance by adding additional model capacity during training without affecting the computational resources required at test time. | coherence | 2006.10598 | 2 |
|
We evaluate SSNs using seven network architectures across diverse tasks that include image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | <meaning-changed> We evaluate SSNs using seven network architectures across diverse tasks that include image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | We evaluate SSNs in key NPAS settings using seven network architectures across diverse tasks that include image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | meaning-changed | 2006.10598 | 2 |
We evaluate SSNs using seven network architectures across diverse tasks that include image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | <clarity> We evaluate SSNs using seven network architectures across diverse tasks that include image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | We evaluate SSNs using seven network architectures across diverse tasks including image classification, bidirectional image-sentence retrieval, and phrase grounding, creating high performing models even when using as little as 1\% of the parameters. | clarity | 2006.10598 | 2 |
We set a new state of the art on both the 100 hour subset of Librispeech as well as on TIMIT phoneme recognition . | <meaning-changed> We set a new state of the art on both the 100 hour subset of Librispeech as well as on TIMIT phoneme recognition . | Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/noisy test sets . | meaning-changed | 2006.11477 | 1 |
When lowering the amount of labeled data to one hour, our model outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. | <clarity> When lowering the amount of labeled data to one hour, our model outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. | When lowering the amount of labeled data to one hour, wav2vec 2.0 outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. | clarity | 2006.11477 | 1 |
Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.7 / 10.1 WER on the noisy/clean test sets of Librispeech. | <meaning-changed> Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.7 / 10.1 WER on the noisy/clean test sets of Librispeech. | Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 10.1 WER on the noisy/clean test sets of Librispeech. | meaning-changed | 2006.11477 | 1 |
Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.7 / 10.1 WER on the noisy/clean test sets of Librispeech. | <meaning-changed> Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.7 / 10.1 WER on the noisy/clean test sets of Librispeech. | Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.7 / 8.6 WER on the noisy/clean test sets of Librispeech. | meaning-changed | 2006.11477 | 1 |
This demonstrates the feasibility of speech recognition with limited amounts of labeled data . Fine-tuning on all of Librispeech achieves 1.9/3.5 WER using a simple baseline model architecture. We will release code and models . | <clarity> This demonstrates the feasibility of speech recognition with limited amounts of labeled data . Fine-tuning on all of Librispeech achieves 1.9/3.5 WER using a simple baseline model architecture. We will release code and models . | This demonstrates the feasibility of speech recognition with limited amounts of labeled data . | clarity | 2006.11477 | 1 |
Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/ noisy test sets. | <clarity> Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/ noisy test sets. | Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/ other test sets. | clarity | 2006.11477 | 2 |
Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 8.6 WERon the noisy/clean test sets of Librispeech . | <meaning-changed> Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 8.6 WERon the noisy/clean test sets of Librispeech . | Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8 / 8.6 WERon the noisy/clean test sets of Librispeech . | meaning-changed | 2006.11477 | 2 |
Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 8.6 WERon the noisy/clean test sets of Librispeech . | <meaning-changed> Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 8.6 WERon the noisy/clean test sets of Librispeech . | Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 5.2 / 8.2 WER . | meaning-changed | 2006.11477 | 2 |
How to explicitly encode positional information into neural networks is an important problem in natural language processing. In the Transformer model , the positional information is simply encoded as embedding vectors, which are used in the input layer, or encoded as a bias term in the self-attention module. | <clarity> How to explicitly encode positional information into neural networks is an important problem in natural language processing. In the Transformer model , the positional information is simply encoded as embedding vectors, which are used in the input layer, or encoded as a bias term in the self-attention module. | How to explicitly encode positional information into neural networks is important in learning the representation of natural languages, such as BERT. Based on the Transformer architecture , the positional information is simply encoded as embedding vectors, which are used in the input layer, or encoded as a bias term in the self-attention module. | clarity | 2006.15595 | 1 |
