NAACL 2022 论文列表

Text classification has achieved great success with the prosperity of deep learning and pre-trained language models. However, we often encounter labeled data deficiency problems in real-world text-classification tasks. To overcome such challenging scenarios, interest in few-shot learning has increased, whereas most few-shot text classification studies suffer from a difficulty of utilizing pre-trained language models. In the study, we propose a novel learning method for learning how to attend, called LEA, through which meta-level attention aspects are derived based on our meta-learning strategy. This enables the generation of task-specific document embedding with leveraging pre-trained language models even though a few labeled data instances are given. We evaluate our proposed learning method on five benchmark datasets. The results show that the novel method robustly provides the competitive performance compared to recent few-shot learning methods for all the datasets.

The development of over-parameterized pre-trained language models has made a significant contribution toward the success of natural language processing. While over-parameterization of these models is the key to their generalization power, it makes them unsuitable for deployment on low-capacity devices. We push the limits of state-of-the-art Transformer-based pre-trained language model compression using Kronecker decomposition. We present our KroneckerBERT, a compressed version of the BERT_BASE model obtained by compressing the embedding layer and the linear mappings in the multi-head attention, and the feed-forward network modules in the Transformer layers. Our KroneckerBERT is trained via a very efficient two-stage knowledge distillation scheme using far fewer data samples than state-of-the-art models like MobileBERT and TinyBERT. We evaluate the performance of KroneckerBERT on well-known NLP benchmarks. We show that our KroneckerBERT with compression factors of 7.7x and 21x outperforms state-of-the-art compression methods on the GLUE and SQuAD benchmarks. In particular, using only 13% of the teacher model parameters, it retain more than 99% of the accuracy on the majority of GLUE tasks.

Knowledge-grounded dialogue systems are challenging to build due to the lack of training data and heterogeneous knowledge sources. Existing systems perform poorly on unseen topics due to limited topics covered in the training data. In addition, it is challenging to generalize to the domains that require different types of knowledge sources. To address the above challenges, we present PLUG, a language model that homogenizes different knowledge sources to a unified knowledge representation for knowledge-grounded dialogue generation tasks. We first retrieve relevant information from heterogeneous knowledge sources (e.g., wiki, dictionary, or knowledge graph); Then the retrieved knowledge is transformed into text and concatenated with dialogue history to feed into the language model for generating responses. PLUG is pre-trained on a large-scale knowledge-grounded dialogue corpus. The empirical evaluation on two benchmarks shows that PLUG generalizes well across different knowledge-grounded dialogue tasks. It achieves comparable performance with state-of-the-art methods in the fully-supervised setting and significantly outperforms other approaches in zero-shot and few-shot settings.

During their pre-flight briefings, aircraft pilots must analyse a long list of NoTAMs (NOtice To AirMen) indicating potential hazards along the flight route, sometimes up to pages for long-haul flights. NOTAM free-text fields typically have a very special phrasing, with lots of acronyms and domain-specific vocabulary, which makes it differ significantly from standard English. In this paper, we pretrain language models derived from BERT on circa 1 million unlabeled NOTAMs and reuse the learnt representations on three downstream tasks valuable for pilots: criticality prediction, named entity recognition and translation into a structured language called Airlang. This self-supervised approach, where smaller amounts of labeled data are enough for task-specific fine-tuning, is well suited in the aeronautical context since expert annotations are expensive and time-consuming. We present evaluation scores across the tasks showing a high potential for an operational usability of such models (by pilots, airlines or service providers), which is a first to the best of our knowledge.

