EMNLP 2020 论文列表

The Data-to-Text task aims to generate human-readable text for describing some given structured data enabling more interpretability. However, the typical generation task is confined to a few particular domains since it requires well-aligned data which is difficult and expensive to obtain. Using partially-aligned data is an alternative way of solving the dataset scarcity problem. This kind of data is much easier to obtain since it can be produced automatically. However, using this kind of data induces the over-generation problem posing difficulties for existing models, which tends to add unrelated excerpts during the generation procedure. In order to effectively utilize automatically annotated partially-aligned datasets, we extend the traditional generation task to a refined task called Partially-Aligned Data-to-Text Generation (PADTG) which is more practical since it utilizes automatically annotated data for training and thus considerably expands the application domains. To tackle this new task, we propose a novel distant supervision generation framework. It firstly estimates the input data’s supportiveness for each target word with an estimator and then applies a supportiveness adaptor and a rebalanced beam search to harness the over-generation problem in the training and generation phases respectively. We also contribute a partially-aligned dataset (The data and source code of this paper can be obtained from https://github.com/fuzihaofzh/distant_supervision_nlg) by sampling sentences from Wikipedia and automatically extracting corresponding KB triples for each sentence from Wikidata. The experimental results show that our framework outperforms all baseline models as well as verify the feasibility of utilizing partially-aligned data.

Despite recent advances in neural text generation, encoding the rich diversity in human language remains elusive. We argue that the sub-optimal text generation is mainly attributable to the imbalanced token distribution, which particularly misdirects the learning model when trained with the maximum-likelihood objective. As a simple yet effective remedy, we propose two novel methods, Fˆ2-Softmax and MefMax, for a balanced training even with the skewed frequency distribution. MefMax assigns tokens uniquely to frequency classes, trying to group tokens with similar frequencies and equalize frequency mass between the classes. Fˆ2-Softmax then decomposes a probability distribution of the target token into a product of two conditional probabilities of (1) frequency class, and (2) token from the target frequency class. Models learn more uniform probability distributions because they are confined to subsets of vocabularies. Significant performance gains on seven relevant metrics suggest the supremacy of our approach in improving not only the diversity but also the quality of generated texts.

Despite the success of existing referenced metrics (e.g., BLEU and MoverScore), they correlate poorly with human judgments for open-ended text generation including story or dialog generation because of the notorious one-to-many issue: there are many plausible outputs for the same input, which may differ substantially in literal or semantics from the limited number of given references. To alleviate this issue, we propose UNION, a learnable UNreferenced metrIc for evaluating Open-eNded story generation, which measures the quality of a generated story without any reference. Built on top of BERT, UNION is trained to distinguish human-written stories from negative samples and recover the perturbation in negative stories. We propose an approach of constructing negative samples by mimicking the errors commonly observed in existing NLG models, including repeated plots, conflicting logic, and long-range incoherence. Experiments on two story datasets demonstrate that UNION is a reliable measure for evaluating the quality of generated stories, which correlates better with human judgments and is more generalizable than existing state-of-the-art metrics.

Neural language models are often trained with maximum likelihood estimation (MLE), where the next word is generated conditioned on the ground-truth word tokens. During testing, however, the model is instead conditioned on previously generated tokens, resulting in what is termed exposure bias. To reduce this gap between training and testing, we propose using optimal transport (OT) to match the sequences generated in these two modes. We examine the necessity of adding Student-Forcing scheme during training with an imitation learning interpretation. An extension is further proposed to improve the OT learning for long sequences, based on the structural and contextual information of the text sequences. The effectiveness of the proposed method is validated on machine translation, text summarization, and text generation tasks.