In the self-attention module, the word correlation and positional correlation are computed separately with different parameterizations and then added together. | <meaning-changed> In the self-attention module, the word correlation and positional correlation are computed separately with different parameterizations and then added together. | In the self-attention module, the word contextual correlation and positional correlation are computed separately with different parameterizations and then added together. | meaning-changed | 2006.15595 | 1 |
This design removes the noisy word-position correlation and gives more expressiveness to characterize the relationship between words/positions by using different projection matrices. | <clarity> This design removes the noisy word-position correlation and gives more expressiveness to characterize the relationship between words/positions by using different projection matrices. | This design removes the addition over heterogeneous embeddings in the input, which may potentially bring randomness, and gives more expressiveness to characterize the relationship between words/positions by using different projection matrices. | clarity | 2006.15595 | 1 |
To combat COVID-19, clinicians and scientists all need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | <coherence> To combat COVID-19, clinicians and scientists all need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | To combat COVID-19, both clinicians and scientists all need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | coherence | 2007.00576 | 1 |
To combat COVID-19, clinicians and scientists all need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | <clarity> To combat COVID-19, clinicians and scientists all need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | To combat COVID-19, clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | clarity | 2007.00576 | 1 |
We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | <coherence> We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | We have developed a novel and comprehensive knowledge discovery framework, COVID-KG extract fine-grained multimedia knowledge elements (entities, relations and events) . | coherence | 2007.00576 | 1 |
We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | <meaning-changed> We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to to extract fine-grained multimedia knowledge elements (entities, relations and events) . | meaning-changed | 2007.00576 | 1 |
We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | <clarity> We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) . | We have developed a novel and comprehensive knowledge discovery framework, COVID-KG, which leverages novel semantic representation and external ontologies to represent text and images in the input literature data, and then performs various extraction components to extract fine-grained multimedia knowledge elements (entities, relations and events) from scientific literature . | clarity | 2007.00576 | 1 |
We then exploit the constructed multimedia KGs for question answering and report generation, using drug repurposing as a case study. | <clarity> We then exploit the constructed multimedia KGs for question answering and report generation, using drug repurposing as a case study. | We then exploit the constructed multimedia knowledge graphs (KGs) for question answering and report generation, using drug repurposing as a case study. | clarity | 2007.00576 | 1 |
All of the data, KGs, resources, and shared services are publicly available. | <clarity> All of the data, KGs, resources, and shared services are publicly available. | All of the data, KGs, reports, resources and shared services are publicly available. | clarity | 2007.00576 | 1 |
To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | <fluency> To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | To combat COVID-19, both clinicians and scientists need to digest vast amounts of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | fluency | 2007.00576 | 2 |
To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | <clarity> To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in scientific literature to understand the disease mechanism and the related biological functions. | clarity | 2007.00576 | 2 |
To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | <coherence> To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and the related biological functions. | To combat COVID-19, both clinicians and scientists need to digest the vast amount of relevant biomedical knowledge in literature to understand the disease mechanism and related biological functions. | coherence | 2007.00576 | 2 |
We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | <meaning-changed> We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | We have developed a novel and comprehensive knowledge discovery framework, COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | meaning-changed | 2007.00576 | 2 |
We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | <meaning-changed> We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities and their visual chemical structures, relations , relations and events) from scientific literature. | meaning-changed | 2007.00576 | 2 |
We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | <fluency> We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , relations and events) from scientific literature. | We have developed a novel and comprehensive knowledge discovery framework, textbf COVID-KG to extract fine-grained multimedia knowledge elements (entities , and events) from scientific literature. | fluency | 2007.00576 | 2 |