Recent explorations of large-scale pre-trained language models (PLMs) have revealed the power of PLMs with huge amounts of parameters, setting off a wave of training ever-larger PLMs. However, it requires tremendous computational resources to train a large-scale PLM, which may be practically unaffordable. In addition, existing large-scale PLMs are mainly trained from scratch individually, ignoring that many well-trained PLMs are available. To this end, we explore the question how could existing PLMs benefit training large-scale PLMs in future. Specifically, we introduce a pre-training framework named “knowledge inheritance” (KI) and explore how could knowledge distillation serve as auxiliary supervision during pre-training to efficiently learn larger PLMs. Experimental results demonstrate the superiority of KI in training efficiency. We also conduct empirical analyses to explore the effects of teacher PLMs’ pre-training settings, including model architecture, pre-training data, etc. Finally, we show that KI could be applied to domain adaptation and knowledge transfer.

This paper presents an effort within our company of developing knowledge extraction pipeline for English, which can be further used for constructing an entreprise-specific knowledge base. We present a system consisting of entity detection and linking, coreference resolution, and relation extraction based on the Wikidata schema. We highlight existing challenges of knowledge extraction by evaluating the deployed pipeline on real-world data. We also make available a database, which can serve as a new resource for sentential relation extraction, and we underline the importance of having balanced data for training classification models.

Existing studies in dialogue system research mostly treat task-oriented dialogue and chit-chat as separate domains. Towards building a human-like assistant that can converse naturally and seamlessly with users, it is important to build a dialogue system that conducts both types of conversations effectively. In this work, we investigate how task-oriented dialogue and knowledge-grounded chit-chat can be effectively integrated into a single model. To this end, we create a new dataset, KETOD (Knowledge-Enriched Task-Oriented Dialogue), where we naturally enrich task-oriented dialogues with chit-chat based on relevant entity knowledge. We also propose two new models, SimpleToDPlus and Combiner, for the proposed task. Experimental results on both automatic and human evaluations show that the proposed methods can significantly improve the performance in knowledge-enriched response generation while maintaining a competitive task-oriented dialog performance. We believe our new dataset will be a valuable resource for future studies. Our dataset and code are publicly available at https://github.com/facebookresearch/ketod.

Self-supervised vision-and-language pretraining (VLP) aims to learn transferable multi-modal representations from large-scale image-text data and to achieve strong performances on a broad scope of vision-language tasks after finetuning. Previous mainstream VLP approaches typically adopt a two-step strategy relying on external object detectors to encode images in a multi-modal Transformer framework, which suffer from restrictive object concept space, limited image context and inefficient computation. In this paper, we propose an object-aware end-to-end VLP framework, which directly feeds image grid features from CNNs into the Transformer and learns the multi-modal representations jointly. More importantly, we propose to perform object knowledge distillation to facilitate learning cross-modal alignment at different semantic levels. To achieve that, we design two novel pretext tasks by taking object features and their semantic labels from external detectors as supervision: 1.) Object-guided masked vision modeling task focuses on enforcing object-aware representation learning in the multi-modal Transformer; 2.) Phrase-region alignment task aims to improve cross-modal alignment by utilizing the similarities between noun phrases and object labels in the linguistic space. Extensive experiments on a wide range of vision-language tasks demonstrate the efficacy of our proposed framework, and we achieve competitive or superior performances over the existing pretraining strategies.

Political perspective detection has become an increasingly important task that can help combat echo chambers and political polarization. Previous approaches generally focus on leveraging textual content to identify stances, while they fail to reason with background knowledge or leverage the rich semantic and syntactic textual labels in news articles. In light of these limitations, we propose KCD, a political perspective detection approach to enable multi-hop knowledge reasoning and incorporate textual cues as paragraph-level labels. Specifically, we firstly generate random walks on external knowledge graphs and infuse them with news text representations. We then construct a heterogeneous information network to jointly model news content as well as semantic, syntactic and entity cues in news articles. Finally, we adopt relational graph neural networks for graph-level representation learning and conduct political perspective detection. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods on two benchmark datasets. We further examine the effect of knowledge walks and textual cues and how they contribute to our approach’s data efficiency.

