EMNLP 2021 论文列表

Interactive machine reading comprehension (iMRC) is machine comprehension tasks where knowledge sources are partially observable. An agent must interact with an environment sequentially to gather necessary knowledge in order to answer a question. We hypothesize that graph representations are good inductive biases, which can serve as an agent’s memory mechanism in iMRC tasks. We explore four different categories of graphs that can capture text information at various levels. We describe methods that dynamically build and update these graphs during information gathering, as well as neural models to encode graph representations in RL agents. Extensive experiments on iSQuAD suggest that graph representations can result in significant performance improvements for RL agents.

Recent years has witnessed the remarkable success in end-to-end task-oriented dialog system, especially when incorporating external knowledge information. However, the quality of most existing models’ generated response is still limited, mainly due to their lack of fine-grained reasoning on deterministic knowledge (w.r.t. conceptual tokens), which makes them difficult to capture the concept shifts and identify user’s real intention in cross-task scenarios. To address these issues, we propose a novel intention mechanism to better model deterministic entity knowledge. Based on such a mechanism, we further propose an intention reasoning network (IR-Net), which consists of joint and multi-hop reasoning, to obtain intention-aware representations of conceptual tokens that can be used to capture the concept shifts involved in task-oriented conversations, so as to effectively identify user’s intention and generate more accurate responses. Experimental results verify the effectiveness of IR-Net, showing that it achieves the state-of-the-art performance on two representative multi-domain dialog datasets.

While much research has been done in text-to-image synthesis, little work has been done to explore the usage of linguistic structure of the input text. Such information is even more important for story visualization since its inputs have an explicit narrative structure that needs to be translated into an image sequence (or visual story). Prior work in this domain has shown that there is ample room for improvement in the generated image sequence in terms of visual quality, consistency and relevance. In this paper, we first explore the use of constituency parse trees using a Transformer-based recurrent architecture for encoding structured input. Second, we augment the structured input with commonsense information and study the impact of this external knowledge on the generation of visual story. Third, we also incorporate visual structure via bounding boxes and dense captioning to provide feedback about the characters/objects in generated images within a dual learning setup. We show that off-the-shelf dense-captioning models trained on Visual Genome can improve the spatial structure of images from a different target domain without needing fine-tuning. We train the model end-to-end using intra-story contrastive loss (between words and image sub-regions) and show significant improvements in visual quality. Finally, we provide an analysis of the linguistic and visuo-spatial information.

Recently graph-based methods have been adopted for Abstractive Text Summarization. However, existing graph-based methods only consider either word relations or structure information, which neglect the correlation between them. To simultaneously capture the word relations and structure information from sentences, we propose a novel Dual Graph network for Abstractive Sentence Summarization. Specifically, we first construct semantic scenario graph and semantic word relation graph based on FrameNet, and subsequently learn their representations and design graph fusion method to enhance their correlation and obtain better semantic representation for summary generation. Experimental results show our model outperforms existing state-of-the-art methods on two popular benchmark datasets, i.e., Gigaword and DUC 2004.

Neural Word Sense Disambiguation (WSD) has recently been shown to benefit from the incorporation of pre-existing knowledge, such as that coming from the WordNet graph. However, state-of-the-art approaches have been successful in exploiting only the local structure of the graph, with only close neighbors of a given synset influencing the prediction. In this work, we improve a classification model by recomputing logits as a function of both the vanilla independently produced logits and the global WordNet graph. We achieve this by incorporating an online neural approximated PageRank, which enables us to refine edge weights as well. This method exploits the global graph structure while keeping space requirements linear in the number of edges. We obtain strong improvements, matching the current state of the art. Code is available at https://github.com/SapienzaNLP/neural-pagerank-wsd

Scripts are structured sequences of events together with the participants, which are extracted from the texts. Script event prediction aims to predict the subsequent event given the historical events in the script. Two kinds of information facilitate this task, namely, the event-level information and the script-level information. At the event level, existing studies view an event as a verb with its participants, while neglecting other useful properties, such as the state of the participants. At the script level, most existing studies only consider a single event sequence corresponding to one common protagonist. In this paper, we propose a Transformer-based model, called MCPredictor, which integrates deep event-level and script-level information for script event prediction. At the event level, MCPredictor utilizes the rich information in the text to obtain more comprehensive event semantic representations. At the script-level, it considers multiple event sequences corresponding to different participants of the subsequent event. The experimental results on the widely-used New York Times corpus demonstrate the effectiveness and superiority of the proposed model.

In order to alleviate the huge demand for annotated datasets for different tasks, many recent natural language processing datasets have adopted automated pipelines for fast-tracking usable data. However, model training with such datasets poses a challenge because popular optimization objectives are not robust to label noise induced in the annotation generation process. Several noise-robust losses have been proposed and evaluated on tasks in computer vision, but they generally use a single dataset-wise hyperparamter to control the strength of noise resistance. This work proposes novel instance-adaptive training frameworks to change single dataset-wise hyperparameters of noise resistance in such losses to be instance-wise. Such instance-wise noise resistance hyperparameters are predicted by special instance-level label quality predictors, which are trained along with the main classification models. Experiments on noisy and corrupted NLP datasets show that proposed instance-adaptive training frameworks help increase the noise-robustness provided by such losses, promoting the use of the frameworks and associated losses in NLP models trained with noisy data.

