ACL 2023 论文列表

Chain-of-thought prompting (e.g., “Let’s think step-by-ste”) primes large language models to verbalize rationalization for their predictions. While chain-of-thought can lead to dramatic performance gains, benefits appear to emerge only for sufficiently large models (beyond 50B parameters). We show that orders-of-magnitude smaller models (125M—1.3B parameters) can still benefit from chain-of-thought prompting. To achieve this, we introduce Symbolic Chain-of-Thought Distillation (SCoTD), a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. Experiments across several commonsense benchmarks show that: 1) SCoTD enhances the performance of the student model in both supervised and few-shot settings, and especially for challenge sets; 2) sampling many reasoning chains per instance from the teacher is paramount; and 3) after distillation, student chain-of-thoughts are judged by humans as comparable to the teacher, despite orders of magnitude fewer parameters. We test several hypotheses regarding what properties of chain-of-thought samples are important, e.g., diversity vs. teacher likelihood vs. open-endedness. We release our corpus of chain-of-thought samples and code.

Augmenting language models with a retrieval mechanism has been shown to significantly improve their performance while keeping the number of parameters low. Retrieval-augmented models commonly rely on a semantic retrieval mechanism based on the similarity between dense representations of the query chunk and potential neighbors. In this paper, we study the state-of-the-art Retro model and observe that its performance gain is better explained by surface-level similarities, such as token overlap. Inspired by this, we replace the semantic retrieval in Retro with a surface-level method based on BM25, obtaining a significant reduction in perplexity. As full BM25 retrieval can be computationally costly for large datasets, we also apply it in a re-ranking scenario, gaining part of the perplexity reduction with minimal computational overhead.

Extracting generalized and robust representations is a major challenge in emotion recognition in conversations (ERC). To address this, we propose a supervised adversarial contrastive learning (SACL) framework for learning class-spread structured representations in a supervised manner. SACL applies contrast-aware adversarial training to generate worst-case samples and uses joint class-spread contrastive learning to extract structured representations. It can effectively utilize label-level feature consistency and retain fine-grained intra-class features. To avoid the negative impact of adversarial perturbations on context-dependent data, we design a contextual adversarial training (CAT) strategy to learn more diverse features from context and enhance the model’s context robustness. Under the framework with CAT, we develop a sequence-based SACL-LSTM to learn label-consistent and context-robust features for ERC. Experiments on three datasets show that SACL-LSTM achieves state-of-the-art performance on ERC. Extended experiments prove the effectiveness of SACL and CAT.

The goal of multimodal abstractive summarization (MAS) is to produce a concise summary given the multimodal data (text and vision). Existing studies on MAS mainly focus on how to effectively use the extracted visual features, having achieved impressive success on the high-resource English dataset. However, less attention has been paid to the quality of the visual features to the summary, which may limit the model performance, especially in the low- and zero-resource scenarios. In this paper, we propose to improve the summary quality through summary-oriented visual features. To this end, we devise two auxiliary tasks including vision to summary task and masked image modeling task. Together with the main summarization task, we optimize the MAS model via the training objectives of all these tasks. By these means, the MAS model can be enhanced by capturing the summary-oriented visual features, thereby yielding more accurate summaries. Experiments on 44 languages, covering mid-high-, low-, and zero-resource scenarios, verify the effectiveness and superiority of the proposed approach, which achieves state-of-the-art performance under all scenarios. Additionally, we will contribute a large-scale multilingual multimodal abstractive summarization (MM-Sum) dataset to the research community.

Large language models, particularly GPT-3, are able to produce high quality summaries ofgeneral domain news articles in few- and zero-shot settings. However, it is unclear if such models are similarly capable in more specialized domains such as biomedicine. In this paper we enlist domain experts (individuals with medical training) to evaluate summaries of biomedical articles generated by GPT-3, given no supervision. We consider bothsingle- and multi-document settings. In the former, GPT-3 is tasked with generating regular and plain-language summaries of articles describing randomized controlled trials; in thelatter, we assess the degree to which GPT-3 is able to synthesize evidence reported acrossa collection of articles. We design an annotation scheme for evaluating model outputs, withan emphasis on assessing the factual accuracy of generated summaries. We find that whileGPT-3 is able to summarize and simplify single biomedical articles faithfully, it strugglesto provide accurate aggregations of findings over multiple documents. We release all data,code, and annotations used in this work.