The primary focus of recent work with large-scale transformers has been on optimizing the amount of information packed into the model’s parameters. In this work, we ask a complementary question: Can multimodal transformers leverage explicit knowledge in their reasoning? Existing, primarily unimodal, methods have explored approaches under the paradigm of knowledge retrieval followed by answer prediction, but leave open questions about the quality and relevance of the retrieved knowledge used, and how the reasoning processes over implicit and explicit knowledge should be integrated. To address these challenges, we propose a - Knowledge Augmented Transformer (KAT) - which achieves a strong state-of-the-art result (+6% absolute) on the open-domain multimodal task of OK-VQA. Our approach integrates implicit and explicit knowledge in an encoder-decoder architecture, while still jointly reasoning over both knowledge sources during answer generation. Additionally, explicit knowledge integration improves interpretability of model predictions in our analysis.

Pre-trained language models (PLMs) have achieved remarkable success on various natural language understanding tasks. Simple fine-tuning of PLMs, on the other hand, might be suboptimal for domain-specific tasks because they cannot possibly cover knowledge from all domains. While adaptive pre-training of PLMs can help them obtain domain-specific knowledge, it requires a large training cost. Moreover, adaptive pre-training can harm the PLM’s performance on the downstream task by causing catastrophic forgetting of its general knowledge. To overcome such limitations of adaptive pre-training for PLM adaption, we propose a novel domain adaption framework for PLMs coined as Knowledge-Augmented Language model Adaptation (KALA), which modulates the intermediate hidden representations of PLMs with domain knowledge, consisting of entities and their relational facts. We validate the performance of our KALA on question answering and named entity recognition tasks on multiple datasets across various domains. The results show that, despite being computationally efficient, our KALA largely outperforms adaptive pre-training.

Abstractive summarization can generate high quality results with the development of the neural network. However, generating factual consistency summaries is a challenging task for abstractive summarization. Recent studies extract the additional information with off-the-shelf tools from the source document as a clue to guide the summary generation, which shows effectiveness to improve the faithfulness. Unlike these work, we present a novel framework based on conditional variational autoencoders, which induces the guidance information and generates the summary equipment with the guidance synchronously. Experiments on XSUM and CNNDM dataset show that our approach can generate relevant and fluent summaries which is more faithful than the existing state-of-the-art approaches, according to multiple factual consistency metrics.

Existing KG-augmented models for commonsense question answering primarily focus on designing elaborate Graph Neural Networks (GNNs) to model knowledge graphs (KGs). However, they ignore (i) the effectively fusing and reasoning over question context representations and the KG representations, and (ii) automatically selecting relevant nodes from the noisy KGs during reasoning. In this paper, we propose a novel model, JointLK, which solves the above limitations through the joint reasoning of LM and GNN and the dynamic KGs pruning mechanism. Specifically, JointLK performs joint reasoning between LM and GNN through a novel dense bidirectional attention module, in which each question token attends on KG nodes and each KG node attends on question tokens, and the two modal representations fuse and update mutually by multi-step interactions. Then, the dynamic pruning module uses the attention weights generated by joint reasoning to prune irrelevant KG nodes recursively. We evaluate JointLK on the CommonsenseQA and OpenBookQA datasets, and demonstrate its improvements to the existing LM and LM+KG models, as well as its capability to perform interpretable reasoning.

Previous studies on the timeline summarization (TLS) task ignored the information interaction between sentences and dates, and adopted pre-defined unlearnable representations for them. They also considered date selection and event detection as two independent tasks, which makes it impossible to integrate their advantages and obtain a globally optimal summary. In this paper, we present a joint learning-based heterogeneous graph attention network for TLS (HeterTls), in which date selection and event detection are combined into a unified framework to improve the extraction accuracy and remove redundant sentences simultaneously. Our heterogeneous graph involves multiple types of nodes, the representations of which are iteratively learned across the heterogeneous graph attention layer. We evaluated our model on four datasets, and found that it significantly outperformed the current state-of-the-art baselines with regard to ROUGE scores and date selection metrics.

