ACL 2023 论文列表

Lexically-constrained NMT (LNMT) aims to incorporate user-provided terminology into translations. Despite its practical advantages, existing work has not evaluated LNMT models under challenging real-world conditions. In this paper, we focus on two important but understudied issues that lie in the current evaluation process of LNMT studies. The model needs to cope with challenging lexical constraints that are “homographs” or “unseen” during training. To this end, we first design a homograph disambiguation module to differentiate the meanings of homographs. Moreover, we propose PLUMCOT which integrates contextually rich information about unseen lexical constraints from pre-trained language models and strengthens a copy mechanism of the pointer network via direct supervision of a copying score. We also release HOLLY, an evaluation benchmark for assessing the ability of model to cope with “homographic” and “unseen” lexical constraints. Experiments on HOLLY and the previous test setup show the effectiveness of our method. The effects of PLUMCOT are shown to be remarkable in “unseen” constraints. Our dataset is available at https://github.com/papago-lab/HOLLY-benchmark.

Measurement of interaction quality is a critical task for the improvement of large-scale spoken dialog systems. Existing approaches to dialog quality estimation either focus on evaluating the quality of individual turns, or collect dialog-level quality measurements from end users immediately following an interaction. In contrast to these approaches, we introduce a new dialog-level annotation workflow called Dialog Quality Annotation (DQA). DQA expert annotators evaluate the quality of dialogs as a whole, and also label dialogs for attributes such as goal completion and user sentiment. In this contribution, we show that: (i) while dialog quality cannot be completely decomposed into dialog-level attributes, there is a strong relationship between some objective dialog attributes and judgments of dialog quality; (ii) for the task of dialog-level quality estimation, a supervised model trained on dialog-level annotations outperforms methods based purely on aggregating turn-level features; and (iii) the proposed evaluation model shows better domain generalization ability compared to the baselines. On the basis of these results, we argue that having high-quality human-annotated data is an important component of evaluating interaction quality for large industrial-scale voice assistant platforms.

Voice dictation is an increasingly important text input modality. Existing systems that allow both dictation and editing-by-voice restrict their command language to flat templates invoked by trigger words. In this work, we study the feasibility of allowing users to interrupt their dictation with spoken editing commands in open-ended natural language. We introduce a new task and dataset, TERTiUS, to experiment with such systems. To support this flexibility in real-time, a system must incrementally segment and classify spans of speech as either dictation or command, and interpret the spans that are commands. We experiment with using large pre-trained language models to predict the edited text, or alternatively, to predict a small text-editing program. Experiments show a natural trade-off between model accuracy and latency: a smaller model achieves 30% end-state accuracy with 1.3 seconds of latency, while a larger model achieves 55% end-state accuracy with 7 seconds of latency.

The pretrained language model (PLM) based metrics have been successfully used in evaluating language generation tasks. Recent studies of the human evaluation community show that considering both major errors (e.g. mistranslated tokens) and minor errors (e.g. imperfections in fluency) can produce high-quality judgments. This inspires us to approach the final goal of the automatic metrics (human-like evaluations) by fine-grained error analysis. In this paper, we argue that the ability to estimate sentence confidence is the tip of the iceberg for PLM-based metrics. And it can be used to refine the generated sentence toward higher confidence and more reference-grounded, where the costs of refining and approaching reference are used to determine the major and minor errors, respectively. To this end, we take BARTScore as the testbed and present an innovative solution to marry the unexploited sentence refining capacity of BARTScore and human-like error analysis, where the final score consists of both the evaluations of major and minor errors. Experiments show that our solution consistently and significantly improves BARTScore, and outperforms top-scoring metrics in 19/25 test settings. Analyses demonstrate our method robustly and efficiently approaches human-like evaluations, enjoying better interpretability. Our code and scripts will be publicly released in https://github.com/Coldmist-Lu/ErrorAnalysis_NLGEvaluation.