Our framework also provides detailed contextual sentences, subfigures and knowledge subgraphs as evidence . | <fluency> Our framework also provides detailed contextual sentences, subfigures and knowledge subgraphs as evidence . | Our framework also provides detailed contextual sentences, subfigures , and knowledge subgraphs as evidence . | fluency | 2007.00576 | 2 |
Our framework also provides detailed contextual sentences, subfigures and knowledge subgraphs as evidence . All of the data, KGs, reports, resources and shared services are publicly available . | <coherence> Our framework also provides detailed contextual sentences, subfigures and knowledge subgraphs as evidence . All of the data, KGs, reports, resources and shared services are publicly available . | Our framework also provides detailed contextual sentences, subfigures and knowledge subgraphs as evidence . | coherence | 2007.00576 | 2 |
Using the presence or frequency of keywords is a classic approach in the formal analysis of text, but has the drawback of glossing over the relationality of word meanings. | <coherence> Using the presence or frequency of keywords is a classic approach in the formal analysis of text, but has the drawback of glossing over the relationality of word meanings. | Using the frequency of keywords is a classic approach in the formal analysis of text, but has the drawback of glossing over the relationality of word meanings. | coherence | 2007.04508 | 1 |
Word embedding models overcome this problem by constructing a standardized meaning space where words are assigned a location based on relations of similarity to , and difference from, other words based on how they are used in natural language samples. | <clarity> Word embedding models overcome this problem by constructing a standardized meaning space where words are assigned a location based on relations of similarity to , and difference from, other words based on how they are used in natural language samples. | Word embedding models overcome this problem by constructing a standardized and continuous "meaning-space" where words are assigned a location based on relations of similarity to , and difference from, other words based on how they are used in natural language samples. | clarity | 2007.04508 | 1 |
Word embedding models overcome this problem by constructing a standardized meaning space where words are assigned a location based on relations of similarity to , and difference from, other words based on how they are used in natural language samples. | <coherence> Word embedding models overcome this problem by constructing a standardized meaning space where words are assigned a location based on relations of similarity to , and difference from, other words based on how they are used in natural language samples. | Word embedding models overcome this problem by constructing a standardized meaning space where words are assigned a location based on relations of similarity to other words based on how they are used in natural language samples. | coherence | 2007.04508 | 1 |
We show how word embeddings can be put to the task of interpretation via two kinds of navigation. | <meaning-changed> We show how word embeddings can be put to the task of interpretation via two kinds of navigation. | We show how word embeddings are commensurate with prevailing theories of meaning in sociology and can be put to the task of interpretation via two kinds of navigation. | meaning-changed | 2007.04508 | 1 |
First, one can hold terms constant and measure how the embedding space moves around them--much like astronomers measured the changing of celestial bodies with the seasons. | <others> First, one can hold terms constant and measure how the embedding space moves around them--much like astronomers measured the changing of celestial bodies with the seasons. | First, one can hold terms constant and measure how the embedding space moves around them -- much like astronomers measured the changing of celestial bodies with the seasons. | others | 2007.04508 | 1 |
Second, one can also hold the embedding space constant and see how documents or authors move relative to it--just as ships use the stars on a given night to determine their location. | <others> Second, one can also hold the embedding space constant and see how documents or authors move relative to it--just as ships use the stars on a given night to determine their location. | Second, one can also hold the embedding space constant and see how documents or authors move relative to it -- just as ships use the stars on a given night to determine their location. | others | 2007.04508 | 1 |
Using the empirical case of immigration discourse in the United States, we demonstrate the merits of these two broad strategies to advance formal approaches to cultural analysis . | <meaning-changed> Using the empirical case of immigration discourse in the United States, we demonstrate the merits of these two broad strategies to advance formal approaches to cultural analysis . | Using the empirical case of immigration discourse in the United States, we demonstrate the merits of these two broad strategies for advancing important topics in cultural theory, including social marking, media fields, echo chambers, and cultural diffusion and change more broadly . | meaning-changed | 2007.04508 | 1 |