Despite the subjective nature of many NLP tasks, most NLU evaluations have focused on using the majority label with presumably high agreement as the ground truth. Less attention has been paid to the distribution of human opinions. We collect ChaosNLI, a dataset with a total of 464,500 annotations to study Collective HumAn OpinionS in oft-used NLI evaluation sets. This dataset is created by collecting 100 annotations per example for 3,113 examples in SNLI and MNLI and 1,532 examples in αNLI. Analysis reveals that: (1) high human disagreement exists in a noticeable amount of examples in these datasets; (2) the state-of-the-art models lack the ability to recover the distribution over human labels; (3) models achieve near-perfect accuracy on the subset of data with a high level of human agreement, whereas they can barely beat a random guess on the data with low levels of human agreement, which compose most of the common errors made by state-of-the-art models on the evaluation sets. This questions the validity of improving model performance on old metrics for the low-agreement part of evaluation datasets. Hence, we argue for a detailed examination of human agreement in future data collection efforts, and evaluating model outputs against the distribution over collective human opinions.

Pre-trained contextual representations like BERT have achieved great success in natural language processing. However, the sentence embeddings from the pre-trained language models without fine-tuning have been found to poorly capture semantic meaning of sentences. In this paper, we argue that the semantic information in the BERT embeddings is not fully exploited. We first reveal the theoretical connection between the masked language model pre-training objective and the semantic similarity task theoretically, and then analyze the BERT sentence embeddings empirically. We find that BERT always induces a non-smooth anisotropic semantic space of sentences, which harms its performance of semantic similarity. To address this issue, we propose to transform the anisotropic sentence embedding distribution to a smooth and isotropic Gaussian distribution through normalizing flows that are learned with an unsupervised objective. Experimental results show that our proposed BERT-flow method obtains significant performance gains over the state-of-the-art sentence embeddings on a variety of semantic textual similarity tasks. The code is available at https://github.com/bohanli/BERT-flow.

Negation is underrepresented in existing natural language inference benchmarks. Additionally, one can often ignore the few negations in existing benchmarks and still make the right inference judgments. In this paper, we present a new benchmark for natural language inference in which negation plays a critical role. We also show that state-of-the-art transformers struggle making inference judgments with the new pairs.

Natural language is characterized by compositionality: the meaning of a complex expression is constructed from the meanings of its constituent parts. To facilitate the evaluation of the compositional abilities of language processing architectures, we introduce COGS, a semantic parsing dataset based on a fragment of English. The evaluation portion of COGS contains multiple systematic gaps that can only be addressed by compositional generalization; these include new combinations of familiar syntactic structures, or new combinations of familiar words and familiar structures. In experiments with Transformers and LSTMs, we found that in-distribution accuracy on the COGS test set was near-perfect (96–99%), but generalization accuracy was substantially lower (16–35%) and showed high sensitivity to random seed (+-6–8%). These findings indicate that contemporary standard NLP models are limited in their compositional generalization capacity, and position COGS as a good way to measure progress.

Recognizing the flow of time in a story is a crucial aspect of understanding it. Prior work related to time has primarily focused on identifying temporal expressions or relative sequencing of events, but here we propose computationally annotating each line of a book with wall clock times, even in the absence of explicit time-descriptive phrases. To do so, we construct a data set of hourly time phrases from 52,183 fictional books. We then construct a time-of-day classification model that achieves an average error of 2.27 hours. Furthermore, we show that by analyzing a book in whole using dynamic programming of breakpoints, we can roughly partition a book into segments that each correspond to a particular time-of-day. This approach improves upon baselines by over two hour. Finally, we apply our model to a corpus of literature categorized by different periods in history, to show interesting trends of hourly activity throughout the past. Among several observations we find that the fraction of events taking place past 10 P.M jumps past 1880 - coincident with the advent of the electric light bulb and city lights.

