EMNLP 2022 论文列表

Knowledge distillation addresses the problem of transferring knowledge from a teacher model to a student model.In this process, we typically have multiple types of knowledge extracted from the teacher model.The problem is to make full use of them to train the student model.Our preliminary study shows that: (1) not all of the knowledge is necessary for learning a good student model, and (2) knowledge distillation can benefit from certain knowledge at different training steps.In response to these, we propose an actor-critic approach to selecting appropriate knowledge to transfer during the process of knowledge distillation.In addition, we offer a refinement of the training algorithm to ease the computational burden.Experimental results on the GLUE datasets show that our method outperforms several strong knowledge distillation baselines significantly.

Although explainable artificial intelligence (XAI) has achieved remarkable developments in recent years, there are few efforts have been devoted to the following problems, namely, i) how to develop an explainable method that could explain the black-box in a model-agnostic way? and ii) how to improve the performance and interpretability of the black-box using such explanations instead of pre-collected important attributions? To explore the potential solution, we propose a model-agnostic explanation method termed as Sparse Contrastive Coding (SCC) and verify its effectiveness in text classification and natural language inference. In brief, SCC explains the feature attributions which characterize the importance of words based on the hidden states of each layer of the model. With such word-level explainability, SCC adaptively divides the input sentences into foregrounds and backgrounds in terms of task relevance. Through maximizing the similarity between the foregrounds and input sentences while minimizing the similarity between the backgrounds and input sentences, SSC employs a supervised contrastive learning loss to boost the interpretability and performance of the model. Extensive experiments show the superiority of our method over five state-of-the-art methods in terms of interpretability and classification measurements. The code is available at https://pengxi.me.

Pretrained Language Models (LMs) have demonstrated ability to perform numerical reasoning by extrapolating from a few examples in few-shot settings. However, the extent to which this extrapolation relies on robust reasoning is unclear. In this paper, we investigate how well these models reason with terms that are less frequent in the pretraining data. In particular, we examine the correlations between the model performance on test instances and the frequency of terms from those instances in the pretraining data. We measure the strength of this correlation for a number of GPT-based language models (pretrained on the Pile dataset) on various numerical deduction tasks (e.g., arithmetic and unit conversion). Our results consistently demonstrate that models are more accurate on instances whose terms are more prevalent, in some cases above 70% (absolute) more accurate on the top 10% frequent terms in comparison to the bottom 10%. Overall, although LMs appear successful at few-shot numerical reasoning, our results raise the question of how much models actually generalize beyond pretraining data, and we encourage researchers to take the pretraining data into account when interpreting evaluation results.

Commonsense knowledge about the typicalfunctions of physical objects allows people tomake inferences during sentence understanding.For example, we infer that “Sam enjoyedthe book” means that Sam enjoyed reading thebook, even though the action is implicit. Priorresearch has focused on learning the prototypicalfunctions of physical objects in order toenable inferences about implicit actions. Butmany sentences refer to objects even when theyare not used (e.g., “The book fell”). We arguethat NLP systems need to recognize whether anobject is being used before inferring how theobject is used. We define a new task called ObjectUse Classification that determines whethera physical object mentioned in a sentence wasused or likely will be used. We introduce a newdataset for this task and present a classificationmodel that exploits data augmentation methodsand FrameNet when fine-tuning a pre-trainedlanguage model. We also show that object useclassification combined with knowledge aboutthe prototypical functions of objects has thepotential to yield very good inferences aboutimplicit and anticipated actions.

Adversarial examples in NLP are receiving increasing research attention. One line of investigation is the generation of word-level adversarial examples against fine-tuned Transformer models that preserve naturalness and grammaticality. Previous work found that human- and machine-generated adversarial examples are comparable in their naturalness and grammatical correctness. Most notably, humans were able to generate adversarial examples much more effortlessly than automated attacks. In this paper, we provide a detailed analysis of exactly how humans create these adversarial examples. By exploring the behavioural patterns of human workers during the generation process, we identify statistically significant tendencies based on which words humans prefer to select for adversarial replacement (e.g., word frequencies, word saliencies, sentiment) as well as where and when words are replaced in an input sequence. With our findings, we seek to inspire efforts that harness human strategies for more robust NLP models.