Motivation: NLP continues improving substantially through auto-regressive and auto-encoding Language Models . These LMsrequire expensive computing resources for self-supervised or un-supervised learning from huge unlabelled text corpora. The information learned is transferred through so-called embeddings to downstream prediction tasks. Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | <clarity> Motivation: NLP continues improving substantially through auto-regressive and auto-encoding Language Models . These LMsrequire expensive computing resources for self-supervised or un-supervised learning from huge unlabelled text corpora. The information learned is transferred through so-called embeddings to downstream prediction tasks. Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | Computational biology and bioinformatics provide vast data gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | clarity | 2007.06225 | 1 |
Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | <clarity> Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | Bioinformatics provide vast gold-mines from protein sequences, ideal for Language Models (LMs) taken from Natural Language Processing (NLP). These LMs reach for new prediction frontiers at low inference cost . | clarity | 2007.06225 | 1 |
Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | <fluency> Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference cost . | Bioinformatics provide vast gold-mines of structured and sequentially ordered text data leading to extraordinarily successful protein sequence LMs that promise new frontiers for generative and predictive tasks at low inference costs . | fluency | 2007.06225 | 1 |
Here, we addressed two questions: (1) To which extent can HPC up-scale protein LMs to larger databases and larger models? (2) To which extent can LMs extract features from single proteins to get closer to the performance of methods using evolutionary information? Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | <clarity> Here, we addressed two questions: (1) To which extent can HPC up-scale protein LMs to larger databases and larger models? (2) To which extent can LMs extract features from single proteins to get closer to the performance of methods using evolutionary information? Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | clarity | 2007.06225 | 1 |
Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | <coherence> Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | Methodology: Here, we trained two auto-regressive language models (Transformer-XL , XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | coherence | 2007.06225 | 1 |
Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | <clarity> Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( Bert, Albert) on data from UniRef and BFD containing up to 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | clarity | 2007.06225 | 1 |
Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | <clarity> Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids (words) from 2.1 billion protein sequences ( BFD ). | clarity | 2007.06225 | 1 |
Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | <meaning-changed> Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( BFD ). | Methodology: Here, we trained two auto-regressive language models (Transformer-XL and XLNet) and two auto-encoder models ( BERT and Albert) using 80 billion amino acids from 200 million protein sequences (UniRef100) and 393 billion amino acids from 2.1 billion protein sequences ( 22- and 112-times the entire English Wikipedia ). | meaning-changed | 2007.06225 | 1 |
The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | <meaning-changed> The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | The LMs were trained on the Summit supercomputer at Oak Ridge National Laboratory (ORNL), using 936 nodes (total 5616 GPUs and one TPU Pod , using V3-512 cores. | meaning-changed | 2007.06225 | 1 |
The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | <fluency> The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | The LMs were trained on the Summit supercomputer , using 5616 GPUs ) and one TPU Pod , using V3-512 cores. | fluency | 2007.06225 | 1 |
The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | <fluency> The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. | The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod ( V3-512 cores. | fluency | 2007.06225 | 1 |
The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | <clarity> The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 cores. Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | The LMs were trained on the Summit supercomputer , using 5616 GPUs and one TPU Pod , using V3-512 or V3-1024). We validated the advantage of up-scaling LMs to larger models supported by bigger data by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | clarity | 2007.06225 | 1 |
Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | <clarity> Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | Results: The results of training these LMs on proteins was assessed by predicting secondary structure (3-states: Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | clarity | 2007.06225 | 1 |
Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | <meaning-changed> Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 76-84, 8-states: Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | meaning-changed | 2007.06225 | 1 |
Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | <meaning-changed> Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 63-72), localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | Results: The results of training these LMs on proteins was assessed by predicting secondary structure in three- and eight-states ( Q3= 75-83, Q8= 65-73), sub-cellular localization for 10 cellular compartments (Q10=74) and whether a protein is membrane-bound or water-soluble (Q2=89). | meaning-changed | 2007.06225 | 1 |
Dimensionality reduction revealed that the LM-embeddings from unlabelled data (only protein sequences) captured important biophysical properties of the protein alphabet, namely the amino acids, and their well orchestrated interplay in governing the shapeof proteins. | <fluency> Dimensionality reduction revealed that the LM-embeddings from unlabelled data (only protein sequences) captured important biophysical properties of the protein alphabet, namely the amino acids, and their well orchestrated interplay in governing the shapeof proteins. | Dimensionality reduction revealed that the LM-embeddings from unlabeled data (only protein sequences) captured important biophysical properties of the protein alphabet, namely the amino acids, and their well orchestrated interplay in governing the shapeof proteins. | fluency | 2007.06225 | 1 |
Dimensionality reduction revealed that the LM-embeddings from unlabelled data (only protein sequences) captured important biophysical properties of the protein alphabet, namely the amino acids, and their well orchestrated interplay in governing the shapeof proteins. In the analogy of NLP, this implied having learned some of the grammar of the language of life realized in protein sequences. | <clarity> Dimensionality reduction revealed that the LM-embeddings from unlabelled data (only protein sequences) captured important biophysical properties of the protein alphabet, namely the amino acids, and their well orchestrated interplay in governing the shapeof proteins. In the analogy of NLP, this implied having learned some of the grammar of the language of life realized in protein sequences. | Dimensionality reduction revealed that the LM-embeddings from unlabelled data (only protein sequences) captured important biophysical properties governing protein shape. This implied learning some of the grammar of the language of life realized in protein sequences. | clarity | 2007.06225 | 1 |
In the analogy of NLP, this implied having learned some of the grammar of the language of life realized in protein sequences. | <meaning-changed> In the analogy of NLP, this implied having learned some of the grammar of the language of life realized in protein sequences. | In the analogy of NLP, this implied having learned some of the grammar of the language of life realized in protein sequences. The successful up-scaling of protein LMs through HPC to larger data sets slightly reduced the gap between models trained on evolutionary information and LMs. The official GitHub repository: URL | meaning-changed | 2007.06225 | 1 |
Current models are too strongly linked to the text-based patterns in large corpora, and too weakly linked to the desires, goals, and beliefs that people use words in order to express . | <clarity> Current models are too strongly linked to the text-based patterns in large corpora, and too weakly linked to the desires, goals, and beliefs that people use words in order to express . | Current models are too strongly linked to the text-based patterns in large corpora, and too weakly linked to the desires, goals, and beliefs that people express through words . | clarity | 2008.01766 | 1 |
Word meanings must also be grounded in vision and action, and capable of flexible combinations , in ways that current systems are not. | <fluency> Word meanings must also be grounded in vision and action, and capable of flexible combinations , in ways that current systems are not. | Word meanings must also be grounded in vision and action, and capable of flexible combinations in ways that current systems are not. | fluency | 2008.01766 | 1 |
Machines show an increasingly broad set of linguistic competencies, thanks to recent progress in Natural Language Processing (NLP). | <clarity> Machines show an increasingly broad set of linguistic competencies, thanks to recent progress in Natural Language Processing (NLP). | Machines have achieved a broad and growing set of linguistic competencies, thanks to recent progress in Natural Language Processing (NLP). | clarity | 2008.01766 | 2 |
Many algorithms stem from past computational work in psychology, raising the question of whether they understand words as people do . | <meaning-changed> Many algorithms stem from past computational work in psychology, raising the question of whether they understand words as people do . | Psychologists have shown increasing interest in such models, comparing their output to psychological judgments such as similarity, association, priming, and comprehension, raising the question of whether they understand words as people do . | meaning-changed | 2008.01766 | 2 |
Many algorithms stem from past computational work in psychology, raising the question of whether they understand words as people do . | <meaning-changed> Many algorithms stem from past computational work in psychology, raising the question of whether they understand words as people do . | Many algorithms stem from past computational work in psychology, raising the question of whether the models could serve as psychological theories . | meaning-changed | 2008.01766 | 2 |
