ACL 2023 论文列表

Language identification (LID) is a fundamental step in many natural language processing pipelines. However, current LID systems are far from perfect, particularly on lower-resource languages. We present a LID model which achieves a macro-average F1 score of 0.93 and a false positive rate of 0.033% across 201 languages, outperforming previous work. We achieve this by training on a curated dataset of monolingual data, which we audit manually to ensure reliability. We make both the model and the dataset available to the research community. Finally, we carry out detailed analysis into our model’s performance, both in comparison to existing open models and by language class.

With a growing focus on morphological inflection systems for languages where high-quality data is scarce, training data noise is a serious but so far largely ignored concern. We aim at closing this gap by investigating the types of noise encountered within a pipeline for truly unsupervised morphological paradigm completion and its impact on morphological inflection systems: First, we propose an error taxonomy and annotation pipeline for inflection training data. Then, we compare the effect of different types of noise on multiple state-of-the- art inflection models. Finally, we propose a novel character-level masked language modeling (CMLM) pretraining objective and explore its impact on the models’ resistance to noise. Our experiments show that various architectures are impacted differently by separate types of noise, but encoder-decoders tend to be more robust to noise than models trained with a copy bias. CMLM pretraining helps transformers, but has lower impact on LSTMs.

Recent studies on automatic note generation have shown that doctors can save significant amounts of time when using automatic clinical note generation (Knoll et al., 2022). Summarization models have been used for this task to generate clinical notes as summaries of doctor-patient conversations (Krishna et al., 2021; Cai et al., 2022). However, assessing which model would best serve clinicians in their daily practice is still a challenging task due to the large set of possible correct summaries, and the potential limitations of automatic evaluation metrics. In this paper we study evaluation methods and metrics for the automatic generation of clinical notes from medical conversation. In particular, we propose new task-specific metrics and we compare them to SOTA evaluation metrics in text summarization and generation, including: (i) knowledge-graph embedding-based metrics, (ii) customized model-based metrics with domain-specific weights, (iii) domain-adapted/fine-tuned metrics, and (iv) ensemble metrics. To study the correlation between the automatic metrics and manual judgments, we evaluate automatic notes/summaries by comparing the system and reference facts and computing the factual correctness, and the hallucination and omission rates for critical medical facts. This study relied on seven datasets manually annotated by domain experts. Our experiments show that automatic evaluation metrics can have substantially different behaviors on different types of clinical notes datasets. However, the results highlight one stable subset of metrics as the most correlated with human judgments with a relevant aggregation of different evaluation criteria.

Controlled generation refers to the problem of creating text that contains stylistic or semantic attributes of interest. Many approaches reduce this problem to training a predictor of the desired attribute. For example, researchers hoping to deploy a large language model to produce non-toxic content may use a toxicity classifier to filter generated text. In practice, the generated text to classify, which is determined by user prompts, may come from a wide range of distributions. In this paper, we show that the performance of controlled generation may be poor if the distributions of text in response to user prompts differ from the distribution the predictor was trained on. To address this problem, we cast controlled generation under distribution shift as an invariant learning problem: the most effective predictor should be invariant across multiple text environments. We then discuss a natural solution that arises from this characterization and propose heuristics for selecting natural environments. We study this characterization and the proposed method empirically using both synthetic and real data. Experiments demonstrate both the challenge of distribution shift in controlled generation and the potential of invariance methods in this setting.

One crucial aspect of democracy is fair information sharing. While it is hard to prevent biases in news, they should be identified for better transparency. We propose an approach to automatically characterize biases that takes into account structural differences and that is efficient for long texts. This yields new ways to provide explanations for a textual classifier, going beyond mere lexical cues. We show that: (i) the use of discourse-based structure-aware document representations compare well to local, computationally heavy, or domain-specific models on classification tasks that deal with textual bias (ii) our approach based on different levels of granularity allows for the generation of better explanations of model decisions, both at the lexical and structural level, while addressing the challenge posed by long texts.

Recent years have witnessed the thriving of pretrained Transformer-based language models for understanding semi-structured tables, with several applications, such as Table Question Answering (TableQA).These models are typically trained on joint tables and surrounding natural language text, by linearizing table content into sequences comprising special tokens and cell information. This yields very long sequences which increase system inefficiency, and moreover, simply truncating long sequences results in information loss for downstream tasks. We propose Inner Table Retriever (ITR), a general-purpose approach for handling long tables in TableQA that extracts sub-tables to preserve the most relevant information for a question. We show that ITR can be easily integrated into existing systems to improve their accuracy with up to 1.3-4.8% and achieve state-of-the-art results in two benchmarks, i.e., 63.4% in WikiTableQuestions and 92.1% in WikiSQL. Additionally, we show that ITR makes TableQA systems more robust to reduced model capacity and to different ordering of columns and rows. We make our code available at: https://github.com/amazon-science/robust-tableqa.