The task of response selection in multi-turn dialogue is to find the best option from all candidates. In order to improve the reasoning ability of the model, previous studies pay more attention to using explicit algorithms to model the dependencies between utterances, which are deterministic, limited and inflexible. In addition, few studies consider differences between the options before and after reasoning. In this paper, we propose an Implicit Relational Reasoning Graph Network to address these issues, which consists of the Utterance Relational Reasoner (URR) and the Option Dual Comparator (ODC). URR aims to implicitly extract dependencies between utterances, as well as utterances and options, and make reasoning with relational graph convolutional networks. ODC focuses on perceiving the difference between the options through dual comparison, which can eliminate the interference of the noise options. Experimental results on two multi-turn dialogue reasoning benchmark datasets MuTual and MuTualplus show that our method significantly improves the baseline of four pre-trained language models and achieves state-of-the-art performance. The model surpasses human performance for the first time on the MuTual dataset.

Evaluation metrics shine the light on the best models and thus strongly influence the research directions, such as the recently developed dialogue metrics USR, FED, and GRADE. However, most current metrics evaluate the dialogue data as isolated and static because they only focus on a single quality or several qualities. To mitigate the problem, this paper proposes an interpretable, multi-faceted, and controllable framework IM^2 (Interpretable and Multi-category Integrated Metric) to combine a large number of metrics which are good at measuring different qualities. The IM^2 framework first divides current popular dialogue qualities into different categories and then applies or proposes dialogue metrics to measure the qualities within each category and finally generates an overall IM^2 score. An initial version of IM^2 was submitted to the AAAI 2022 Track5.1@DSTC10 challenge and took the 2^nd place on both of the development and test leaderboard. After the competition, we develop more metrics and improve the performance of our model. We compare IM^2 with other 13 current dialogue metrics and experimental results show that IM^2 correlates more strongly with human judgments than any of them on each evaluated dataset.

We introduce a new open information extraction (OIE) benchmark for pre-trained language models (LM). Recent studies have demonstrated that pre-trained LMs, such as BERT and GPT, may store linguistic and relational knowledge. In particular, LMs are able to answer “fill-in-the-blank” questions when given a pre-defined relation category. Instead of focusing on pre-defined relations, we create an OIE benchmark aiming to fully examine the open relational information present in the pre-trained LMs. We accomplish this by turning pre-trained LMs into zero-shot OIE systems. Surprisingly, pre-trained LMs are able to obtain competitive performance on both standard OIE datasets (CaRB and Re-OIE2016) and two new large-scale factual OIE datasets (TAC KBP-OIE and Wikidata-OIE) that we establish via distant supervision. For instance, the zero-shot pre-trained LMs outperform the F1 score of the state-of-the-art supervised OIE methods on our factual OIE datasets without needing to use any training sets.

Machine Reading Comprehension (MRC) has become one of the essential tasks in Natural Language Understanding (NLU) as it is often included in several NLU benchmarks (Liang et al., 2020; Wilie et al., 2020). However, most MRC datasets only have answerable question type, overlooking the importance of unanswerable questions. MRC models trained only on answerable questions will select the span that is most likely to be the answer, even when the answer does not actually exist in the given passage (Rajpurkar et al., 2018). This problem especially remains in medium- to low-resource languages like Indonesian. Existing Indonesian MRC datasets (Purwarianti et al., 2007; Clark et al., 2020) are still inadequate because of the small size and limited question types, i.e., they only cover answerable questions. To fill this gap, we build a new Indonesian MRC dataset called I(n)don’tKnow- MRC (IDK-MRC) by combining the automatic and manual unanswerable question generation to minimize the cost of manual dataset construction while maintaining the dataset quality. Combined with the existing answerable questions, IDK-MRC consists of more than 10K questions in total. Our analysis shows that our dataset significantly improves the performance of Indonesian MRC models, showing a large improvement for unanswerable questions.