In this paper , we compare how humans and machines represent the meaning of words. | <style> In this paper , we compare how humans and machines represent the meaning of words. | In this article , we compare how humans and machines represent the meaning of words. | style | 2008.01766 | 2 |
We argue that contemporary NLP systems are promising models of human word similarity, but they fall short in many other respects. | <clarity> We argue that contemporary NLP systems are promising models of human word similarity, but they fall short in many other respects. | We argue that contemporary NLP systems are fairly successful models of human word similarity, but they fall short in many other respects. | clarity | 2008.01766 | 2 |
Word meanings must also be grounded in vision and action , and capable of flexible combinations in ways that current systems are not. | <clarity> Word meanings must also be grounded in vision and action , and capable of flexible combinations in ways that current systems are not. | Word meanings must also be grounded in perception and action and be capable of flexible combinations in ways that current systems are not. | clarity | 2008.01766 | 2 |
We pose concrete challenges for developing machines with a more human-like, conceptual basis for word meaning . | <meaning-changed> We pose concrete challenges for developing machines with a more human-like, conceptual basis for word meaning . | We discuss more promising approaches to grounding NLP systems and argue that they will be more successful with a more human-like, conceptual basis for word meaning . | meaning-changed | 2008.01766 | 2 |
We pose concrete challenges for developing machines with a more human-like, conceptual basis for word meaning . We also discuss implications for cognitive science and NLP . | <coherence> We pose concrete challenges for developing machines with a more human-like, conceptual basis for word meaning . We also discuss implications for cognitive science and NLP . | We pose concrete challenges for developing machines with a more human-like, conceptual basis for word meaning . | coherence | 2008.01766 | 2 |
Non-autoregressive neural machine translation achieves remarkable inference acceleration compared to autoregressive models. However, current non-autoregressive models still fall behind their autoregressive counterparts in prediction accuracy. | <coherence> Non-autoregressive neural machine translation achieves remarkable inference acceleration compared to autoregressive models. However, current non-autoregressive models still fall behind their autoregressive counterparts in prediction accuracy. | Although non-autoregressive models still fall behind their autoregressive counterparts in prediction accuracy. | coherence | 2008.07905 | 1 |
However, current non-autoregressive models still fall behind their autoregressive counterparts in prediction accuracy. | <meaning-changed> However, current non-autoregressive models still fall behind their autoregressive counterparts in prediction accuracy. | However, current non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. | meaning-changed | 2008.07905 | 1 |
We attribute the accuracy gaps to two disadvantages of non-autoregressive models : a) learning simultaneous generation under the overly strong conditional independence assumption; | <meaning-changed> We attribute the accuracy gaps to two disadvantages of non-autoregressive models : a) learning simultaneous generation under the overly strong conditional independence assumption; | The non-autoregressive models with the best accuracy currently rely on multiple decoding iterations, which largely sacrifice the inference speed of non-autoregressive models : a) learning simultaneous generation under the overly strong conditional independence assumption; | meaning-changed | 2008.07905 | 1 |
We attribute the accuracy gaps to two disadvantages of non-autoregressive models : a) learning simultaneous generation under the overly strong conditional independence assumption; b) lacking explicit target language modeling. In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | <meaning-changed> We attribute the accuracy gaps to two disadvantages of non-autoregressive models : a) learning simultaneous generation under the overly strong conditional independence assumption; b) lacking explicit target language modeling. In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | We attribute the accuracy gaps to two disadvantages of non-autoregressive models . Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | meaning-changed | 2008.07905 | 1 |
In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | <meaning-changed> In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | In this paper , we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | meaning-changed | 2008.07905 | 1 |
In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | <clarity> In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on three benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | clarity | 2008.07905 | 1 |
In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | <clarity> In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without sacrificing any inference efficiency . | clarity | 2008.07905 | 1 |
In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | <clarity> In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without sacrificing any inference efficiency . | In this paper , we propose Glancing Transformer (GLAT) to address the above disadvantages, which reduces the difficulty of learning simultaneous generation and introduces explicit target language modeling in the non-autoregressive setting at the same time . Experiments on several benchmarks demonstrate that our approach significantly improves the accuracy of non-autoregressive models without multiple decoding iterations . | clarity | 2008.07905 | 1 |
