ACL 2021 论文列表

Neural machine translation models are often biased toward the limited translation references seen during training. To amend this form of overfitting, in this paper we propose fine-tuning the models with a novel training objective based on the recently-proposed BERTScore evaluation metric. BERTScore is a scoring function based on contextual embeddings that overcomes the typical limitations of n-gram-based metrics (e.g. synonyms, paraphrases), allowing translations that are different from the references, yet close in the contextual embedding space, to be treated as substantially correct. To be able to use BERTScore as a training objective, we propose three approaches for generating soft predictions, allowing the network to remain completely differentiable end-to-end. Experiments carried out over four, diverse language pairs show improvements of up to 0.58 pp (3.28%) in BLEU score and up to 0.76 pp (0.98%) in BERTScore (F_BERT) when fine-tuning a strong baseline.

We present BERTGen, a novel, generative, decoder-only model which extends BERT by fusing multimodal and multilingual pre-trained models VL-BERT and M-BERT, respectively. BERTGen is auto-regressively trained for language generation tasks, namely image captioning, machine translation and multimodal machine translation, under a multi-task setting. With a comprehensive set of evaluations, we show that BERTGen outperforms many strong baselines across the tasks explored. We also show BERTGen’s ability for zero-shot language generation, where it exhibits competitive performance to supervised counterparts. Finally, we conduct ablation studies which demonstrate that BERTGen substantially benefits from multi-tasking and effectively transfers relevant inductive biases from the pre-trained models.

Transformer-based language models (TLMs), such as BERT, ALBERT and GPT-3, have shown strong performance in a wide range of NLP tasks and currently dominate the field of NLP. However, many researchers wonder whether these models can maintain their dominance forever. Of course, we do not have answers now, but, as an attempt to find better neural architectures and training schemes, we pretrain a simple CNN using a GAN-style learning scheme and Wikipedia data, and then integrate it with standard TLMs. We show that on the GLUE tasks, the combination of our pretrained CNN with ALBERT outperforms the original ALBERT and achieves a similar performance to that of SOTA. Furthermore, on open-domain QA (Quasar-T and SearchQA), the combination of the CNN with ALBERT or RoBERTa achieved stronger performance than SOTA and the original TLMs. We hope that this work provides a hint for developing a novel strong network architecture along with its training scheme. Our source code and models are available at https://github.com/nict-wisdom/bertac.

Analogies play a central role in human commonsense reasoning. The ability to recognize analogies such as “eye is to seeing what ear is to hearing”, sometimes referred to as analogical proportions, shape how we structure knowledge and understand language. Surprisingly, however, the task of identifying such analogies has not yet received much attention in the language model era. In this paper, we analyze the capabilities of transformer-based language models on this unsupervised task, using benchmarks obtained from educational settings, as well as more commonly used datasets. We find that off-the-shelf language models can identify analogies to a certain extent, but struggle with abstract and complex relations, and results are highly sensitive to model architecture and hyperparameters. Overall the best results were obtained with GPT-2 and RoBERTa, while configurations using BERT were not able to outperform word embedding models. Our results raise important questions for future work about how, and to what extent, pre-trained language models capture knowledge about abstract semantic relations.

Abstractive summarization for long-document or multi-document remains challenging for the Seq2Seq architecture, as Seq2Seq is not good at analyzing long-distance relations in text. In this paper, we present BASS, a novel framework for Boosting Abstractive Summarization based on a unified Semantic graph, which aggregates co-referent phrases distributing across a long range of context and conveys rich relations between phrases. Further, a graph-based encoder-decoder model is proposed to improve both the document representation and summary generation process by leveraging the graph structure. Specifically, several graph augmentation methods are designed to encode both the explicit and implicit relations in the text while the graph-propagation attention mechanism is developed in the decoder to select salient content into the summary. Empirical results show that the proposed architecture brings substantial improvements for both long-document and multi-document summarization tasks.

In this paper, we focus on the problem of citing sentence generation, which entails generating a short text to capture the salient information in a cited paper and the connection between the citing and cited paper. We present BACO, a BAckground knowledge- and COntent-based framework for citing sentence generation, which considers two types of information: (1) background knowledge by leveraging structural information from a citation network; and (2) content, which represents in-depth information about what to cite and why to cite. First, a citation network is encoded to provide background knowledge. Second, we apply salience estimation to identify what to cite by estimating the importance of sentences in the cited paper. During the decoding stage, both types of information are combined to facilitate the text generation, and then we conduct a joint training for the generator and citation function classification to make the model aware of why to cite. Our experimental results show that our framework outperforms comparative baselines.

