EMNLP 2020 论文列表

Recent advances in open-domain dialogue systems rely on the success of neural models that are trained on large-scale data. However, collecting large-scale dialogue data is usually time-consuming and labor-intensive. To address this data dilemma, we propose a novel data augmentation method for training open-domain dialogue models by utilizing unpaired data. Specifically, a data-level distillation process is first proposed to construct augmented dialogues where both post and response are retrieved from the unpaired data. A ranking module is employed to filter out low-quality dialogues. Further, a model-level distillation process is employed to distill a teacher model trained on high-quality paired data to augmented dialogue pairs, thereby preventing dialogue models from being affected by the noise in the augmented data. Automatic and manual evaluation indicates that our method can produce high-quality dialogue pairs with diverse contents, and the proposed data-level and model-level dialogue distillation can improve the performance of competitive baselines.

Open-domain dialogue generation suffers from the data insufficiency problem due to the vast size of potential responses. In this paper, we propose to explore potential responses by counterfactual reasoning. Given an observed response, the counterfactual reasoning model automatically infers the outcome of an alternative policy that could have been taken. The resulting counterfactual response synthesized in hindsight is of higher quality than the response synthesized from scratch. Training on the counterfactual responses under the adversarial learning framework helps to explore the high-reward area of the potential response space. An empirical study on the DailyDialog dataset shows that our approach significantly outperforms the HRED model as well as the conventional adversarial learning approaches.

Knowledge selection plays an important role in knowledge-grounded dialogue, which is a challenging task to generate more informative responses by leveraging external knowledge. Recently, latent variable models have been proposed to deal with the diversity of knowledge selection by using both prior and posterior distributions over knowledge and achieve promising performance. However, these models suffer from a huge gap between prior and posterior knowledge selection. Firstly, the prior selection module may not learn to select knowledge properly because of lacking the necessary posterior information. Secondly, latent variable models suffer from the exposure bias that dialogue generation is based on the knowledge selected from the posterior distribution at training but from the prior distribution at inference. Here, we deal with these issues on two aspects: (1) We enhance the prior selection module with the necessary posterior information obtained from the specially designed Posterior Information Prediction Module (PIPM); (2) We propose a Knowledge Distillation Based Training Strategy (KDBTS) to train the decoder with the knowledge selected from the prior distribution, removing the exposure bias of knowledge selection. Experimental results on two knowledge-grounded dialogue datasets show that both PIPM and KDBTS achieve performance improvement over the state-of-the-art latent variable model and their combination shows further improvement.

Recent works have shown that generative data augmentation, where synthetic samples generated from deep generative models complement the training dataset, benefit NLP tasks. In this work, we extend this approach to the task of dialog state tracking for goaloriented dialogs. Due to the inherent hierarchical structure of goal-oriented dialogs over utterances and related annotations, the deep generative model must be capable of capturing the coherence among different hierarchies and types of dialog features. We propose the Variational Hierarchical Dialog Autoencoder (VHDA) for modeling the complete aspects of goal-oriented dialogs, including linguistic features and underlying structured annotations, namely speaker information, dialog acts, and goals. The proposed architecture is designed to model each aspect of goal-oriented dialogs using inter-connected latent variables and learns to generate coherent goal-oriented dialogs from the latent spaces. To overcome training issues that arise from training complex variational models, we propose appropriate training strategies. Experiments on various dialog datasets show that our model improves the downstream dialog trackers’ robustness via generative data augmentation. We also discover additional benefits of our unified approach to modeling goal-oriented dialogs – dialog response generation and user simulation, where our model outperforms previous strong baselines.