Existing research for question generation encodes the input text as a sequence of tokens without explicitly modeling fact information. These models tend to generate irrelevant and uninformative questions. In this paper, we explore to incorporate facts in the text for question generation in a comprehensive way. We present a novel task of question generation given a query path in the knowledge graph constructed from the input text. We divide the task into two steps, namely, query representation learning and query-based question generation. We formulate query representation learning as a sequence labeling problem for identifying the involved facts to form a query and employ an RNN-based generator for question generation. We first train the two modules jointly in an end-to-end fashion, and further enforce the interaction between these two modules in a variational framework. We construct the experimental datasets on top of SQuAD and results show that our model outperforms other state-of-the-art approaches, and the performance margin is larger when target questions are complex. Human evaluation also proves that our model is able to generate relevant and informative questions.

Simultaneously modeling source code and natural language has many exciting applications in automated software development and understanding. Pursuant to achieving such technology, we introduce PyMT5, the Python method text-to-text transfer transformer, which is trained to translate between all pairs of Python method feature combinations: a single model that can both predict whole methods from natural language documentation strings (docstrings) and summarize code into docstrings of any common style. We present an analysis and modeling effort of a large-scale parallel corpus of 26 million Python methods and 7.7 million method-docstring pairs, demonstrating that for docstring and method generation, PyMT5 outperforms similarly-sized auto-regressive language models (GPT2) which were English pre-trained or randomly initialized. On the CodeSearchNet test set, our best model predicts 92.1% syntactically correct method bodies, achieved a BLEU score of 8.59 for method generation and 16.3 for docstring generation (summarization), and achieved a ROUGE-L F-score of 24.8 for method generation and 36.7 for docstring generation.

In this paper, we study automatic rumor detection for in social media at the event level where an event consists of a sequence of posts organized according to the posting time. It is common that the state of an event is dynamically evolving. However, most of the existing methods to this task ignored this problem, and established a global representation based on all the posts in the event’s life cycle. Such coarse-grained methods failed to capture the event’s unique features in different states. To address this limitation, we propose a state-independent and time-evolving Network (STN) for rumor detection based on fine-grained event state detection and segmentation. Given an event composed of a sequence of posts, STN first predicts the corresponding sequence of states and segments the event into several state-independent sub-events. For each sub-event, STN independently trains an encoder to learn the feature representation for that sub-event and incrementally fuses the representation of the current sub-event with previous ones for rumor prediction. This framework can more accurately learn the representation of an event in the initial stage and enable early rumor detection. Experiments on two benchmark datasets show that STN can significantly improve the rumor detection accuracy in comparison with some strong baseline systems. We also design a new evaluation metric to measure the performance of early rumor detection, under which STN shows a higher advantage in comparison.

Data augmentation is proven to be effective in many NLU tasks, especially for those suffering from data scarcity. In this paper, we present a powerful and easy to deploy text augmentation framework, Data Boost, which augments data through reinforcement learning guided conditional generation. We evaluate Data Boost on three diverse text classification tasks under five different classifier architectures. The result shows that Data Boost can boost the performance of classifiers especially in low-resource data scenarios. For instance, Data Boost improves F1 for the three tasks by 8.7% on average when given only 10% of the whole data for training. We also compare Data Boost with six prior text augmentation methods. Through human evaluations (N=178), we confirm that Data Boost augmentation has comparable quality as the original data with respect to readability and class consistency.

Advances on deep generative models have attracted significant research interest in neural topic modeling. The recently proposed Adversarial-neural Topic Model models topics with an adversarially trained generator network and employs Dirichlet prior to capture the semantic patterns in latent topics. It is effective in discovering coherent topics but unable to infer topic distributions for given documents or utilize available document labels. To overcome such limitations, we propose Topic Modeling with Cycle-consistent Adversarial Training (ToMCAT) and its supervised version sToMCAT. ToMCAT employs a generator network to interpret topics and an encoder network to infer document topics. Adversarial training and cycle-consistent constraints are used to encourage the generator and the encoder to produce realistic samples that coordinate with each other. sToMCAT extends ToMCAT by incorporating document labels into the topic modeling process to help discover more coherent topics. The effectiveness of the proposed models is evaluated on unsupervised/supervised topic modeling and text classification. The experimental results show that our models can produce both coherent and informative topics, outperforming a number of competitive baselines.