Radiology report summarization (RRS) is a growing area of research. Given the Findings section of a radiology report, the goal is to generate a summary (called an Impression section) that highlights the key observations and conclusions of the radiology study. However, RRS currently faces essential limitations. First, many prior studies conduct experiments on private datasets, preventing reproduction of results and fair comparisons across different systems and solutions. Second, most prior approaches are evaluated solely on chest X-rays. To address these limitations, we propose a dataset (MIMIC-RRS) involving three new modalities and seven new anatomies based on the MIMIC-III and MIMIC-CXR datasets. We then conduct extensive experiments to evaluate the performance of models both within and across modality-anatomy pairs in MIMIC-RRS. In addition, we evaluate their clinical efficacy via RadGraph, a factual correctness metric.

Millions of users are active on social media. To allow users to better showcase themselves and network with others, we explore the auto-generation of social media self-introduction, a short sentence outlining a user’s personal interests. While most prior work profiling users with tags (e.g., ages), we investigate sentence-level self-introductions to provide a more natural and engaging way for users to know each other. Here we exploit a user’s tweeting history to generate their self-introduction. The task is non-trivial because the history content may be lengthy, noisy, and exhibit various personal interests. To address this challenge, we propose a novel unified topic-guided encoder-decoder (UTGED) framework; it models latent topics to reflect salient user interest, whose topic mixture then guides encoding a user’s history and topic words control decoding their self-introduction. For experiments, we collect a large-scale Twitter dataset, and extensive results show the superiority of our UTGED to the advanced encoder-decoder models without topic modeling.

The task of Stance Detection is concerned with identifying the attitudes expressed by an author towards a target of interest. This task spans a variety of domains ranging from social media opinion identification to detecting the stance for a legal claim. However, the framing of the task varies within these domains in terms of the data collection protocol, the label dictionary and the number of available annotations. Furthermore, these stance annotations are significantly imbalanced on a per-topic and inter-topic basis. These make multi-domain stance detection challenging, requiring standardization and domain adaptation. To overcome this challenge, we propose Topic Efficient StancE Detection (TESTED), consisting of a topic-guided diversity sampling technique used for creating a multi-domain data efficient training set and a contrastive objective that is used for fine-tuning a stance classifier using the produced set. We evaluate the method on an existing benchmark of 16 datasets with in-domain, i.e. all topics seen and out-of-domain, i.e. unseen topics, experiments. The results show that the method outperforms the state-of-the-art with an average of 3.5 F1 points increase in-domain and is more generalizable with an averaged 10.2 F1 on out-of-domain evaluation while using <10% of the training data. We show that our sampling technique mitigates both inter- and per-topic class imbalances. Finally, our analysis demonstrates that the contrastive learning objective allows the model for a more pronounced segmentation of samples with varying labels.

Understanding emotion expressions in multimodal signals is key for machines to have a better understanding of human communication. While language, visual and acoustic modalities can provide clues from different perspectives, the visual modality is shown to make minimal contribution to the performance in the emotion recognition field due to its high dimensionality. Therefore, we first leverage the strong multimodality backbone VATT to project the visual signal to the common space with language and acoustic signals. Also, we propose content-oriented features Topic and Speaking style on top of it to approach the subjectivity issues. Experiments conducted on the benchmark dataset MOSEI show our model can outperform SOTA results and effectively incorporate visual signals and handle subjectivity issues by serving as content “normalization”.

In recent years, language models have become increasingly larger and more complex. However, the input representations for these models continue to rely on simple and greedy subword tokenization methods. In this paper, we propose a novel tokenization method that factorizes subwords onto discrete triplets using a VQ-VAE model. The effectiveness of the proposed tokenization method, referred to as the Factorizer, is evaluated on language modeling and morpho-syntactic tasks for 7 diverse languages. Results indicate that this method is more appropriate and robust for morphological tasks than the commonly used byte-pair encoding (BPE) tokenization algorithm.

Subword tokenization is a key part of most NLP pipelines. However, little is known about why some tokenizer and hyperparameter combinations lead to improved downstream model performance over others. We propose that good tokenizers lead to efficient channel usage, where the channel is the means by which some input is conveyed to the model and efficiency can be quantified in information-theoretic terms as the ratio of the Shannon entropy to the maximum entropy of the subword distribution. Nevertheless, an optimal encoding according to Shannon entropy assigns extremely long codes to low-frequency subwords and very short codes to high-frequency subwords.Defining efficiency in terms of Rényi entropy, on the other hand, penalizes distributions with either very high or very low-frequency subwords.We posit that (1) extremely high-frequency subwords are problematic because their meaning is not distinct and (2) that low-frequency subwords may not appear frequently enough for their meaning to be learned properly; encodings that induce unigram distributions with either can harm model performance. In machine translation, we find that across multiple tokenizers, the Rényi entropy has a very strong correlation with BLEU: 0.82 in comparison to just -0.30 for compressed length.