In this paper, we propose Minimalist Transfer Learning (MinTL) to simplify the system design process of task-oriented dialogue systems and alleviate the over-dependency on annotated data. MinTL is a simple yet effective transfer learning framework, which allows us to plug-and-play pre-trained seq2seq models, and jointly learn dialogue state tracking and dialogue response generation. Unlike previous approaches, which use a copy mechanism to “carryover” the old dialogue states to the new one, we introduce Levenshtein belief spans (Lev), that allows efficient dialogue state tracking with a minimal generation length. We instantiate our learning framework with two pre-trained backbones: T5 and BART, and evaluate them on MultiWOZ. Extensive experiments demonstrate that: 1) our systems establish new state-of-the-art results on end-to-end response generation, 2) MinTL-based systems are more robust than baseline methods in the low resource setting, and they achieve competitive results with only 20% training data, and 3) Lev greatly improves the inference efficiency.

We study knowledge-grounded dialogue generation with pre-trained language models. To leverage the redundant external knowledge under capacity constraint, we propose equipping response generation defined by a pre-trained language model with a knowledge selection module, and an unsupervised approach to jointly optimizing knowledge selection and response generation with unlabeled dialogues. Empirical results on two benchmarks indicate that our model can significantly outperform state-of-the-art methods in both automatic evaluation and human judgment.

Vision-and-language navigation requires an agent to navigate through a real 3D environment following natural language instructions. Despite significant advances, few previous works are able to fully utilize the strong correspondence between the visual and textual sequences. Meanwhile, due to the lack of intermediate supervision, the agent’s performance at following each part of the instruction cannot be assessed during navigation. In this work, we focus on the granularity of the visual and language sequences as well as the traceability of agents through the completion of an instruction. We provide agents with fine-grained annotations during training and find that they are able to follow the instruction better and have a higher chance of reaching the target at test time. We enrich the benchmark dataset Room-to-Room (R2R) with sub-instructions and their corresponding paths. To make use of this data, we propose effective sub-instruction attention and shifting modules that select and attend to a single sub-instruction at each time-step. We implement our sub-instruction modules in four state-of-the-art agents, compare with their baseline models, and show that our proposed method improves the performance of all four agents. We release the Fine-Grained R2R dataset (FGR2R) and the code at https://github.com/YicongHong/Fine-Grained-R2R.

In this paper, we consider the syntactic properties of languages emerged in referential games, using unsupervised grammar induction (UGI) techniques originally designed to analyse natural language. We show that the considered UGI techniques are appropriate to analyse emergent languages and we then study if the languages that emerge in a typical referential game setup exhibit syntactic structure, and to what extent this depends on the maximum message length and number of symbols that the agents are allowed to use. Our experiments demonstrate that a certain message length and vocabulary size are required for structure to emerge, but they also illustrate that more sophisticated game scenarios are required to obtain syntactic properties more akin to those observed in human language. We argue that UGI techniques should be part of the standard toolkit for analysing emergent languages and release a comprehensive library to facilitate such analysis for future researchers.

Visual dialog is a challenging vision-language task, where a dialog agent needs to answer a series of questions through reasoning on the image content and dialog history. Prior work has mostly focused on various attention mechanisms to model such intricate interactions. By contrast, in this work, we propose VD-BERT, a simple yet effective framework of unified vision-dialog Transformer that leverages the pretrained BERT language models for Visual Dialog tasks. The model is unified in that (1) it captures all the interactions between the image and the multi-turn dialog using a single-stream Transformer encoder, and (2) it supports both answer ranking and answer generation seamlessly through the same architecture. More crucially, we adapt BERT for the effective fusion of vision and dialog contents via visually grounded training. Without the need of pretraining on external vision-language data, our model yields new state of the art, achieving the top position in both single-model and ensemble settings (74.54 and 75.35 NDCG scores) on the visual dialog leaderboard. Our code and pretrained models are released at https://github.com/salesforce/VD-BERT.