Event trigger detection, entity mention recognition, event argument extraction, and relation extraction are the four important tasks in information extraction that have been performed jointly (Joint Information Extraction - JointIE) to avoid error propagation and leverage dependencies between the task instances (i.e., event triggers, entity mentions, relations, and event arguments). However, previous JointIE models often assume heuristic manually-designed dependency between the task instances and mean-field factorization for the joint distribution of instance labels, thus unable to capture optimal dependencies among instances and labels to improve representation learning and IE performance. To overcome these limitations, we propose to induce a dependency graph among task instances from data to boost representation learning. To better capture dependencies between instance labels, we propose to directly estimate their joint distribution via Conditional Random Fields. Noise Contrastive Estimation is introduced to address the maximization of the intractable joint likelihood for model training. Finally, to improve the decoding with greedy or beam search in prior work, we present Simulated Annealing to better find the globally optimal assignment for instance labels at decoding time. Experimental results show that our proposed model outperforms previous models on multiple IE tasks across 5 datasets and 2 languages.

The softmax layer in neural machine translation is designed to model the distribution over mutually exclusive tokens. Machine translation, however, is intrinsically uncertain: the same source sentence can have multiple semantically equivalent translations. Therefore, we propose to replace the softmax activation with a multi-label classification layer that can model ambiguity more effectively. We call our loss function Single-label Contrastive Objective for Non-Exclusive Sequences (SCONES). We show that the multi-label output layer can still be trained on single reference training data using the SCONES loss function. SCONES yields consistent BLEU score gains across six translation directions, particularly for medium-resource language pairs and small beam sizes. By using smaller beam sizes we can speed up inference by a factor of 3.9x and still match or improve the BLEU score obtained using softmax. Furthermore, we demonstrate that SCONES can be used to train NMT models that assign the highest probability to adequate translations, thus mitigating the “beam search curse”. Additional experiments on synthetic language pairs with varying levels of uncertainty suggest that the improvements from SCONES can be attributed to better handling of ambiguity.

Recursive noun phrases (NPs) have interesting semantic properties. For example, “my favorite new movie” is not necessarily my favorite movie, whereas “my new favorite movie” is. This is common sense to humans, yet it is unknown whether language models have such knowledge. We introduce the Recursive Noun Phrase Challenge (RNPC), a dataset of three textual inference tasks involving textual entailment and event plausibility comparison, precisely targeting the understanding of recursive NPs. When evaluated on RNPC, state-of-the-art Transformer models only perform around chance. Still, we show that such knowledge is learnable with appropriate data. We further probe the models for relevant linguistic features that can be learned from our tasks, including modifier semantic category and modifier scope. Finally, models trained on RNPC achieve strong zero-shot performance on an extrinsic Harm Detection evaluation task, showing the usefulness of the understanding of recursive NPs in downstream applications.

Recent work incorporates pre-trained word embeddings such as BERT embeddings into Neural Topic Models (NTMs), generating highly coherent topics. However, with high-quality contextualized document representations, do we really need sophisticated neural models to obtain coherent and interpretable topics? In this paper, we conduct thorough experiments showing that directly clustering high-quality sentence embeddings with an appropriate word selecting method can generate more coherent and diverse topics than NTMs, achieving also higher efficiency and simplicity.

Current pre-trained models applied for summarization are prone to factual inconsistencies that misrepresent the source text. Evaluating the factual consistency of summaries is thus necessary to develop better models. However, the human evaluation setup for evaluating factual consistency has not been standardized. To determine the factors that affect the reliability of the human evaluation, we crowdsource evaluations for factual consistency across state-of-the-art models on two news summarization datasets using the rating-based Likert Scale and ranking-based Best-Worst Scaling. Our analysis reveals that the ranking-based Best-Worst Scaling offers a more reliable measure of summary quality across datasets and that the reliability of Likert ratings highly depends on the target dataset and the evaluation design. To improve crowdsourcing reliability, we extend the scale of the Likert rating and present a scoring algorithm for Best-Worst Scaling that we call value learning. Our crowdsourcing guidelines will be publicly available to facilitate future work on factual consistency in summarization.