Leveraging additional unlabeled data to boost model performance is common practice in machine learning and natural language processing. For generation tasks, if there is overlap between the additional data and the target text evaluation data, then training on the additional data is training on answers of the test set. This leads to overly-inflated scores with the additional data compared to real-world testing scenarios and problems when comparing models. We study the AMR dataset and Gigaword, which is popularly used for improving AMR-to-text generators, and find significant overlap between Gigaword and a subset of the AMR dataset. We propose methods for excluding parts of Gigaword to remove this overlap, and show that our approach leads to a more realistic evaluation of the task of AMR-to-text generation. Going forward, we give simple best-practice recommendations for leveraging additional data in AMR-to-text generation.

The ICD coding task aims at assigning codes of the International Classification of Diseases in clinical notes. Since manual coding is very laborious and prone to errors, many methods have been proposed for the automatic ICD coding task. However, existing works either ignore the long-tail of code frequency or the noisy clinical notes. To address the above issues, we propose an Interactive Shared Representation Network with Self-Distillation Mechanism. Specifically, an interactive shared representation network targets building connections among codes while modeling the co-occurrence, consequently alleviating the long-tail problem. Moreover, to cope with the noisy text issue, we encourage the model to focus on the clinical note’s noteworthy part and extract valuable information through a self-distillation learning mechanism. Experimental results on two MIMIC datasets demonstrate the effectiveness of our method.

Most fact checking models for automatic fake news detection are based on reasoning: given a claim with associated evidence, the models aim to estimate the claim veracity based on the supporting or refuting content within the evidence. When these models perform well, it is generally assumed to be due to the models having learned to reason over the evidence with regards to the claim. In this paper, we investigate this assumption of reasoning, by exploring the relationship and importance of both claim and evidence. Surprisingly, we find on political fact checking datasets that most often the highest effectiveness is obtained by utilizing only the evidence, as the impact of including the claim is either negligible or harmful to the effectiveness. This highlights an important problem in what constitutes evidence in existing approaches for automatic fake news detection.

Research on the application of NLP in symbol-based Augmentative and Alternative Communication (AAC) tools for improving social interaction support is scarce. We contribute a novel method for generating context-related vocabulary from photographs of personally relevant events aimed at supporting people with language impairments in retelling their past experiences. Performance was calculated with information retrieval concepts on the relevance of vocabulary generated for communicating a corpus of 9730 narrative phrases about events depicted in 1946 photographs. In comparison to a baseline generation composed of frequent English words, our method generated vocabulary with a 4.6 gain in mean average precision, regardless of the level of contextual information in the input photographs, and 6.9 for photographs in which contextual information was extracted correctly. We conclude by discussing how our findings provide insights for system optimization and usage.

Pretrained contextualized embeddings are powerful word representations for structured prediction tasks. Recent work found that better word representations can be obtained by concatenating different types of embeddings. However, the selection of embeddings to form the best concatenated representation usually varies depending on the task and the collection of candidate embeddings, and the ever-increasing number of embedding types makes it a more difficult problem. In this paper, we propose Automated Concatenation of Embeddings (ACE) to automate the process of finding better concatenations of embeddings for structured prediction tasks, based on a formulation inspired by recent progress on neural architecture search. Specifically, a controller alternately samples a concatenation of embeddings, according to its current belief of the effectiveness of individual embedding types in consideration for a task, and updates the belief based on a reward. We follow strategies in reinforcement learning to optimize the parameters of the controller and compute the reward based on the accuracy of a task model, which is fed with the sampled concatenation as input and trained on a task dataset. Empirical results on 6 tasks and 21 datasets show that our approach outperforms strong baselines and achieves state-of-the-art performance with fine-tuned embeddings in all the evaluations.

Pre-trained language models (PLMs) have achieved great success in natural language processing. Most of PLMs follow the default setting of architecture hyper-parameters (e.g., the hidden dimension is a quarter of the intermediate dimension in feed-forward sub-networks) in BERT. Few studies have been conducted to explore the design of architecture hyper-parameters in BERT, especially for the more efficient PLMs with tiny sizes, which are essential for practical deployment on resource-constrained devices. In this paper, we adopt the one-shot Neural Architecture Search (NAS) to automatically search architecture hyper-parameters. Specifically, we carefully design the techniques of one-shot learning and the search space to provide an adaptive and efficient development way of tiny PLMs for various latency constraints. We name our method AutoTinyBERT and evaluate its effectiveness on the GLUE and SQuAD benchmarks. The extensive experiments show that our method outperforms both the SOTA search-based baseline (NAS-BERT) and the SOTA distillation-based methods (such as DistilBERT, TinyBERT, MiniLM, and MobileBERT). In addition, based on the obtained architectures, we propose a more efficient development method that is even faster than the development of a single PLM. The source code and models will be publicly available upon publication.