Social media produces large amounts of contents every day. To help users quickly capture what they need, keyphrase prediction is receiving a growing attention. Nevertheless, most prior efforts focus on text modeling, largely ignoring the rich features embedded in the matching images. In this work, we explore the joint effects of texts and images in predicting the keyphrases for a multimedia post. To better align social media style texts and images, we propose: (1) a novel Multi-Modality MultiHead Attention (M3H-Att) to capture the intricate cross-media interactions; (2) image wordings, in forms of optical characters and image attributes, to bridge the two modalities. Moreover, we design a unified framework to leverage the outputs of keyphrase classification and generation and couple their advantages. Extensive experiments on a large-scale dataset newly collected from Twitter show that our model significantly outperforms the previous state of the art based on traditional attention mechanisms. Further analyses show that our multi-head attention is able to attend information from various aspects and boost classification or generation in diverse scenarios.

In real-world dialogue, first-person visual information about where the other speakers are and what they are paying attention to is crucial to understand their intentions. Non-verbal responses also play an important role in social interactions. In this paper, we propose a visually-grounded first-person dialogue (VFD) dataset with verbal and non-verbal responses. The VFD dataset provides manually annotated (1) first-person images of agents, (2) utterances of human speakers, (3) eye-gaze locations of the speakers, and (4) the agents’ verbal and non-verbal responses. We present experimental results obtained using the proposed VFD dataset and recent neural network models (e.g., BERT, ResNet). The results demonstrate that first-person vision helps neural network models correctly understand human intentions, and the production of non-verbal responses is a challenging task like that of verbal responses. Our dataset is publicly available.

Physical common sense plays an essential role in the cognition abilities of robots for human-robot interaction. Machine learning methods have shown promising results on physical commonsense learning in natural language processing but still suffer from model generalization. In this paper, we formulate physical commonsense learning as a knowledge graph completion problem to better use the latent relationships among training samples. Compared with completing general knowledge graphs, completing a physical commonsense knowledge graph has three unique characteristics: training data are scarce, not all facts can be mined from existing texts, and the number of relationships is small. To deal with these problems, we first use a pre-training language model BERT to augment training data, and then employ constrained tucker factorization to model complex relationships by constraining types and adding negative relationships. We compare our method with existing state-of-the-art knowledge graph embedding methods and show its superior performance.

In the task of Visual Question Answering (VQA), most state-of-the-art models tend to learn spurious correlations in the training set and achieve poor performance in out-of-distribution test data. Some methods of generating counterfactual samples have been proposed to alleviate this problem. However, the counterfactual samples generated by most previous methods are simply added to the training data for augmentation and are not fully utilized. Therefore, we introduce a novel self-supervised contrastive learning mechanism to learn the relationship between original samples, factual samples and counterfactual samples. With the better cross-modal joint embeddings learned from the auxiliary training objective, the reasoning capability and robustness of the VQA model are boosted significantly. We evaluate the effectiveness of our method by surpassing current state-of-the-art models on the VQA-CP dataset, a diagnostic benchmark for assessing the VQA model’s robustness.

Chart Question Answering (CQA) is the task of answering natural language questions about visualisations in the chart image. Recent solutions, inspired by VQA approaches, rely on image-based attention for question/answering while ignoring the inherent chart structure. We propose STL-CQA which improves the question/answering through sequential elements localization, question encoding and then, a structural transformer-based learning approach. We conduct extensive experiments while proposing pre-training tasks, methodology and also an improved dataset with more complex and balanced questions of different types. The proposed methodology shows a significant accuracy improvement compared to the state-of-the-art approaches on various chart Q/A datasets, while outperforming even human baseline on the DVQA Dataset. We also demonstrate interpretability while examining different components in the inference pipeline.