Transformer has been demonstrated effective in Neural Machine Translation (NMT). However, it is memory-consuming and time-consuming in edge devices, resulting in some difficulties for real-time feedback. To compress and accelerate Transformer, we propose a Hybrid Tensor-Train (HTT) decomposition, which retains full rank and meanwhile reduces operations and parameters. A Transformer using HTT, named Hypoformer, consistently and notably outperforms the recent light-weight SOTA methods on three standard translation tasks under different parameter and speed scales. In extreme low resource scenarios, Hypoformer has 7.1 points absolute improvement in BLEU and 1.27 X speedup than vanilla Transformer on IWSLT’14 De-En task.

We investigate input-conditioned hypernetworks for multi-tasking in NLP, generating parameter-efficient adaptations for a decoder using a hypernetwork conditioned on the output of an encoder. This approach produces a unique decoder adaptation for every input instance, allowing the network a larger degree of flexibility than prior work that only produces one decoder adaptation per task. We apply our method to sequence classification tasks, extractive QA, and summarisation and find that it surpasses previous parameter efficient fine-tuning methods and often outperforms fully finetuning the underlying model. An analysis of the embeddings used by our hypernetwork shows that they are sensitive to output label and type, suggesting that our approach better maps from encoder representations to output labels. Our code is publicly available at https://github.com/allenai/hyperdecoders.

Massively multilingual models are promising for transfer learning across tasks and languages. However, existing methods are unable to fully leverage training data when it is available in different task-language combinations. To exploit such heterogeneous supervision, we propose Hyper-X, a single hypernetwork that unifies multi-task and multilingual learning with efficient adaptation. It generates weights for adapter modules conditioned on both tasks and language embeddings. By learning to combine task and language-specific knowledge, our model enables zero-shot transfer for unseen languages and task-language combinations. Our experiments on a diverse set of languages demonstrate that Hyper-X achieves the best or competitive gain when a mixture of multiple resources is available, while on par with strong baseline in the standard scenario. Hyper-X is also considerably more efficient in terms of parameters and resources compared to methods that train separate adapters. Finally, Hyper-X consistently produces strong results in few-shot scenarios for new languages, showing the versatility of our approach beyond zero-shot transfer.

Summarization systems make numerous “decisions” about summary properties during inference, e.g. degree of copying, specificity and length of outputs, etc. However, these are implicitly encoded within model parameters and specific styles cannot be enforced. To address this, we introduce HydraSum, a new summarization architecture that extends the single decoder framework of current models to a mixture-of-experts version with multiple decoders. We show that HydraSum’s multiple decoders automatically learn contrasting summary styles when trained under the standard training objective without any extra supervision. Through experiments on three summarization datasets (CNN, Newsroom and XSum), we show that HydraSum provides a simple mechanism to obtain stylistically-diverse summaries by sampling from either individual decoders or their mixtures, outperforming baseline models. Finally, we demonstrate that a small modification to the gating strategy during training can enforce an even stricter style partitioning, e.g. high- vs low-abstractiveness or high- vs low-specificity, allowing users to sample from a larger area in the generation space and vary summary styles along multiple dimensions.

The shift of public debate to the digital sphere has been accompanied by a rise in online hate speech. While many promising approaches for hate speech classification have been proposed, studies often focus only on a single language, usually English, and do not address three key concerns: post-deployment performance, classifier maintenance and infrastructural limitations. In this paper, we introduce a new human-in-the-loop BERT-based hate speech classification pipeline and trace its development from initial data collection and annotation all the way to post-deployment. Our classifier, trained using data from our original corpus of over 422k examples, is specifically developed for the inherently multilingual setting of Switzerland and outperforms with its F1 score of 80.5 the currently best-performing BERT-based multilingual classifier by 5.8 F1 points in German and 3.6 F1 points in French. Our systematic evaluations over a 12-month period further highlight the vital importance of continuous, human-in-the-loop classifier maintenance to ensure robust hate speech classification post-deployment.

