EMNLP 2020 论文列表

Privacy policy documents are long and verbose. A question answering (QA) system can assist users in finding the information that is relevant and important to them. Prior studies in this domain frame the QA task as retrieving the most relevant text segment or a list of sentences from the policy document given a question. On the contrary, we argue that providing users with a short text span from policy documents reduces the burden of searching the target information from a lengthy text segment. In this paper, we present PolicyQA, a dataset that contains 25,017 reading comprehension style examples curated from an existing corpus of 115 website privacy policies. PolicyQA provides 714 human-annotated questions written for a wide range of privacy practices. We evaluate two existing neural QA models and perform rigorous analysis to reveal the advantages and challenges offered by PolicyQA.

Attention mechanisms have improved the performance of NLP tasks while allowing models to remain explainable. Self-attention is currently widely used, however interpretability is difficult due to the numerous attention distributions. Recent work has shown that model representations can benefit from label-specific information, while facilitating interpretation of predictions. We introduce the Label Attention Layer: a new form of self-attention where attention heads represent labels. We test our novel layer by running constituency and dependency parsing experiments and show our new model obtains new state-of-the-art results for both tasks on both the Penn Treebank (PTB) and Chinese Treebank. Additionally, our model requires fewer self-attention layers compared to existing work. Finally, we find that the Label Attention heads learn relations between syntactic categories and show pathways to analyze errors.

Traditional (unstructured) pruning methods for a Transformer model focus on regularizing the individual weights by penalizing them toward zero. In this work, we explore spectral-normalized identity priors (SNIP), a structured pruning approach which penalizes an entire residual module in a Transformer model toward an identity mapping. Our method identifies and discards unimportant non-linear mappings in the residual connections by applying a thresholding operator on the function norm, and is applicable to any structured module including a single attention head, an entire attention blocks, or a feed-forward subnetwork. Furthermore, we introduce spectral normalization to stabilize the distribution of the post-activation values of the Transformer layers, further improving the pruning effectiveness of the proposed methodology. We conduct experiments with BERT on 5 GLUE benchmark tasks to demonstrate that SNIP achieves effective pruning results while maintaining comparable performance. Specifically, we improve the performance over the state-of-the-art by 0.5 to 1.0% on average at 50% compression ratio.

This work proposes the use of a pretrained sequence-to-sequence model for document ranking. Our approach is fundamentally different from a commonly adopted classification-based formulation based on encoder-only pretrained transformer architectures such as BERT. We show how a sequence-to-sequence model can be trained to generate relevance labels as “target tokens”, and how the underlying logits of these target tokens can be interpreted as relevance probabilities for ranking. Experimental results on the MS MARCO passage ranking task show that our ranking approach is superior to strong encoder-only models. On three other document retrieval test collections, we demonstrate a zero-shot transfer-based approach that outperforms previous state-of-the-art models requiring in-domain cross-validation. Furthermore, we find that our approach significantly outperforms an encoder-only architecture in a data-poor setting. We investigate this observation in more detail by varying target tokens to probe the model’s use of latent knowledge. Surprisingly, we find that the choice of target tokens impacts effectiveness, even for words that are closely related semantically. This finding sheds some light on why our sequence-to-sequence formulation for document ranking is effective. Code and models are available at pygaggle.ai.

Vision-and-Language Navigation (VLN) is a natural language grounding task where an agent learns to follow language instructions and navigate to specified destinations in real-world environments. A key challenge is to recognize and stop at the correct location, especially for complicated outdoor environments. Existing methods treat the STOP action equally as other actions, which results in undesirable behaviors that the agent often fails to stop at the destination even though it might be on the right path. Therefore, we propose Learning to Stop (L2Stop), a simple yet effective policy module that differentiates STOP and other actions. Our approach achieves the new state of the art on a challenging urban VLN dataset Touchdown, outperforming the baseline by 6.89% (absolute improvement) on Success weighted by Edit Distance (SED).

We present a method for creating parallel data to train Seq2Seq neural networks for sentiment transfer. Most systems for this task, which can be viewed as monolingual machine translation (MT), have relied on unsupervised methods, such as Generative Adversarial Networks (GANs)-inspired approaches, for coping with the lack of parallel corpora. Given that the literature shows that Seq2Seq methods have been consistently outperforming unsupervised methods in MT-related tasks, in this work we exploit the use of semantic similarity computation for converting non-parallel data onto a parallel corpus. That allows us to train a transformer neural network for the sentiment transfer task, and compare its performance against unsupervised approaches. With experiments conducted on two well-known public datasets, i.e. Yelp and Amazon, we demonstrate that the proposed methodology outperforms existing unsupervised methods very consistently in fluency, and presents competitive results in terms of sentiment conversion and content preservation. We believe that this works opens up an opportunity for seq2seq neural networks to be better exploited in problems for which they have not been applied owing to the lack of parallel training data.