In particular, GLAT achieves 30.91 BLEU on WMT 2014 German-English, which narrows the gap between autoregressive models and non-autoregressive models to less than 0.5 BLEU score . | <clarity> In particular, GLAT achieves 30.91 BLEU on WMT 2014 German-English, which narrows the gap between autoregressive models and non-autoregressive models to less than 0.5 BLEU score . | In particular, GLAT achieves state-of-the-art results among non-iterative models and even outperforms top iterative counterparts in some specific benchmarks . | clarity | 2008.07905 | 1 |
Although non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. | <clarity> Although non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. | Recent work on non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. | clarity | 2008.07905 | 2 |
Although non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. The non-autoregressive models with the best accuracy currently rely on multiple decoding iterations, which largely sacrifice the inference speed of non-autoregressive models. Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | <meaning-changed> Although non-autoregressive models with one-iteration generation achieve remarkable inference speed-up, they still fall behind their autoregressive counterparts in prediction accuracy. The non-autoregressive models with the best accuracy currently rely on multiple decoding iterations, which largely sacrifice the inference speed of non-autoregressive models. Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | Although non-autoregressive neural machine translation (NAT) aims at improving the efficiency by parallel decoding without sacrificing the quality. However, existing NAT methods are either inferior to Transformer or require multiple decoding passes, leading to reduced speedup. We propose the Glancing Language Model (GLM), a method to learn word interdependency for single-pass parallel generation models. With GLM, we develop Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | meaning-changed | 2008.07905 | 2 |
Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | <meaning-changed> Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) for machine translation. With only single-pass parallel decoding, GLAT is able to generate high-quality translation with 8-15 times speedup . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | meaning-changed | 2008.07905 | 2 |
Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | <clarity> Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. | Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on multiple WMT language directions show that GLAT outperforms all previous single pass non-autoregressive models without multiple decoding iterations. | clarity | 2008.07905 | 2 |
Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. In particular, GLAT achieves state-of-the-art results among non-iterative models and even outperforms top iterative counterparts in some specific benchmarks . | <meaning-changed> Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive models without multiple decoding iterations. In particular, GLAT achieves state-of-the-art results among non-iterative models and even outperforms top iterative counterparts in some specific benchmarks . | Inspired by the way of learning word dependencies in autoregressive and iterative-decoding models, we propose Glancing Transformer (GLAT) with a glancing language model (GLM), which learns to capture the word dependency gradually . Experiments on three benchmarks demonstrate that our approach can significantly improve the accuracy of non-autoregressive methods, and is nearly comparable to Transformer, reducing the gap to 0.25-0.9 BLEU points . | meaning-changed | 2008.07905 | 2 |
However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | <clarity> However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | However, to build a real-world intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | clarity | 2008.11015 | 1 |
However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | <clarity> However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | However, to build an intelligent assistant that recommends commonly composed charts, it should take the challenges of efficiency , imbalanced data and open vocabulary exist . | clarity | 2008.11015 | 1 |
However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | <clarity> However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data and open vocabulary exist . | However, to build an intelligent assistant that recommends commonly composed charts, the fundamental problems of "multi-dialect" unification , imbalanced data hungry and table context into consideration . | clarity | 2008.11015 | 1 |
On a large spreadsheet corpus with 196k tables and 306k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | <meaning-changed> On a large spreadsheet corpus with 196k tables and 306k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | meaning-changed | 2008.11015 | 1 |
Table2Charts has >0.61 recall at top-3 and >0.49 recall at top-1 for both single-type and multi-type chart recommendation tasks . | <meaning-changed> Table2Charts has >0.61 recall at top-3 and >0.49 recall at top-1 for both single-type and multi-type chart recommendation tasks . | Table2Charts outperforms other chart recommendation systems in both multi-type chart recommendation tasks . | meaning-changed | 2008.11015 | 1 |
