EMNLP 2020 Papers

In this paper, we focus on the imbalance issue, which is rarely studied in aspect term extraction and aspect sentiment classification when regarding them as sequence labeling tasks. Besides, previous works usually ignore the interaction between aspect terms when labeling polarities. We propose a GRadient hArmonized and CascadEd labeling model (GRACE) to solve these problems. Specifically, a cascaded labeling module is developed to enhance the interchange between aspect terms and improve the attention of sentiment tokens when labeling sentiment polarities. The polarities sequence is designed to depend on the generated aspect terms labels. To alleviate the imbalance issue, we extend the gradient harmonized mechanism used in object detection to the aspect-based sentiment analysis by adjusting the weight of each label dynamically. The proposed GRACE adopts a post-pretraining BERT as its backbone. Experimental results demonstrate that the proposed model achieves consistency improvement on multiple benchmark datasets and generates state-of-the-art results.

In this paper, we propose a sequence contrast loss driven text generation framework, which learns the difference between real texts and generated texts and uses that difference. Specifically, our discriminator contains a discriminative sequence generator instead of a binary classifier, and measures the ‘relative realism’ of generated texts against real texts by making use of them simultaneously. Moreover, our generator uses discriminative sequences to directly improve itself, which not only replaces the gradient propagation process from the discriminator to the generator, but also avoids the time-consuming sampling process of estimating rewards in some previous methods. We conduct extensive experiments with various metrics, substantiating that our framework brings improvements in terms of training stability and the quality of generated texts.

Domain adaptation or transfer learning using pre-trained language models such as BERT has proven to be an effective approach for many natural language processing tasks. In this work, we propose to formulate word sense disambiguation as a relevance ranking task, and fine-tune BERT on sequence-pair ranking task to select the most probable sense definition given a context sentence and a list of candidate sense definitions. We also introduce a data augmentation technique for WSD using existing example sentences from WordNet. Using the proposed training objective and data augmentation technique, our models are able to achieve state-of-the-art results on the English all-words benchmark datasets.

We investigate the problem of generating informative questions in information-asymmetric conversations. Unlike previous work on question generation which largely assumes knowledge of what the answer might be, we are interested in the scenario where the questioner is not given the context from which answers are drawn, but must reason pragmatically about how to acquire new information, given the shared conversation history. We identify two core challenges: (1) formally defining the informativeness of potential questions, and (2) exploring the prohibitively large space of potential questions to find the good candidates. To generate pragmatic questions, we use reinforcement learning to optimize an informativeness metric we propose, combined with a reward function designed to promote more specific questions. We demonstrate that the resulting pragmatic questioner substantially improves the informativeness and specificity of questions generated over a baseline model, as evaluated by our metrics as well as humans.

Online search engines are a popular source of medical information for users, where users can enter questions and obtain relevant answers. It is desirable to generate answer summaries for online search engines, particularly summaries that can reveal direct answers to questions. Moreover, answer summaries are expected to reveal the most relevant information in response to questions; hence, the summaries should be generated with a focus on the question, which is a challenging topic-focused summarization task. In this paper, we propose an approach that utilizes graph convolution networks and question-focused dual attention for Chinese medical answer summarization. We first organize the original long answer text into a medical concept graph with graph convolution networks to better understand the internal structure of the text and the correlation between medical concepts. Then, we introduce a question-focused dual attention mechanism to generate summaries relevant to questions. Experimental results demonstrate that the proposed model can generate more coherent and informative summaries compared with baseline models.

State-of-the-art neural machine translation methods employ massive amounts of parameters. Drastically reducing computational costs of such methods without affecting performance has been up to this point unsuccessful. To this end, we propose FullyQT: an all-inclusive quantization strategy for the Transformer. To the best of our knowledge, we are the first to show that it is possible to avoid any loss in translation quality with a fully quantized Transformer. Indeed, compared to full-precision, our 8-bit models score greater or equal BLEU on most tasks. Comparing ourselves to all previously proposed methods, we achieve state-of-the-art quantization results.