Natural language processing (NLP) researchers develop models of grammar, meaning and communication based on written text. Due to task and data differences, what is considered text can vary substantially across studies. A conceptual framework for systematically capturing these differences is lacking. We argue that clarity on the notion of text is crucial for reproducible and generalizable NLP. Towards that goal, we propose common terminology to discuss the production and transformation of textual data, and introduce a two-tier taxonomy of linguistic and non-linguistic elements that are available in textual sources and can be used in NLP modeling. We apply this taxonomy to survey existing work that extends the notion of text beyond the conservative language-centered view. We outline key desiderata and challenges of the emerging inclusive approach to text in NLP, and suggest community-level reporting as a crucial next step to consolidate the discussion.

Back-translation is a data augmentation technique that has been shown to improve model quality through the creation of synthetic training bitext. Early studies showed the promise of the technique and follow on studies have produced additional refinements. We have undertaken a broad investigation using back-translation to train models from 60 languages into English; the majority of these languages are considered moderate- or low-resource languages. We observed consistent gains, though compared to prior work we saw conspicuous gains in quite a number of lower-resourced languages. We analyzed differences in translations between baseline and back-translation models, and observed many indications of improved translation quality. Translation of both rare and common terms is improved, and these improvements occur despite the less natural synthetic source-language text used in training.

Recently, multi-aspect controllable text generation that controls the generated text in multiple aspects (e.g., sentiment, topic, and keywords) has attracted increasing attention. Although methods based on parameter efficient tuning like prefix-tuning could achieve multi-aspect controlling in a plug-and-play way, the mutual interference of multiple prefixes leads to significant degeneration of constraints and limits their extensibility to training-time unseen aspect combinations. In this work, we provide a theoretical lower bound for the interference and empirically found that the interference grows with the number of layers where prefixes are inserted. Based on these analyses, we propose using trainable gates to normalize the intervention of prefixes to restrain the growing interference. As a result, controlling training-time unseen combinations of aspects can be realized by simply concatenating corresponding plugins such that new constraints can be extended at a lower cost. In addition, we propose a unified way to process both categorical and free-form constraints. Experiments on text generation and machine translation demonstrate the superiority of our approach over baselines on constraint accuracy, text quality, and extensibility.

Text-to-SQL translates user queries into SQL statements that can retrieve relevant answers from relational databases. Recent approaches to Text-to-SQL rely on pre-trained language models that are computationally expensive and technically challenging to deploy in real-world applications that require real-time or on-device processing capabilities. In this paper, we perform a focused study on the feasibility of applying recent model compression techniques to sketch-based and sequence-to-sequence Text-to-SQL models. Our results reveal that sketch-based Text-to-SQL models generally have higher inference efficiency and respond better to model compression than sequence-to-sequence models, making them ideal for real-world deployments, especially in use cases with simple SQL statements.

Standard methods for multi-label text classification largely rely on encoder-only pre-trained language models, whereas encoder-decoder models have proven more effective in other classification tasks. In this study, we compare four methods for multi-label classification, two based on an encoder only, and two based on an encoder-decoder. We carry out experiments on four datasets—two in the legal domain and two in the biomedical domain, each with two levels of label granularity— and always depart from the same pre-trained model, T5. Our results show that encoder-decoder methods outperform encoder-only methods, with a growing advantage on more complex datasets and labeling schemes of finer granularity. Using encoder-decoder models in a non-autoregressive fashion, in particular, yields the best performance overall, so we further study this approach through ablations to better understand its strengths.

Aspect-Based Sentiment Analysis (ABSA) aims to recognize fine-grained opinions and sentiments of users, which is an important problem in sentiment analysis. Recent work has shown that Sentiment-enhanced Pre-Training (SPT) can substantially improve the performance of various ABSA tasks. However, there is currently a lack of comprehensive evaluation and fair comparison of existing SPT approaches. Therefore, this paper performs an empirical study to investigate the effectiveness of different SPT approaches. First, we develop an effective knowledge-mining method and leverage it to build a large-scale knowledge-annotated SPT corpus. Second, we systematically analyze the impact of integrating sentiment knowledge and other linguistic knowledge in pre-training. For each type of sentiment knowledge, we also examine and compare multiple integration methods. Finally, we conduct extensive experiments on a wide range of ABSA tasks to see how much SPT can facilitate the understanding of aspect-level sentiments.

Question and answer generation (QAG) consists of generating a set of question-answer pairs given a context (e.g. a paragraph). This task has a variety of applications, such as data augmentation for question answering (QA) models, information retrieval and education. In this paper, we establish baselines with three different QAG methodologies that leverage sequence-to-sequence language model (LM) fine-tuning. Experiments show that an end-to-end QAG model, which is computationally light at both training and inference times, is generally robust and outperforms other more convoluted approaches. However, there are differences depending on the underlying generative LM. Finally, our analysis shows that QA models fine-tuned solely on generated question-answer pairs can be competitive when compared to supervised QA models trained on human-labeled data.