Table2Charts has >0.61 recall at top-3 and >0.49 recall at top-1 for both single-type and multi-type chart recommendation tasks . | <meaning-changed> Table2Charts has >0.61 recall at top-3 and >0.49 recall at top-1 for both single-type and multi-type chart recommendation tasks . | Table2Charts has >0.61 recall at top-3 and >0.49 recall at top-1 for both single-type and multi-type task (with almost doubled recall numbers R@3=0.62 and R@1=0.44) and human evaluations . | meaning-changed | 2008.11015 | 1 |
However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | <clarity> However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | However, to recommend commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | clarity | 2008.11015 | 2 |
However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | <clarity> However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | However, to build a real-world intelligent assistant that recommends commonly composed charts in real world, one should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | clarity | 2008.11015 | 2 |
However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | <clarity> However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data hungry and table context into consideration. | However, to build a real-world intelligent assistant that recommends commonly composed charts , it should take the challenges of efficiency, imbalanced data and table context into consideration. | clarity | 2008.11015 | 2 |
On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | <meaning-changed> On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | On a large spreadsheet corpus with 165k tables and 266k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | meaning-changed | 2008.11015 | 2 |
On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | <meaning-changed> On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that tasks on different chart types could mutually enhance each other. | On a large spreadsheet corpus with 167k tables and 271k charts, we show that Table2Charts could learn a shared representation of table fields so that recommendation tasks on different chart types could mutually enhance each other. | meaning-changed | 2008.11015 | 2 |
Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | <clarity> Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | Table2Charts outperforms other chart recommendation systems in both multi-type task (with doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | clarity | 2008.11015 | 2 |
Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | <meaning-changed> Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.61 and R@1= 0.44 ) and human evaluations. | meaning-changed | 2008.11015 | 2 |
Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | <meaning-changed> Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.44 ) and human evaluations. | Table2Charts outperforms other chart recommendation systems in both multi-type task (with almost doubled recall numbers R@3= 0.62 and R@1= 0.43 ) and human evaluations. | meaning-changed | 2008.11015 | 2 |
One of the main conclusions of our analysis is that BERT performs a decent job in capturing high-level sense distinctions , 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 performs a decent job in capturing high-level sense distinctions , 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 | 1 |
One of the main conclusions of our analysis is that BERT performs a decent job in capturing high-level sense distinctions , 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 performs a decent job in capturing high-level sense distinctions , even when a limited number of examples is available for each word sense. | One of the main conclusions of our analysis is that BERT performs a decent job in capturing high-level sense distinctions accurately , even when a limited number of examples is available for each word sense. | clarity | 2008.11608 | 1 |
We also perform an in-depth comparison of the two main language model based WSD strategies, i.e., fine-tuning and feature extraction, finding that the latter approach is more robust with respect to sense bias and it can better exploit limited available training data . | <meaning-changed> We also perform an in-depth comparison of the two main language model based WSD strategies, i.e., fine-tuning and feature extraction, finding that the latter approach is more robust with respect to sense bias and it can better exploit limited available training data . | We also perform an in-depth comparison of the two main language model based WSD strategies, i.e., fine-tuning and feature extraction, finding that the latter approach is more robust with respect to sense bias and it can better exploit limited available training data . 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 . | meaning-changed | 2008.11608 | 1 |
However, there is still little knowledge about their capabilities and potential limitations for encoding and recovering word senses. | <fluency> However, there is still little knowledge about their capabilities and potential limitations for encoding and recovering word senses. | However, there is still little knowledge about their capabilities and potential limitations in encoding and recovering word senses. | fluency | 2008.11608 | 2 |