Multilingual language models have recently gained attention as a promising solution for representing multiple languages in a single model. In this paper, we propose new criteria to evaluate the quality of lexical representation and vocabulary overlap observed in sub-word tokenizers.Our findings show that the overlap of vocabulary across languages can be actually detrimental to certain downstream tasks (POS, dependency tree labeling). In contrast, NER and sentence-level tasks (cross-lingual retrieval, NLI) benefit from sharing vocabulary. We also observe that the coverage of the language-specific tokens in the multilingual vocabulary significantly impacts the word-level tasks. Our study offers a deeper understanding of the role of tokenizers in multilingual language models and guidelines for future model developers to choose the most suitable tokenizer for their specific application before undertaking costly model pre-training.

While there is much recent interest in studying why Transformer-based large language models make predictions the way they do, the complex computations performed within each layer have made their behavior somewhat opaque. To mitigate this opacity, this work presents a linear decomposition of final hidden states from autoregressive language models based on each initial input token, which is exact for virtually all contemporary Transformer architectures. This decomposition allows the definition of probability distributions that ablate the contribution of specific input tokens, which can be used to analyze their influence on model probabilities over a sequence of upcoming words with only one forward pass from the model. Using the change in next-word probability as a measure of importance, this work first examines which context words make the biggest contribution to language model predictions. Regression experiments suggest that Transformer-based language models rely primarily on collocational associations, followed by linguistic factors such as syntactic dependencies and coreference relationships in making next-word predictions. Additionally, analyses using these measures to predict syntactic dependencies and coreferent mention spans show that collocational association and repetitions of the same token largely explain the language models’ predictions on these tasks.

Adaptive training approaches, widely used in sequence-to-sequence models, commonly reweigh the losses of different target tokens based on priors, e.g. word frequency. However, most of them do not consider the variation of learning difficulty in different training steps, and overly emphasize the learning of difficult one-hot labels, making the learning deterministic and sub-optimal. In response, we present Token-Level Self-Evolution Training (SE), a simple and effective dynamic training method to fully and wisely exploit the knowledge from data. SE focuses on dynamically learning the under-explored tokens for each forward pass and adaptively regularizes the training by introducing a novel token-specific label smoothing approach. Empirically, SE yields consistent and significant improvements in three tasks, i.e. machine translation, summarization, and grammatical error correction. Encouragingly, we achieve averaging +0.93 BLEU improvement on three machine translation tasks. Analyses confirm that, besides improving lexical accuracy, SE enhances generation diversity and model generalization.

Sense embedding learning methods learn multiple vectors for a given ambiguous word, corresponding to its different word senses. For this purpose, different methods have been proposed in prior work on sense embedding learning that use different sense inventories, sense-tagged corpora and learning methods. However, not all existing sense embeddings cover all senses of ambiguous words equally well due to the discrepancies in their training resources. To address this problem, we propose the first-ever meta-sense embedding method – Neighbour Preserving Meta-Sense Embeddings, which learns meta-sense embeddings by combining multiple independently trained source sense embeddings such that the sense neighbourhoods computed from the source embeddings are preserved in the meta-embedding space. Our proposed method can combine source sense embeddings that cover different sets of word senses. Experimental results on Word Sense Disambiguation (WSD) and Word-in-Context (WiC) tasks show that the proposed meta-sense embedding method consistently outperforms several competitive baselines. An anonymised version of the source code implementation for our proposed method is submitted to reviewing system. Both source code and the learnt meta-sense embeddings will be publicly released upon paper acceptance.