A popular multimedia news format nowadays is providing users with a lively video and a corresponding news article, which is employed by influential news media including CNN, BBC, and social media including Twitter and Weibo. In such a case, automatically choosing a proper cover frame of the video and generating an appropriate textual summary of the article can help editors save time, and readers make the decision more effectively. Hence, in this paper, we propose the task of Video-based Multimodal Summarization with Multimodal Output (VMSMO) to tackle such a problem. The main challenge in this task is to jointly model the temporal dependency of video with semantic meaning of article. To this end, we propose a Dual-Interaction-based Multimodal Summarizer (DIMS), consisting of a dual interaction module and multimodal generator. In the dual interaction module, we propose a conditional self-attention mechanism that captures local semantic information within video and a global-attention mechanism that handles the semantic relationship between news text and video from a high level. Extensive experiments conducted on a large-scale real-world VMSMO dataset show that DIMS achieves the state-of-the-art performance in terms of both automatic metrics and human evaluations.

Automated evaluation metrics as a stand-in for manual evaluation are an essential part of the development of text-generation tasks such as text summarization. However, while the field has progressed, our standard metrics have not – for nearly 20 years ROUGE has been the standard evaluation in most summarization papers. In this paper, we make an attempt to re-evaluate the evaluation method for text summarization: assessing the reliability of automatic metrics using top-scoring system outputs, both abstractive and extractive, on recently popular datasets for both system-level and summary-level evaluation settings. We find that conclusions about evaluation metrics on older datasets do not necessarily hold on modern datasets and systems. We release a dataset of human judgments that are collected from 25 top-scoring neural summarization systems (14 abstractive and 11 extractive).

The most common metrics for assessing summarization algorithms do not account for whether summaries are factually consistent with source documents. We propose a weakly-supervised, model-based approach for verifying factual consistency and identifying conflicts between source documents and generated summaries. Training data is generated by applying a series of rule-based transformations to the sentences of source documents. The factual consistency model is then trained jointly for three tasks: 1) predict whether each summary sentence is factually consistent or not, 2) in either case, extract a span in the source document to support this consistency prediction, 3) for each summary sentence that is deemed inconsistent, extract the inconsistent span from it. Transferring this model to summaries generated by several neural models reveals that this highly scalable approach outperforms previous models, including those trained with strong supervision using datasets from related domains, such as natural language inference and fact checking. Additionally, human evaluation shows that the auxiliary span extraction tasks provide useful assistance in the process of verifying factual consistency. We also release a manually annotated dataset for factual consistency verification, code for training data generation, and trained model weights at https://github.com/salesforce/factCC.

Pre-trained neural abstractive summarization systems have dominated extractive strategies on news summarization performance, at least in terms of ROUGE. However, system-generated abstractive summaries often face the pitfall of factual inconsistency: generating incorrect facts with respect to the source text. To address this challenge, we propose Span-Fact, a suite of two factual correction models that leverages knowledge learned from question answering models to make corrections in system-generated summaries via span selection. Our models employ single or multi-masking strategies to either iteratively or auto-regressively replace entities in order to ensure semantic consistency w.r.t. the source text, while retaining the syntactic structure of summaries generated by abstractive summarization models. Experiments show that our models significantly boost the factual consistency of system-generated summaries without sacrificing summary quality in terms of both automatic metrics and human evaluation.

We present a method to produce abstractive summaries of long documents that exceed several thousand words via neural abstractive summarization. We perform a simple extractive step before generating a summary, which is then used to condition the transformer language model on relevant information before being tasked with generating a summary. We also show that this approach produces more abstractive summaries compared to prior work that employs a copy mechanism while still achieving higher ROUGE scores. We provide extensive comparisons with strong baseline methods, prior state of the art work as well as multiple variants of our approach including those using only transformers, only extractive techniques and combinations of the two. We examine these models using four different summarization tasks and datasets: arXiv papers, PubMed papers, the Newsroom and BigPatent datasets. We find that transformer based methods produce summaries with fewer n-gram copies, leading to n-gram copying statistics that are more similar to human generated abstracts. We include a human evaluation, finding that transformers are ranked highly for coherence and fluency, but purely extractive methods score higher for informativeness and relevance. We hope that these architectures and experiments may serve as strong points of comparison for future work. Note: The abstract above was collaboratively written by the authors and one of the models presented in this paper based on an earlier draft of this paper.