The increasingly large size of modern pre-trained language models not only makes them inherit more human-like biases from the training corpora, but also makes it computationally expensive to mitigate such biases. In this paper, we investigate recent parameter-efficient methods in combination with counterfactual data augmentation (CDA) for bias mitigation. We conduct extensive experiments with prefix tuning, prompt tuning, and adapter tuning on different language models and bias types to evaluate their debiasing performance and abilities to preserve the internal knowledge of a pre-trained model. We find that the parameter-efficient methods (i) are effective in mitigating gender bias, where adapter tuning is consistently the most effective one and prompt tuning is more suitable for GPT-2 than BERT, (ii) areless effective when it comes to racial and religious bias, which may be attributed to the limitations of CDA, and (iii) can perform similarly to or sometimes better than full fine-tuning with improved time and memory efficiency, as well as maintain the internal knowledge in BERT and GPT-2, evaluated via fact retrieval and downstream fine-tuning.

Task-oriented semantic parsing has drawn a lot of interest from the NLP community, and especially the voice assistant industry as it enables representing the meaning of user requests with arbitrarily nested semantics, including multiple intents and compound entities. SOTA models are large seq2seq transformers and require hundreds of thousands of annotated examples to be trained. However annotating such data to bootstrap new domains or languages is expensive and error-prone, especially for requests made of nested semantics. In addition large models easily break the tight latency constraints imposed in a user-facing production environment. As part of this work we explore leveraging external knowledge to improve model accuracy in low-resource and low-compute settings. We demonstrate that using knowledge-enhanced encoders inside seq2seq models does not result in performance gains by itself, but jointly learning to uncover entities in addition to the parse generation is a simple yet effective way of improving performance across the board. We show this is especially true in the low-compute scarce-data setting and for entity-rich domains, with relative gains up to 74.48% on the TOPv2 dataset.

Named entity recognition (NER) is the task to detect and classify entity spans in the text. When entity spans overlap between each other, the task is named as nested NER. Span-based methods have been widely used to tackle nested NER. Most of these methods get a score matrix, where each entry corresponds to a span. However, previous work ignores spatial relations in the score matrix. In this paper, we propose using Convolutional Neural Network (CNN) to model these spatial relations. Despite being simple, experiments in three commonly used nested NER datasets show that our model surpasses several recently proposed methods with the same pre-trained encoders. Further analysis shows that using CNN can help the model find more nested entities. Besides, we find that different papers use different sentence tokenizations for the three nested NER datasets, which will influence the comparison. Thus, we release a pre-processing script to facilitate future comparison.

The effectiveness of contrastive learning technology in natural language processing tasks is yet to be explored and analyzed. How to construct positive and negative samples correctly and reasonably is the core challenge of contrastive learning. It is even harder to discover contrastive objects in multi-label text classification tasks. There are very few contrastive losses proposed previously. In this paper, we investigate the problem from a different angle by proposing five novel contrastive losses for multi-label text classification tasks. These are Strict Contrastive Loss (SCL), Intra-label Contrastive Loss (ICL), Jaccard Similarity Contrastive Loss (JSCL), Jaccard Similarity Probability Contrastive Loss (JSPCL), and Stepwise Label Contrastive Loss (SLCL). We explore the effectiveness of contrastive learning for multi-label text classification tasks by the employment of these novel losses and provide a set of baseline models for deploying contrastive learning techniques on specific tasks. We further perform an interpretable analysis of our approach to show how different components of contrastive learning losses play their roles. The experimental results show that our proposed contrastive losses can bring improvement to multi-label text classification tasks. Our work also explores how contrastive learning should be adapted for multi-label text classification tasks.

With a surge in identifying suicidal risk and its severity in social media posts, we argue that a more consequential and explainable research is required for optimal impact on clinical psychology practice and personalized mental healthcare. The success of computational intelligence techniques for inferring mental illness from social media resources, points to natural language processing as a lens for determining Interpersonal Risk Factors (IRF) in human writings. Motivated with limited availability of datasets for social NLP research community, we construct and release a new annotated dataset with human-labelled explanations and classification of IRF affecting mental disturbance on social media: (i) Thwarted Belongingness (TBe), and (ii) Perceived Burdensomeness (PBu). We establish baseline models on our dataset facilitating future research directions to develop real-time personalized AI models by detecting patterns of TBe and PBu in emotional spectrum of user’s historical social media profile.

Recent works have introduced Abstract Meaning Representation (AMR) for Document-level Event Argument Extraction (Doc-level EAE), since AMR provides a useful interpretation of complex semantic structures and helps to capture long-distance dependency. However, in these works AMR is used only implicitly, for instance, as additional features or training signals. Motivated by the fact that all event structures can be inferred from AMR, this work reformulates EAE as a link prediction problem on AMR graphs. Since AMR is a generic structure and does not perfectly suit EAE, we propose a novel graph structure, Tailored AMR Graph (TAG), which compresses less informative subgraphs and edge types, integrates span information, and highlights surrounding events in the same document. With TAG, we further propose a novel method using graph neural networks as a link prediction model to find event arguments. Our extensive experiments on WikiEvents and RAMS show that this simpler approach outperforms the state-of-the-art models by 3.63pt and 2.33pt F1, respectively, and do so with reduced 56% inference time.