Subword segmenters like BPE operate as a preprocessing step in neural machine translation and other (conditional) language models. They are applied to datasets before training, so translation or text generation quality relies on the quality of segmentations. We propose a departure from this paradigm, called subword segmental machine translation (SSMT). SSMT unifies subword segmentation and MT in a single trainable model. It learns to segment target sentence words while jointly learning to generate target sentences. To use SSMT during inference we propose dynamic decoding, a text generation algorithm that adapts segmentations as it generates translations. Experiments across 6 translation directions show that SSMT improves chrF scores for morphologically rich agglutinative languages. Gains are strongest in the very low-resource scenario. SSMT also learns subwords that are closer to morphemes compared to baselines and proves more robust on a test set constructed for evaluating morphological compositional generalisation.

Measuring semantic change has thus far remained a task where methods using contextual embeddings have struggled to improve upon simpler techniques relying only on static word vectors. Moreover, many of the previously proposed approaches suffer from downsides related to scalability and ease of interpretation. We present a simplified approach to measuring semantic change using contextual embeddings, relying only on the most probable substitutes for masked terms. Not only is this approach directly interpretable, it is also far more efficient in terms of storage, achieves superior average performance across the most frequently cited datasets for this task, and allows for more nuanced investigation of change than is possible with static word vectors.

k-nearest-neighbor machine translation (kNN-MT) (Khandelwal et al., 2021) boosts the translation performance of trained neural machine translation (NMT) models by incorporating example-search into the decoding algorithm. However, decoding is seriously time-consuming, i.e., roughly 100 to 1,000 times slower than standard NMT, because neighbor tokens are retrieved from all target tokens of parallel data in each timestep. In this paper, we propose “Subset kNN-MT”, which improves the decoding speed of kNN-MT by two methods: (1) retrieving neighbor target tokens from a subset that is the set of neighbor sentences of the input sentence, not from all sentences, and (2) efficient distance computation technique that is suitable for subset neighbor search using a look-up table. Our proposed method achieved a speed-up of up to 132.2 times and an improvement in BLEU score of up to 1.6 compared with kNN-MT in the WMT’19 De-En translation task and the domain adaptation tasks in De-En and En-Ja.

Human-annotated data plays a critical role in the fairness of AI systems, including those that deal with life-altering decisions or moderating human-created web/social media content. Conventionally, annotator disagreements are resolved before any learning takes place. However, researchers are increasingly identifying annotator disagreement as pervasive and meaningful. They also question the performance of a system when annotators disagree. Particularly when minority views are disregarded, especially among groups that may already be underrepresented in the annotator population. In this paper, we introduce CrowdOpinion, an unsupervised learning based approach that uses language features and label distributions to pool similar items into larger samples of label distributions. We experiment with four generative and one density-based clustering method, applied to five linear combinations of label distributions and features. We use five publicly available benchmark datasets (with varying levels of annotator disagreements) from social media (Twitter, Gab, and Reddit). We also experiment in the wild using a dataset from Facebook, where annotations come from the platform itself by users reacting to posts. We evaluate CrowdOpinion as a label distribution prediction task using KL-divergence and a single-label problem using accuracy measures.

In NLP, recent work has seen increased focus on spurious correlations between various features and labels in training data, and how these influence model behavior. However, the presence and effect of such correlations are typically examined feature by feature. We investigate the cumulative impact on a model of many such intersecting features. Using a new statistical method, we examine whether such spurious patterns in data appear in models trained on the data. We select two tasks— natural language inference and duplicate-question detection— for which any unigram feature on its own should ideally be uninformative, which gives us a large pool of automatically extracted features with which to experiment. The large size of this pool allows us to investigate the intersection of features spuriously associated with (potentially different) labels. We then apply an optimization approach to *reweight* the training data, reducing thousands of spurious correlations, and examine how doing so affects models trained on the reweighted data. Surprisingly, though this method can successfully reduce lexical biases in the training data, we still find strong evidence of corresponding bias in the trained models, including worsened bias for slightly more complex features (bigrams). We close with discussion about the implications of our results on what it means to “debias” training data, and how issues of data quality can affect model bias.