In this work, we explore the way to quickly adjust an existing named entity recognition (NER) system to make it capable of recognizing entity types not defined in the system. As an illustrative example, consider the case that a NER system has been built to recognize person and organization names, and now it requires to additionally recognize job titles. Such a situation is common in the industrial areas, where the entity types required to recognize vary a lot in different products and keep changing. To avoid laborious data labeling and achieve fast adaptation, we propose to adjust the existing NER system using the previously labeled data and entity lexicons of the newly introduced entity types. We formulate such a task as a partially supervised learning problem and accordingly propose an effective algorithm to solve the problem. Comprehensive experimental studies on several public NER datasets validate the effectiveness of our method.

Transformers have shown great success in learning representations for language modelling. However, an open challenge still remains on how to systematically aggregate semantic information (word embedding) with positional (or temporal) information (word orders). In this work, we propose a new architecture to aggregate the two sources of information using cascaded semantic and positional self-attention network (CSPAN) in the context of document classification. The CSPAN uses a semantic self-attention layer cascaded with Bi-LSTM to process the semantic and positional information in a sequential manner, and then adaptively combine them together through a residue connection. Compared with commonly used positional encoding schemes, CSPAN can exploit the interaction between semantics and word positions in a more interpretable and adaptive manner, and the classification performance can be notably improved while simultaneously preserving a compact model size and high convergence rate. We evaluate the CSPAN model on several benchmark data sets for document classification with careful ablation studies, and demonstrate the encouraging results compared with state of the art.

Bidirectional Encoder Representations from Transformers (BERT) has shown marvelous improvements across various NLP tasks, and consecutive variants have been proposed to further improve the performance of the pre-trained language models. In this paper, we target on revisiting Chinese pre-trained language models to examine their effectiveness in a non-English language and release the Chinese pre-trained language model series to the community. We also propose a simple but effective model called MacBERT, which improves upon RoBERTa in several ways, especially the masking strategy that adopts MLM as correction (Mac). We carried out extensive experiments on eight Chinese NLP tasks to revisit the existing pre-trained language models as well as the proposed MacBERT. Experimental results show that MacBERT could achieve state-of-the-art performances on many NLP tasks, and we also ablate details with several findings that may help future research. https://github.com/ymcui/MacBERT

The representations generated by many models of language (word embeddings, recurrent neural networks and transformers) correlate to brain activity recorded while people read. However, these decoding results are usually based on the brain’s reaction to syntactically and semantically sound language stimuli. In this study, we asked: how does an LSTM (long short term memory) language model, trained (by and large) on semantically and syntactically intact language, represent a language sample with degraded semantic or syntactic information? Does the LSTM representation still resemble the brain’s reaction? We found that, even for some kinds of nonsensical language, there is a statistically significant relationship between the brain’s activity and the representations of an LSTM. This indicates that, at least in some instances, LSTMs and the human brain handle nonsensical data similarly.

Mixed Boolean-Arithmetic (MBA) expressions involve both arithmetic calculation (e.g.,plus, minus, multiply) and bitwise computation (e.g., and, or, negate, xor). MBA expressions have been widely applied in software obfuscation, transforming programs from a simple form to a complex form. MBA expressions are challenging to be simplified, because the interleaving bitwise and arithmetic operations causing mathematical reduction laws to be ineffective. Our goal is to recover the original, simple form from an obfuscated MBA expression. In this paper, we first propose NeuReduce, a string to string method based on neural networks to automatically learn and reduce complex MBA expressions. We develop a comprehensive MBA dataset, including one million diversified MBA expression samples and corresponding simplified forms. After training on the dataset, NeuReduce can reduce MBA rules to homelier but mathematically equivalent forms. By comparing with three state-of-the-art MBA reduction methods, our evaluation result shows that NeuReduce outperforms all other tools in terms of accuracy, solving time, and performance overhead.

Federated learning has sparkled new interests in the deep learning society to make use of isolated data sources from independent institutes. With the development of novel training tools, we have successfully deployed federated natural language processing networks on GPU-enabled server clusters. This paper demonstrates federated training of a popular NLP model, TextCNN, with applications in sentence intent classification. Furthermore, differential privacy is introduced to protect participants in the training process, in a manageable manner. Distinguished from previous client-level privacy protection schemes, the proposed differentially private federated learning procedure is defined in the dataset sample level, inherent with the applications among institutions instead of individual users. Optimal settings of hyper-parameters for the federated TextCNN model are studied through comprehensive experiments. We also evaluated the performance of federated TextCNN model under imbalanced data load configuration. Experiments show that, the sampling ratio has a large impact on the performance of the FL models, causing up to 38.4% decrease in the test accuracy, while they are robust to different noise multiplier levels, with less than 3% variance in the test accuracy. It is also found that the FL models are sensitive to data load balancedness among client datasets. When the data load is imbalanced, model performance dropped by up to 10%.