Natural Language Generation (NLG) is a key component in a task-oriented dialogue system, which converts the structured meaning representation (MR) to the natural language. For large-scale conversational systems, where it is common to have over hundreds of intents and thousands of slots, neither template-based approaches nor model-based approaches are scalable. Recently, neural NLGs started leveraging transfer learning and showed promising results in few-shot settings. This paper proposes AugNLG, a novel data augmentation approach that combines a self-trained neural retrieval model with a few-shot learned NLU model, to automatically create MR-to-Text data from open-domain texts. The proposed system mostly outperforms the state-of-the-art methods on the FewshotWOZ data in both BLEU and Slot Error Rate. We further confirm improved results on the FewshotSGD data and provide comprehensive analysis results on key components of our system. Our code and data are available at https://github.com/XinnuoXu/AugNLG.

We present a text representation approach that can combine different views (representations) of the same input through effective data fusion and attention strategies for ranking purposes. We apply our model to the problem of differential diagnosis, which aims to find the most probable diseases that match with clinical descriptions of patients, using data from the Undiagnosed Diseases Network. Our model outperforms several ranking approaches (including a commercially-supported system) by effectively prioritizing and combining representations obtained from traditional and recent text representation techniques. We elaborate on several aspects of our model and shed light on its improved performance.

Shapley Values, a solution to the credit assignment problem in cooperative game theory, are a popular type of explanation in machine learning, having been used to explain the importance of features, embeddings, and even neurons. In NLP, however, leave-one-out and attention-based explanations still predominate. Can we draw a connection between these different methods? We formally prove that — save for the degenerate case — attention weights and leave-one-out values cannot be Shapley Values. Attention flow is a post-processed variant of attention weights obtained by running the max-flow algorithm on the attention graph. Perhaps surprisingly, we prove that attention flows are indeed Shapley Values, at least at the layerwise level. Given the many desirable theoretical qualities of Shapley Values — which has driven their adoption among the ML community — we argue that NLP practitioners should, when possible, adopt attention flow explanations alongside more traditional ones.

Attention mechanisms have achieved substantial improvements in neural machine translation by dynamically selecting relevant inputs for different predictions. However, recent studies have questioned the attention mechanisms’ capability for discovering decisive inputs. In this paper, we propose to calibrate the attention weights by introducing a mask perturbation model that automatically evaluates each input’s contribution to the model outputs. We increase the attention weights assigned to the indispensable tokens, whose removal leads to a dramatic performance decrease. The extensive experiments on the Transformer-based translation have demonstrated the effectiveness of our model. We further find that the calibrated attention weights are more uniform at lower layers to collect multiple information while more concentrated on the specific inputs at higher layers. Detailed analyses also show a great need for calibration in the attention weights with high entropy where the model is unconfident about its decision.

Video Question Answering is a task which requires an AI agent to answer questions grounded in video. This task entails three key challenges: (1) understand the intention of various questions, (2) capturing various elements of the input video (e.g., object, action, causality), and (3) cross-modal grounding between language and vision information. We propose Motion-Appearance Synergistic Networks (MASN), which embed two cross-modal features grounded on motion and appearance information and selectively utilize them depending on the question’s intentions. MASN consists of a motion module, an appearance module, and a motion-appearance fusion module. The motion module computes the action-oriented cross-modal joint representations, while the appearance module focuses on the appearance aspect of the input video. Finally, the motion-appearance fusion module takes each output of the motion module and the appearance module as input, and performs question-guided fusion. As a result, MASN achieves new state-of-the-art performance on the TGIF-QA and MSVD-QA datasets. We also conduct qualitative analysis by visualizing the inference results of MASN.

Accurate assessment of the ability of embedding models to capture idiomaticity may require evaluation at token rather than type level, to account for degrees of idiomaticity and possible ambiguity between literal and idiomatic usages. However, most existing resources with annotation of idiomaticity include ratings only at type level. This paper presents the Noun Compound Type and Token Idiomaticity (NCTTI) dataset, with human annotations for 280 noun compounds in English and 180 in Portuguese at both type and token level. We compiled 8,725 and 5,091 token level annotations for English and Portuguese, respectively, which are strongly correlated with the corresponding scores obtained at type level. The NCTTI dataset is used to explore how vector space models reflect the variability of idiomaticity across sentences. Several experiments using state-of-the-art contextualised models suggest that their representations are not capturing the noun compounds idiomaticity as human annotators. This new multilingual resource also contains suggestions for paraphrases of the noun compounds both at type and token levels, with uses for lexical substitution or disambiguation in context.

Emojis have become ubiquitous in digital communication, due to their visual appeal as well as their ability to vividly convey human emotion, among other factors. This also leads to an increased need for systems and tools to operate on text containing emojis. In this study, we assess this support by considering test sets of tweets with emojis, based on which we perform a series of experiments investigating the ability of prominent NLP and text processing tools to adequately process them. In particular, we consider tokenization, part-of-speech tagging, dependency parsing, as well as sentiment analysis. Our findings show that many systems still have notable shortcomings when operating on text containing emojis.