The increasing sizes of large generative Pre-trained Language Models (PLMs) hinder their deploymentin real-world applications. To obtain efficient PLMs, previous studies mostly focus on pruning the attention heads and feed-forward networks (FFNs) of the Transformer. Nevertheless, we find that in generative PLMs, the hidden dimension shared by many other modules (e.g., embedding layer and layer normalization) contains persistent outliers regardless of the network input. This study comprehensively investigates the structured pruning of generative PLMs with all the above compressible components. To identify redundant network structures, we assign learnable masks over compressible components followed by sparse training. Various sizes of PLMs can be flexibly extracted via different thresholds, and are then task-specifically fine-tuned for further improvement. Extensive experiments on language modeling, summarization and machine translation validate the effectiveness of the proposed method. For example, the pruned BART brings 1.51x/6.96x inference speedup on GPU/CPU with 67% size reduction, and can be further combined with quantization for more than 25× compression.

We present an annotation approach for capturing structured components and arguments inlegal case solutions of German students. Based on the appraisal style, which dictates the structured way of persuasive writing in German law, we propose an annotation scheme with annotation guidelines that identify structured writing in legal case solutions. We conducted an annotation study with two annotators and annotated legal case solutions to capture the structures of a persuasive legal text. Based on our dataset, we trained three transformer-based models to show that the annotated components can be successfully predicted, e.g. to provide users with writing assistance for legal texts. We evaluated a writing support system in which our models were integrated in an online experiment with law students and found positive learning success and users’ perceptions. Finally, we present our freely available corpus of 413 law student case studies to support the development of intelligent writing support systems.

In this work, we enhance higher-order graph-based approaches for span-based semantic role labeling (SRL) by means of structured modeling. To decrease the complexity of higher-order modeling, we decompose the edge from predicate word to argument span into three different edges, predicate-to-head (P2H), predicate-to-tail (P2T), and head-to-tail (H2T), where head/tail means the first/last word of the semantic argument span. As such, we use a CRF-based higher-order dependency parser and leverage Mean-Field Variational Inference (MFVI) for higher-order inference. Moreover, since semantic arguments of predicates are often constituents within a constituency parse tree, we can leverage such nice structural property by defining a TreeCRF distribution over all H2T edges, using the idea of partial marginalization to define structural training loss. We further leverage structured MFVI to enhance inference. We experiment on span-based SRL benchmarks, showing the effectiveness of both higher-order and structured modeling and the combination thereof. In addition, we show superior performance of structured MFVI against vanilla MFVI.

Conditional question answering on long documents aims to find probable answers and identify conditions that need to be satisfied to make the answers correct over long documents. Existing approaches solve this task by segmenting long documents into multiple sections, and attending information at global and local tokens to predict the answers and corresponding conditions. However, the natural structure of the document and discourse relations between sentences in each document section are ignored, which are crucial for condition retrieving across sections, as well as logical interaction over the question and conditions. To address this issue, this paper constructs a Structure-Discourse Hierarchical Graph (SDHG) and conducts bottom-up information propagation. Firstly we build the sentence-level discourse graphs for each section and encode the discourse relations by graph attention. Secondly, we construct a section-level structure graph based on natural structures, and conduct interactions over the question and contexts. Finally different levels of representations are integrated into jointly answer and condition decoding. The experiments on the benchmark ConditionalQA shows our approach gains over the prior state-of-the-art, by 3.0 EM score and 2.4 F1 score on answer measuring, as well as 2.2 EM score and 1.9 F1 score on jointly answer and condition measuring.