Optimizing the phrasing of argumentative text is crucial in higher education and professional development. However, assessing whether and how the different claims in a text should be revised is a hard task, especially for novice writers. In this work, we explore the main challenges to identifying argumentative claims in need of specific revisions. By learning from collaborative editing behaviors in online debates, we seek to capture implicit revision patterns in order to develop approaches aimed at guiding writers in how to further improve their arguments. We systematically compare the ability of common word embedding models to capture the differences between different versions of the same text, and we analyze their impact on various types of writing issues. To deal with the noisy nature of revision-based corpora, we propose a new sampling strategy based on revision distance. Opposed to approaches from prior work, such sampling can be done without employing additional annotations and judgments. Moreover, we provide evidence that using contextual information and domain knowledge can further improve prediction results. How useful a certain type of context is, depends on the issue the claim is suffering from, though.

Embedding models have shown great power in knowledge graph completion (KGC) task. By learning structural constraints for each training triple, these methods implicitly memorize intrinsic relation rules to infer missing links. However, this paper points out that the multi-hop relation rules are hard to be reliably memorized due to the inherent deficiencies of such implicit memorization strategy, making embedding models underperform in predicting links between distant entity pairs. To alleviate this problem, we present Vertical Learning Paradigm (VLP), which extends embedding models by allowing to explicitly copy target information from related factual triples for more accurate prediction. Rather than solely relying on the implicit memory, VLP directly provides additional cues to improve the generalization ability of embedding models, especially making the distant link prediction significantly easier. Moreover, we also propose a novel relative distance based negative sampling technique (ReD) for more effective optimization. Experiments demonstrate the validity and generality of our proposals on two standard benchmarks. Our code is available at https://github.com/rui9812/VLP.

Recent advances in open-domain question answering (ODQA) have demonstrated impressive accuracy on general-purpose domains like Wikipedia. While some work has been investigating how well ODQA models perform when tested for out-of-domain (OOD) generalization, these studies have been conducted only under conservative shifts in data distribution and typically focus on a single component (i.e., retriever or reader) rather than an end-to-end system. This work proposes a more realistic end-to-end domain shift evaluation setting covering five diverse domains. We not only find that end-to-end models fail to generalize but that high retrieval scores often still yield poor answer prediction accuracy. To address these failures, we investigate several interventions, in the form of data augmentations, for improving model adaption and use our evaluation set to elucidate the relationship between the efficacy of an intervention scheme and the particular type of dataset shifts we consider. We propose a generalizability test that estimates the type of shift in a target dataset without training a model in the target domain and that the type of shift is predictive of which data augmentation schemes will be effective for domain adaption. Overall, we find that these interventions increase end-to-end performance by up to ~24 points.

Lifelogs are descriptions of experiences that a person had during their life. Lifelogs are created by fusing data from the multitude of digital services, such as online photos, maps, shopping and content streaming services. Question answering over lifelogs can offer personal assistants a critical resource when they try to provide advice in context. However, obtaining answers to questions over lifelogs is beyond the current state of the art of question answering techniques for a variety of reasons, the most pronounced of which is that lifelogs combine free text with some degree of structure such as temporal and geographical information. We create and publicly release TimelineQA, a benchmark for accelerating progress on querying lifelogs. TimelineQA generates lifelogs of imaginary people. The episodes in the lifelog range from major life episodes such as high school graduation to those that occur on a daily basis such as going for a run. We describe a set of experiments on TimelineQA with several state-of-the-art QA models. Our experiments reveal that for atomic queries, an extractive QA system significantly out-performs a state-of-the-art retrieval-augmented QA system. For multi-hop queries involving aggregates, we show that the best result is obtained with a state-of-the-art table QA technique, assuming the ground truth set of episodes for deriving the answer is available.

Bias research in NLP seeks to analyse models for social biases, thus helping NLP practitioners uncover, measure, and mitigate social harms. We analyse the body of work that uses prompts and templates to assess bias in language models. We draw on a measurement modelling framework to create a taxonomy of attributes that capture what a bias test aims to measure and how that measurement is carried out. By applying this taxonomy to 90 bias tests, we illustrate qualitatively and quantitatively that core aspects of bias test conceptualisations and operationalisations are frequently unstated or ambiguous, carry implicit assumptions, or be mismatched. Our analysis illuminates the scope of possible bias types the field is able to measure, and reveals types that are as yet under-researched. We offer guidance to enable the community to explore a wider section of the possible bias space, and to better close the gap between desired outcomes and experimental design, both for bias and for evaluating language models more broadly.