Recent developments in machine learning have introduced models that approach human performance at the cost of increased architectural complexity. Efforts to make the rationales behind the models’ predictions transparent have inspired an abundance of new explainability techniques. Provided with an already trained model, they compute saliency scores for the words of an input instance. However, there exists no definitive guide on (i) how to choose such a technique given a particular application task and model architecture, and (ii) the benefits and drawbacks of using each such technique. In this paper, we develop a comprehensive list of diagnostic properties for evaluating existing explainability techniques. We then employ the proposed list to compare a set of diverse explainability techniques on downstream text classification tasks and neural network architectures. We also compare the saliency scores assigned by the explainability techniques with human annotations of salient input regions to find relations between a model’s performance and the agreement of its rationales with human ones. Overall, we find that the gradient-based explanations perform best across tasks and model architectures, and we present further insights into the properties of the reviewed explainability techniques.

Attribution methods assess the contribution of inputs to the model prediction. One way to do so is erasure: a subset of inputs is considered irrelevant if it can be removed without affecting the prediction. Though conceptually simple, erasure’s objective is intractable and approximate search remains expensive with modern deep NLP models. Erasure is also susceptible to the hindsight bias: the fact that an input can be dropped does not mean that the model ‘knows’ it can be dropped. The resulting pruning is over-aggressive and does not reflect how the model arrives at the prediction. To deal with these challenges, we introduce Differentiable Masking. DiffMask learns to mask-out subsets of the input while maintaining differentiability. The decision to include or disregard an input token is made with a simple model based on intermediate hidden layers of the analyzed model. First, this makes the approach efficient because we predict rather than search. Second, as with probing classifiers, this reveals what the network ‘knows’ at the corresponding layers. This lets us not only plot attribution heatmaps but also analyze how decisions are formed across network layers. We use DiffMask to study BERT models on sentiment classification and question answering.

BERT and its variants have achieved state-of-the-art performance in various NLP tasks. Since then, various works have been proposed to analyze the linguistic information being captured in BERT. However, the current works do not provide an insight into how BERT is able to achieve near human-level performance on the task of Reading Comprehension based Question Answering. In this work, we attempt to interpret BERT for RCQA. Since BERT layers do not have predefined roles, we define a layer’s role or functionality using Integrated Gradients. Based on the defined roles, we perform a preliminary analysis across all layers. We observed that the initial layers focus on query-passage interaction, whereas later layers focus more on contextual understanding and enhancing the answer prediction. Specifically for quantifier questions (how much/how many), we notice that BERT focuses on confusing words (i.e., on other numerical quantities in the passage) in the later layers, but still manages to predict the answer correctly. The fine-tuning and analysis scripts will be publicly available at https://github.com/iitmnlp/BERT-Analysis-RCQA.

Given the success of Transformer-based models, two directions of study have emerged: interpreting role of individual attention heads and down-sizing the models for efficiency. Our work straddles these two streams: We analyse the importance of basing pruning strategies on the interpreted role of the attention heads. We evaluate this on Transformer and BERT models on multiple NLP tasks. Firstly, we find that a large fraction of the attention heads can be randomly pruned with limited effect on accuracy. Secondly, for Transformers, we find no advantage in pruning attention heads identified to be important based on existing studies that relate importance to the location of a head. On the BERT model too we find no preference for top or bottom layers, though the latter are reported to have higher importance. However, strategies that avoid pruning middle layers and consecutive layers perform better. Finally, during fine-tuning the compensation for pruned attention heads is roughly equally distributed across the un-pruned heads. Our results thus suggest that interpretation of attention heads does not strongly inform pruning.

Large Transformer-based models were shown to be reducible to a smaller number of self-attention heads and layers. We consider this phenomenon from the perspective of the lottery ticket hypothesis, using both structured and magnitude pruning. For fine-tuned BERT, we show that (a) it is possible to find subnetworks achieving performance that is comparable with that of the full model, and (b) similarly-sized subnetworks sampled from the rest of the model perform worse. Strikingly, with structured pruning even the worst possible subnetworks remain highly trainable, indicating that most pre-trained BERT weights are potentially useful. We also study the “good” subnetworks to see if their success can be attributed to superior linguistic knowledge, but find them unstable, and not explained by meaningful self-attention patterns.