This paper presents Structure Aware Dense Retrieval (SANTA) model, which encodes user queries and structured data in one universal embedding space for retrieving structured data. SANTA proposes two pretraining methods to make language models structure-aware and learn effective representations for structured data: 1) Structured Data Alignment, which utilizes the natural alignment relations between structured data and unstructured data for structure-aware pretraining. It contrastively trains language models to represent multi-modal text data and teaches models to distinguish matched structured data for unstructured texts. 2) Masked Entity Prediction, which designs an entity-oriented mask strategy and asks language models to fill in the masked entities. Our experiments show that SANTA achieves state-of-the-art on code search and product search and conducts convincing results in the zero-shot setting. SANTA learns tailored representations for multi-modal text data by aligning structured and unstructured data pairs and capturing structural semantics by masking and predicting entities in the structured data. All codes are available at https://github.com/OpenMatch/OpenMatch.

To present, multilingual language models trained using various pre-training tasks like mask language modeling (MLM) have yielded encouraging results on a wide range of downstream tasks. Despite the promising performances, structural knowledge in cross-lingual corpus is less explored in current works, leading to the semantic misalignment. In this paper, we propose a new pre-training task named Structural Contrast Pretraining (SCP) to align the structural words in a parallel sentence, enhancing the models’ ability to comprehend cross-lingual representations. Concretely, each structural word in source and target languages is regarded as a positive pair in SCP. Since contrastive learning compares positive and negative pairs, an increase in the frequency of negative pairings could enhance the performance of the resulting model. Therefore, we further propose Cross-lingual Momentum Contrast (CL-MoCo) to increase the number of negative pairs by maintaining a large size of the queue. CL-MoCo extends the original Moco approach into cross-lingual training and jointly optimizes the source-to-target language and target-to-source language representations, resulting in a more suitable encoder for cross-lingual transfer. We conduct extensive experiments to validate the proposed approach on three cross-lingual tasks across five datasets such as MLQA, WikiAnn, etc, and results prove the effectiveness of our method.

We have investigated methods utilizing hierarchical structure information representation in the semantic parsing task and have devised a method that reinforces the semantic awareness of a pre-trained language model via a two-step fine-tuning mechanism: hierarchical structure information strengthening and a final specific task. The model used is better than existing ones at learning the contextual representations of utterances embedded within its hierarchical semantic structure and thereby improves system performance. In addition, we created a mechanism using inductive grammar to dynamically prune the unpromising directions in the semantic structure parsing process. Finally, through experimentsOur code will be published when this paper is accepted. on the TOP and TOPv2 (low-resource setting) datasets, we achieved state-of-the-art (SOTA) performance, confirming the effectiveness of our proposed model.

Collaborative stories, which are texts created through the collaborative efforts of multiple authors with different writing styles and intentions, pose unique challenges for NLP models. Understanding and generating such stories remains an underexplored area due to the lack of open-domain corpora. To address this, we introduce StoryWars, a new dataset of over 40,000 collaborative stories written by 9,400 different authors from an online platform. We design 12 task types, comprising 7 understanding and 5 generation task types, on {pasted macro ‘STORYWARS’}, deriving 101 diverse story-related tasks in total as a multi-task benchmark covering all fully-supervised, few-shot, and zero-shot scenarios. Furthermore, we present our instruction-tuned model, InstructStory, for the story tasks showing that instruction tuning, in addition to achieving superior results in zero-shot and few-shot scenarios, can also obtain the best performance on the fully-supervised tasks in StoryWars, establishing strong multi-task benchmark performances on StoryWars.

Non-parallel text style transfer is an important task in natural language generation. However, previous studies concentrate on the token or sentence level, such as sentence sentiment and formality transfer, but neglect long style transfer at the discourse level. Long texts usually involve more complicated author linguistic preferences such as discourse structures than sentences. In this paper, we formulate the task of non-parallel story author-style transfer, which requires transferring an input story into a specified author style while maintaining source semantics. To tackle this problem, we propose a generation model, named StoryTrans, which leverages discourse representations to capture source content information and transfer them to target styles with learnable style embeddings. We use an additional training objective to disentangle stylistic features from the learned discourse representation to prevent the model from degenerating to an auto-encoder. Moreover, to enhance content preservation, we design a mask-and-fill framework to explicitly fuse style-specific keywords of source texts into generation. Furthermore, we constructed new datasets for this task in Chinese and English, respectively. Extensive experiments show that our model outperforms strong baselines in overall performance of style transfer and content preservation.