Two main approaches for evaluating the quality of machine-generated rationales are: 1) using human rationales as a gold standard; and 2) automated metrics based on how rationales affect model behavior. An open question, however, is how human rationales fare with these automatic metrics. Analyzing a variety of datasets and models, we find that human rationales do not necessarily perform well on these metrics. To unpack this finding, we propose improved metrics to account for model-dependent baseline performance. We then propose two methods to further characterize rationale quality, one based on model retraining and one on using “fidelity curves” to reveal properties such as irrelevance and redundancy. Our work leads to actionable suggestions for evaluating and characterizing rationales.

Large datasets have become commonplace in NLP research. However, the increased emphasis on data quantity has made it challenging to assess the quality of data. We introduce Data Maps—a model-based tool to characterize and diagnose datasets. We leverage a largely ignored source of information: the behavior of the model on individual instances during training (training dynamics) for building data maps. This yields two intuitive measures for each example—the model’s confidence in the true class, and the variability of this confidence across epochs—obtained in a single run of training. Experiments on four datasets show that these model-dependent measures reveal three distinct regions in the data map, each with pronounced characteristics. First, our data maps show the presence of “ambiguous” regions with respect to the model, which contribute the most towards out-of-distribution generalization. Second, the most populous regions in the data are “easy to learn” for the model, and play an important role in model optimization. Finally, data maps uncover a region with instances that the model finds “hard to learn”; these often correspond to labeling errors. Our results indicate that a shift in focus from quantity to quality of data could lead to robust models and improved out-of-distribution generalization.

Despite its importance to experimental design, statistical power (the probability that, given a real effect, an experiment will reject the null hypothesis) has largely been ignored by the NLP community. Underpowered experiments make it more difficult to discern the difference between statistical noise and meaningful model improvements, and increase the chances of exaggerated findings. By meta-analyzing a set of existing NLP papers and datasets, we characterize typical power for a variety of settings and conclude that underpowered experiments are common in the NLP literature. In particular, for several tasks in the popular GLUE benchmark, small test sets mean that most attempted comparisons to state of the art models will not be adequately powered. Similarly, based on reasonable assumptions, we find that the most typical experimental design for human rating studies will be underpowered to detect small model differences, of the sort that are frequently studied. For machine translation, we find that typical test sets of 2000 sentences have approximately 75% power to detect differences of 1 BLEU point. To improve the situation going forward, we give an overview of best practices for power analysis in NLP and release a series of notebooks to assist with future power analyses.

There is increasing interest in assessing the linguistic knowledge encoded in neural representations. A popular approach is to attach a diagnostic classifier – or ”probe” – to perform supervised classification from internal representations. However, how to select a good probe is in debate. Hewitt and Liang (2019) showed that a high performance on diagnostic classification itself is insufficient, because it can be attributed to either ”the representation being rich in knowledge”, or ”the probe learning the task”, which Pimentel et al. (2020) challenged. We show this dichotomy is valid information-theoretically. In addition, we find that the ”good probe” criteria proposed by the two papers, selectivity (Hewitt and Liang, 2019) and information gain (Pimentel et al., 2020), are equivalent – the errors of their approaches are identical (modulo irrelevant terms). Empirically, these two selection criteria lead to results that highly agree with each other.