Recent successes in deep generative modeling have led to significant advances in natural language generation (NLG). Incorporating entities into neural generation models has demonstrated great improvements by assisting to infer the summary topic and to generate coherent content. To enhance the role of entity in NLG, in this paper, we aim to model the entity type in the decoding phase to generate contextual words accurately. We develop a novel NLG model to produce a target sequence based on a given list of entities. Our model has a multi-step decoder that injects the entity types into the process of entity mention generation. Experiments on two public news datasets demonstrate type injection performs better than existing type embedding concatenation baselines.

Deep-learning models for language generation tasks tend to produce repetitive output. Various methods have been proposed to encourage lexical diversity during decoding, but this often comes at a cost to the perceived fluency and adequacy of the output. In this work, we propose to ameliorate this cost by using an Imitation Learning approach to explore the level of diversity that a language generation model can reliably produce. Specifically, we augment the decoding process with a meta-classifier trained to distinguish which words at any given timestep will lead to high-quality output. We focus our experiments on concept-to-text generation where models are sensitive to the inclusion of irrelevant words due to the strict relation between input and output. Our analysis shows that previous methods for diversity underperform in this setting, while human evaluation suggests that our proposed method achieves a high level of diversity with minimal effect on the output’s fluency and adequacy.

Among the most critical limitations of deep learning NLP models are their lack of interpretability, and their reliance on spurious correlations. Prior work proposed various approaches to interpreting the black-box models to unveil the spurious correlations, but the research was primarily used in human-computer interaction scenarios. It still remains underexplored whether or how such model interpretations can be used to automatically “unlearn” confounding features. In this work, we propose influence tuning—a procedure that leverages model interpretations to update the model parameters towards a plausible interpretation (rather than an interpretation that relies on spurious patterns in the data) in addition to learning to predict the task labels. We show that in a controlled setup, influence tuning can help deconfounding the model from spurious patterns in data, significantly outperforming baseline methods that use adversarial training.

By exploiting the cross-modal attention, cross-BERT methods have achieved state-of-the-art accuracy in cross-modal retrieval. Nevertheless, the heavy text-image interactions in the cross-BERT model are prohibitively slow for large-scale retrieval. Late-interaction methods trade off retrieval accuracy and efficiency by exploiting cross-modal interaction only in the late stage, attaining a satisfactory retrieval speed. In this work, we propose an inflating and shrinking approach to further boost the efficiency and accuracy of late-interaction methods. The inflating operation plugs several codes in the input of the encoder to exploit the text-image interactions more thoroughly for higher retrieval accuracy. Then the shrinking operation gradually reduces the text-image interactions through knowledge distilling for higher efficiency. Through an inflating operation followed by a shrinking operation, both efficiency and accuracy of a late-interaction model are boosted. Systematic experiments on public benchmarks demonstrate the effectiveness of our inflating and shrinking approach.

Systematic compositionality is an essential mechanism in human language, allowing the recombination of known parts to create novel expressions. However, existing neural models have been shown to lack this basic ability in learning symbolic structures. Motivated by the failure of a Transformer model on the SCAN compositionality challenge (Lake and Baroni, 2018), which requires parsing a command into actions, we propose two auxiliary sequence prediction tasks as additional training supervision. These automatically-generated sequences are more representative of the underlying compositional symbolic structures of the input data. During inference, the model jointly predicts the next action and the next tokens in the auxiliary sequences at each step. Experiments on the SCAN dataset show that our method encourages the Transformer to understand compositional structures of the command, improving its accuracy on multiple challenging splits from ≤ 10% to 100%. With only 418 (5%) training instances, our approach still achieves 97.8% accuracy on the MCD1 split. Therefore, we argue that compositionality can be induced in Transformers given minimal but proper guidance. We also show that a better result is achieved using less contextualized vectors as the attention’s query, providing insights into architecture choices in achieving systematic compositionality. Finally, we show positive generalization results on the grounded-SCAN task (Ruis et al., 2020).

Stereotypical character roles-also known as archetypes or dramatis personae-play an important function in narratives: they facilitate efficient communication with bundles of default characteristics and associations and ease understanding of those characters’ roles in the overall narrative. We present a fully unsupervised k-means clustering approach for learning stereotypical roles given only structural plot information. We demonstrate the technique on Vladimir Propp’s structural theory of Russian folktales (captured in the extended ProppLearner corpus, with 46 tales), showing that our approach can induce six out of seven of Propp’s dramatis personae with F1 measures of up to 0.70 (0.58 average), with an additional category for minor characters. We have explored various feature sets and variations of a cluster evaluation method. The best-performing feature set comprises plot functions, unigrams, tf-idf weights, and embeddings over coreference chain heads. Roles that are mentioned more often (Hero, Villain), or have clearly distinct plot patterns (Princess) are more strongly differentiated than less frequent or distinct roles (Dispatcher, Helper, Donor). Detailed error analysis suggests that the quality of the coreference chain and plot functions annotations are critical for this task. We provide all our data and code for reproducibility.