Fighting online hate speech is a challenge that is usually addressed using Natural Language Processing via automatic detection and removal of hate content. Besides this approach, counter narratives have emerged as an effective tool employed by NGOs to respond to online hate on social media platforms. For this reason, Natural Language Generation is currently being studied as a way to automatize counter narrative writing. However, the existing resources necessary to train NLG models are limited to 2-turn interactions (a hate speech and a counter narrative as response), while in real life, interactions can consist of multiple turns. In this paper, we present a hybrid approach for dialogical data collection, which combines the intervention of human expert annotators over machine generated dialogues obtained using 19 different configurations. The result of this work is DIALOCONAN, the first dataset comprising over 3000 fictitious multi-turn dialogues between a hater and an NGO operator, covering 6 targets of hate.

The multi-head self-attention mechanism of the transformer model has been thoroughly investigated recently. In one vein of study, researchers are interested in understanding why and how transformers work. In another vein, researchers propose new attention augmentation methods to make transformers more accurate, efficient and interpretable. In this paper, we combine these two lines of research in a human-in-the-loop pipeline to first discover important task-specific attention patterns. Then those patterns are injected, not only to smaller models, but also to the original model. The benefits of our pipeline and discovered patterns are demonstrated in two case studies with extractive summarization and topic segmentation. After discovering interpretable patterns in BERT-based models fine-tuned for the two downstream tasks, experiments indicate that when we inject the patterns into attention heads, the models show considerable improvements in accuracy and efficiency.

Timely and effective response to humanitarian crises requires quick and accurate analysis of large amounts of text data – a process that can highly benefit from expert-assisted NLP systems trained on validated and annotated data in the humanitarian response domain. To enable creation of such NLP systems, we introduce and release HumSet, a novel and rich multilingual dataset of humanitarian response documents annotated by experts in the humanitarian response community. The dataset provides documents in three languages (English, French, Spanish) and covers a variety of humanitarian crises from 2018 to 2021 across the globe. For each document, HUMSET provides selected snippets (entries) as well as assigned classes to each entry annotated using common humanitarian information analysis frameworks. HUMSET also provides novel and challenging entry extraction and multi-label entry classification tasks. In this paper, we take a first step towards approaching these tasks and conduct a set of experiments on Pre-trained Language Models (PLM) to establish strong baselines for future research in this domain. The dataset is available at https://blog.thedeep.io/humset/.

The task of multi-document summarization (MDS) aims at models that, given multiple documents as input, are able to generate a summary that combines disperse information, originally spread __across__ these documents. Accordingly, it is expected that both reference summaries in MDS datasets, as well as system summaries, would indeed be based on such dispersed information. In this paper, we argue for quantifying and assessing this expectation. To that end, we propose an automated measure for evaluating the degree to which a summary is “disperse”, in the sense of the number of source documents needed to cover its content. We apply our measure to empirically analyze several popular MDS datasets, with respect to their reference summaries, as well as the output of state-of-the-art systems. Our results show that certain MDS datasets barely require combining information from multiple documents, where a single document often covers the full summary content. Overall, we advocate using our metric for assessing and improving the degree to which summarization datasets require combining multi-document information, and similarly how summarization models actually meet this challenge.

Text-to-image generative models have achieved unprecedented success in generating high-quality images based on natural language descriptions. However, it is shown that these models tend to favor specific social groups when prompted with neutral text descriptions (e.g., ‘a photo of a lawyer’). Following Zhao et al. (2021), we study the effect on the diversity of the generated images when adding ethical intervention that supports equitable judgment (e.g., ‘if all individuals can be a lawyer irrespective of their gender’) in the input prompts. To this end, we introduce an Ethical NaTural Language Interventions in Text-to-Image GENeration (ENTIGEN) benchmark dataset to evaluate the change in image generations conditional on ethical interventions across three social axes – gender, skin color, and culture. Through CLIP-based and human evaluation on minDALL.E, DALL.E-mini and Stable Diffusion, we find that the model generations cover diverse social groups while preserving the image quality. In some cases, the generations would be anti-stereotypical (e.g., models tend to create images with individuals that are perceived as man when fed with prompts about makeup) in the presence of ethical intervention. Preliminary studies indicate that a large change in the model predictions is triggered by certain phrases such as ‘irrespective of gender’ in the context of gender bias in the ethical interventions. We release code and annotated data at https://github.com/Hritikbansal/entigen_emnlp.