In standard methodology for natural language processing, entities in text are typically embedded in dense vector spaces with pre-trained models. The embeddings produced this way are effective when fed into downstream models, but they require end-task fine-tuning and are fundamentally difficult to interpret. In this paper, we present an approach to creating entity representations that are human readable and achieve high performance on entity-related tasks out of the box. Our representations are vectors whose values correspond to posterior probabilities over fine-grained entity types, indicating the confidence of a typing model’s decision that the entity belongs to the corresponding type. We obtain these representations using a fine-grained entity typing model, trained either on supervised ultra-fine entity typing data (Choi et al. 2018) or distantly-supervised examples from Wikipedia. On entity probing tasks involving recognizing entity identity, our embeddings used in parameter-free downstream models achieve competitive performance with ELMo- and BERT-based embeddings in trained models. We also show that it is possible to reduce the size of our type set in a learning-based way for particular domains. Finally, we show that these embeddings can be post-hoc modified through a small number of rules to incorporate domain knowledge and improve performance.

We look into the task of generalizing word embeddings: given a set of pre-trained word vectors over a finite vocabulary, the goal is to predict embedding vectors for out-of-vocabulary words, without extra contextual information. We rely solely on the spellings of words and propose a model, along with an efficient algorithm, that simultaneously models subword segmentation and computes subword-based compositional word embedding. We call the model probabilistic bag-of-subwords (PBoS), as it applies bag-of-subwords for all possible segmentations based on their likelihood. Inspections and affix prediction experiment show that PBoS is able to produce meaningful subword segmentations and subword rankings without any source of explicit morphological knowledge. Word similarity and POS tagging experiments show clear advantages of PBoS over previous subword-level models in the quality of generated word embeddings across languages.

Syntactic information is essential for both sentiment analysis(SA) and aspect-based sentiment analysis(ABSA). Previous work has already achieved great progress utilizing Graph Convolutional Network(GCN) over dependency tree of a sentence. However, these models do not fully exploit the syntactic information obtained from dependency parsing such as the diversified types of dependency relations. The message passing process of GCN should be distinguished based on these syntactic information. To tackle this problem, we design a novel weighted graph convolutional network(WGCN) which can exploit rich syntactic information based on the feature combination. Furthermore, we utilize BERT instead of Bi-LSTM to generate contextualized representations as inputs for GCN and present an alignment method to keep word-level dependencies consistent with wordpiece unit of BERT. With our proposal, we are able to improve the state-of-the-art on four ABSA tasks out of six and two SA tasks out of three.

Multi-hop relation reasoning over knowledge base is to generate effective and interpretable relation prediction through reasoning paths. The current methods usually require sufficient training data (fact triples) for each query relation, impairing their performances over few-shot relations (with limited triples) which are common in knowledge base. To this end, we propose FIRE, a novel few-shot multi-hop relation learning model. FIRE applies reinforcement learning to model the sequential steps of multi-hop reasoning, besides performs heterogeneous structure encoding and knowledge-aware search space pruning. The meta-learning technique is employed to optimize model parameters that could quickly adapt to few-shot relations. Empirical study on two datasets demonstrate that FIRE outperforms state-of-the-art methods.

We study the problem of recognizing visual entities from the textual descriptions of their classes. Specifically, given birds’ images with free-text descriptions of their species, we learn to classify images of previously-unseen species based on specie descriptions. This setup has been studied in the vision community under the name zero-shot learning from text, focusing on learning to transfer knowledge about visual aspects of birds from seen classes to previously-unseen ones. Here, we suggest focusing on the textual description and distilling from the description the most relevant information to effectively match visual features to the parts of the text that discuss them. Specifically, (1) we propose to leverage the similarity between species, reflected in the similarity between text descriptions of the species. (2) we derive visual summaries of the texts, i.e., extractive summaries that focus on the visual features that tend to be reflected in images. We propose a simple attention-based model augmented with the similarity and visual summaries components. Our empirical results consistently and significantly outperform the state-of-the-art on the largest benchmarks for text-based zero-shot learning, illustrating the critical importance of texts for zero-shot image-recognition.

Transformer-based models have brought a radical change to neural machine translation. A key feature of the Transformer architecture is the so-called multi-head attention mechanism, which allows the model to focus simultaneously on different parts of the input. However, recent works have shown that most attention heads learn simple, and often redundant, positional patterns. In this paper, we propose to replace all but one attention head of each encoder layer with simple fixed – non-learnable – attentive patterns that are solely based on position and do not require any external knowledge. Our experiments with different data sizes and multiple language pairs show that fixing the attention heads on the encoder side of the Transformer at training time does not impact the translation quality and even increases BLEU scores by up to 3 points in low-resource scenarios.

Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.