We present IndoNLI, the first human-elicited NLI dataset for Indonesian. We adapt the data collection protocol for MNLI and collect ~18K sentence pairs annotated by crowd workers and experts. The expert-annotated data is used exclusively as a test set. It is designed to provide a challenging test-bed for Indonesian NLI by explicitly incorporating various linguistic phenomena such as numerical reasoning, structural changes, idioms, or temporal and spatial reasoning. Experiment results show that XLM-R outperforms other pre-trained models in our data. The best performance on the expert-annotated data is still far below human performance (13.4% accuracy gap), suggesting that this test set is especially challenging. Furthermore, our analysis shows that our expert-annotated data is more diverse and contains fewer annotation artifacts than the crowd-annotated data. We hope this dataset can help accelerate progress in Indonesian NLP research.

Natural language generation (NLG) benchmarks provide an important avenue to measure progress and develop better NLG systems. Unfortunately, the lack of publicly available NLG benchmarks for low-resource languages poses a challenging barrier for building NLG systems that work well for languages with limited amounts of data. Here we introduce IndoNLG, the first benchmark to measure natural language generation (NLG) progress in three low-resource—yet widely spoken—languages of Indonesia: Indonesian, Javanese, and Sundanese. Altogether, these languages are spoken by more than 100 million native speakers, and hence constitute an important use case of NLG systems today. Concretely, IndoNLG covers six tasks: summarization, question answering, chit-chat, and three different pairs of machine translation (MT) tasks. We collate a clean pretraining corpus of Indonesian, Sundanese, and Javanese datasets, Indo4B-Plus, which is used to pretrain our models: IndoBART and IndoGPT. We show that IndoBART and IndoGPT achieve competitive performance on all tasks—despite using only one-fifth the parameters of a larger multilingual model, mBART-large (Liu et al., 2020). This finding emphasizes the importance of pretraining on closely related, localized languages to achieve more efficient learning and faster inference at very low-resource languages like Javanese and Sundanese.

We present IndoBERTweet, the first large-scale pretrained model for Indonesian Twitter that is trained by extending a monolingually-trained Indonesian BERT model with additive domain-specific vocabulary. We focus in particular on efficient model adaptation under vocabulary mismatch, and benchmark different ways of initializing the BERT embedding layer for new word types. We find that initializing with the average BERT subword embedding makes pretraining five times faster, and is more effective than proposed methods for vocabulary adaptation in terms of extrinsic evaluation over seven Twitter-based datasets.

In recent years pre-trained language models (PLM) such as BERT have proven to be very effective in diverse NLP tasks such as Information Extraction, Sentiment Analysis and Question Answering. Trained with massive general-domain text, these pre-trained language models capture rich syntactic, semantic and discourse information in the text. However, due to the differences between general and specific domain text (e.g., Wikipedia versus clinic notes), these models may not be ideal for domain-specific tasks (e.g., extracting clinical relations). Furthermore, it may require additional medical knowledge to understand clinical text properly. To solve these issues, in this research, we conduct a comprehensive examination of different techniques to add medical knowledge into a pre-trained BERT model for clinical relation extraction. Our best model outperforms the state-of-the-art systems on the benchmark i2b2/VA 2010 clinical relation extraction dataset.

Transformer architecture has become ubiquitous in the natural language processing field. To interpret the Transformer-based models, their attention patterns have been extensively analyzed. However, the Transformer architecture is not only composed of the multi-head attention; other components can also contribute to Transformers’ progressive performance. In this study, we extended the scope of the analysis of Transformers from solely the attention patterns to the whole attention block, i.e., multi-head attention, residual connection, and layer normalization. Our analysis of Transformer-based masked language models shows that the token-to-token interaction performed via attention has less impact on the intermediate representations than previously assumed. These results provide new intuitive explanations of existing reports; for example, discarding the learned attention patterns tends not to adversely affect the performance. The codes of our experiments are publicly available.

The narrative event prediction aims to predict what happens after a sequence of events, which is essential to modeling sophisticated real-world events. Existing studies focus on mining the inter-events relationships while ignoring how the events happened, which we called circumstances. With our observation, the event circumstances indicate what will happen next. To incorporate event circumstances into the narrative event prediction, we propose the CircEvent, which adopts the two multi-head attention to retrieve circumstances at the local and global levels. We also introduce a regularization of attention weights to leverage the alignment between events and local circumstances. The experimental results demonstrate our CircEvent outperforms existing baselines by 12.2%. The further analysis demonstrates the effectiveness of our multi-head attention modules and regularization.