Medical dialogue systems are promising in assisting in telemedicine to increase access to healthcare services, improve the quality of patient care, and reduce medical costs. To facilitate the research and development of medical dialogue systems, we build large-scale medical dialogue datasets – MedDialog, which contain 1) a Chinese dataset with 3.4 million conversations between patients and doctors, 11.3 million utterances, 660.2 million tokens, covering 172 specialties of diseases, and 2) an English dataset with 0.26 million conversations, 0.51 million utterances, 44.53 million tokens, covering 96 specialties of diseases. To our best knowledge, MedDialog is the largest medical dialogue dataset to date. We pretrain several dialogue generation models on the Chinese MedDialog dataset, including Transformer, GPT, BERT-GPT, and compare their performance. It is shown that models trained on MedDialog are able to generate clinically correct and doctor-like medical dialogues. We also study the transferability of models trained on MedDialog to low-resource medical dialogue generation tasks. It is shown that via transfer learning which finetunes the models pretrained on MedDialog, the performance on medical dialogue generation tasks with small datasets can be greatly improved, as shown in human evaluation and automatic evaluation. The datasets and code are available at https://github.com/UCSD-AI4H/Medical-Dialogue-System

Automatically evaluating dialogue coherence is a challenging but high-demand ability for developing high-quality open-domain dialogue systems. However, current evaluation metrics consider only surface features or utterance-level semantics, without explicitly considering the fine-grained topic transition dynamics of dialogue flows. Here, we first consider that the graph structure constituted with topics in a dialogue can accurately depict the underlying communication logic, which is a more natural way to produce persuasive metrics. Capitalized on the topic-level dialogue graph, we propose a new evaluation metric GRADE, which stands for Graph-enhanced Representations for Automatic Dialogue Evaluation. Specifically, GRADE incorporates both coarse-grained utterance-level contextualized representations and fine-grained topic-level graph representations to evaluate dialogue coherence. The graph representations are obtained by reasoning over topic-level dialogue graphs enhanced with the evidence from a commonsense graph, including k-hop neighboring representations and hop-attention weights. Experimental results show that our GRADE significantly outperforms other state-of-the-art metrics on measuring diverse dialogue models in terms of the Pearson and Spearman correlations with human judgments. Besides, we release a new large-scale human evaluation benchmark to facilitate future research on automatic metrics.

Response selection plays a vital role in building retrieval-based conversation systems. Despite that response selection is naturally a learning-to-rank problem, most prior works take a point-wise view and train binary classifiers for this task: each response candidate is labeled either relevant (one) or irrelevant (zero). On the one hand, this formalization can be sub-optimal due to its ignorance of the diversity of response quality. On the other hand, annotating grayscale data for learning-to-rank can be prohibitively expensive and challenging. In this work, we show that grayscale data can be automatically constructed without human effort. Our method employs off-the-shelf response retrieval models and response generation models as automatic grayscale data generators. With the constructed grayscale data, we propose multi-level ranking objectives for training, which can (1) teach a matching model to capture more fine-grained context-response relevance difference and (2) reduce the train-test discrepancy in terms of distractor strength. Our method is simple, effective, and universal. Experiments on three benchmark datasets and four state-of-the-art matching models show that the proposed approach brings significant and consistent performance improvements.

Structured belief states are crucial for user goal tracking and database query in task-oriented dialog systems. However, training belief trackers often requires expensive turn-level annotations of every user utterance. In this paper we aim at alleviating the reliance on belief state labels in building end-to-end dialog systems, by leveraging unlabeled dialog data towards semi-supervised learning. We propose a probabilistic dialog model, called the LAtent BElief State (LABES) model, where belief states are represented as discrete latent variables and jointly modeled with system responses given user inputs. Such latent variable modeling enables us to develop semi-supervised learning under the principled variational learning framework. Furthermore, we introduce LABES-S2S, which is a copy-augmented Seq2Seq model instantiation of LABES. In supervised experiments, LABES-S2S obtains strong results on three benchmark datasets of different scales. In utilizing unlabeled dialog data, semi-supervised LABES-S2S significantly outperforms both supervised-only and semi-supervised baselines. Remarkably, we can reduce the annotation demands to 50% without performance loss on MultiWOZ.