Current text classification methods typically require a good number of human-labeled documents as training data, which can be costly and difficult to obtain in real applications. Humans can perform classification without seeing any labeled examples but only based on a small set of words describing the categories to be classified. In this paper, we explore the potential of only using the label name of each class to train classification models on unlabeled data, without using any labeled documents. We use pre-trained neural language models both as general linguistic knowledge sources for category understanding and as representation learning models for document classification. Our method (1) associates semantically related words with the label names, (2) finds category-indicative words and trains the model to predict their implied categories, and (3) generalizes the model via self-training. We show that our model achieves around 90% accuracy on four benchmark datasets including topic and sentiment classification without using any labeled documents but learning from unlabeled data supervised by at most 3 words (1 in most cases) per class as the label name.

Deep neural network models have helped named entity recognition achieve amazing performance without handcrafting features. However, existing systems require large amounts of human annotated training data. Efforts have been made to replace human annotations with external knowledge (e.g., NE dictionary, part-of-speech tags), while it is another challenge to obtain such effective resources. In this work, we propose a fully unsupervised NE recognition model which only needs to take informative clues from pre-trained word embeddings.We first apply Gaussian Hidden Markov Model and Deep Autoencoding Gaussian Mixture Model on word embeddings for entity span detection and type prediction, and then further design an instance selector based on reinforcement learning to distinguish positive sentences from noisy sentences and then refine these coarse-grained annotations through neural networks. Extensive experiments on two CoNLL benchmark NER datasets (CoNLL-2003 English dataset and CoNLL-2002 Spanish dataset) demonstrate that our proposed light NE recognition model achieves remarkable performance without using any annotated lexicon or corpus.

In this work, we aim at equipping pre-trained language models with structured knowledge. We present two self-supervised tasks learning over raw text with the guidance from knowledge graphs. Building upon entity-level masked language models, our first contribution is an entity masking scheme that exploits relational knowledge underlying the text. This is fulfilled by using a linked knowledge graph to select informative entities and then masking their mentions. In addition, we use knowledge graphs to obtain distractors for the masked entities, and propose a novel distractor-suppressed ranking objective that is optimized jointly with masked language model. In contrast to existing paradigms, our approach uses knowledge graphs implicitly, only during pre-training, to inject language models with structured knowledge via learning from raw text. It is more efficient than retrieval-based methods that perform entity linking and integration during finetuning and inference, and generalizes more effectively than the methods that directly learn from concatenated graph triples. Experiments show that our proposed model achieves improved performance on five benchmarks, including question answering and knowledge base completion.

Current approaches to empathetic response generation view the set of emotions expressed in the input text as a flat structure, where all the emotions are treated uniformly. We argue that empathetic responses often mimic the emotion of the user to a varying degree, depending on its positivity or negativity and content. We show that the consideration of these polarity-based emotion clusters and emotional mimicry results in improved empathy and contextual relevance of the response as compared to the state-of-the-art. Also, we introduce stochasticity into the emotion mixture that yields emotionally more varied empathetic responses than the previous work. We demonstrate the importance of these factors to empathetic response generation using both automatic- and human-based evaluations. The implementation of MIME is publicly available at https://github.com/declare-lab/MIME.

Given the growing ubiquity of emojis in language, there is a need for methods and resources that shed light on their meaning and communicative role. One conspicuous aspect of emojis is their use to convey affect in ways that may otherwise be non-trivial to achieve. In this paper, we seek to explore the connection between emojis and emotions by means of a new dataset consisting of human-solicited association ratings. We additionally conduct experiments to assess to what extent such associations can be inferred from existing data in an unsupervised manner. Our experiments show that this succeeds when high-quality word-level information is available.