ACL 2022 论文列表

Large pre-trained vision-language (VL) models can learn a new task with a handful of examples and generalize to a new task without fine-tuning. However, these VL models are hard to deploy for real-world applications due to their impractically huge sizes and slow inference speed. To solve this limitation, we study prompt-based low-resource learning of VL tasks with our proposed method, FewVLM, relatively smaller than recent few-shot learners. For FewVLM, we pre-train a sequence-to-sequence transformer model with prefix language modeling (PrefixLM) and masked language modeling (MaskedLM).Furthermore, we analyze the effect of diverse prompts for few-shot tasks. Experimental results on VQA show that FewVLM with prompt-based learning outperforms Frozen which is 31x larger than FewVLM by 18.2% point and achieves comparable results to a 246x larger model, PICa.In our analysis, we observe that (1) prompts significantly affect zero-shot performance but marginally affect few-shot performance, (2) models with noisy prompts learn as quickly as hand-crafted prompts given larger training data, and (3) MaskedLM helps VQA tasks while PrefixLM boosts captioning performance. Our code is publicly available at https://github.com/woojeongjin/FewVLM

We present an efficient BERT-based multi-task (MT) framework that is particularly suitable for iterative and incremental development of the tasks. The proposed framework is based on the idea of partial fine-tuning, i.e. only fine-tune some top layers of BERT while keep the other layers frozen. For each task, we train independently a single-task (ST) model using partial fine-tuning. Then we compress the task-specific layers in each ST model using knowledge distillation. Those compressed ST models are finally merged into one MT model so that the frozen layers of the former are shared across the tasks. We exemplify our approach on eight GLUE tasks, demonstrating that it is able to achieve 99.6% of the performance of the full fine-tuning method, while reducing up to two thirds of its overhead.

Previous attempts to build effective semantic parsers for Wizard-of-Oz (WOZ) conversations suffer from the difficulty in acquiring a high-quality, manually annotated training set. Approaches based only on dialogue synthesis are insufficient, as dialogues generated from state-machine based models are poor approximations of real-life conversations. Furthermore, previously proposed dialogue state representations are ambiguous and lack the precision necessary for building an effective agent. This paper proposes a new dialogue representation and a sample-efficient methodology that can predict precise dialogue states in WOZ conversations. We extended the ThingTalk representation to capture all information an agent needs to respond properly. Our training strategy is sample-efficient: we combine (1) few-shot data sparsely sampling the full dialogue space and (2) synthesized data covering a subset space of dialogues generated by a succinct state-based dialogue model. The completeness of the extended ThingTalk language is demonstrated with a fully operational agent, which is also used in training data synthesis. We demonstrate the effectiveness of our methodology on MultiWOZ 3.0, a reannotation of the MultiWOZ 2.1 dataset in ThingTalk. ThingTalk can represent 98% of the test turns, while the simulator can emulate 85% of the validation set. We train a contextual semantic parser using our strategy, and obtain 79% turn-by-turn exact match accuracy on the reannotated test set.

We conduct a feasibility study into the applicability of answer-agnostic question generation models to textbook passages. We show that a significant portion of errors in such systems arise from asking irrelevant or un-interpretable questions and that such errors can be ameliorated by providing summarized input. We find that giving these models human-written summaries instead of the original text results in a significant increase in acceptability of generated questions (33% → 83%) as determined by expert annotators. We also find that, in the absence of human-written summaries, automatic summarization can serve as a good middle ground.

Open-domain question answering is a challenging task with a wide variety of practical applications. Existing modern approaches mostly follow a standard two-stage paradigm: retriever then reader. In this article, we focus on improving the effectiveness of the reader module and propose a novel copy-augmented generative approach that integrates the merits of both extractive and generative readers. In particular, our model is built upon the powerful generative model FiD (CITATION). We enhance the original generative reader by incorporating a pointer network to encourage the model to directly copy words from the retrieved passages. We conduct experiments on the two benchmark datasets, Natural Questions and TriviaQA, and the empirical results demonstrate the performance gains of our proposed approach.

Learning high-quality sentence representations is a fundamental problem of natural language processing which could benefit a wide range of downstream tasks. Though the BERT-like pre-trained language models have achieved great success, using their sentence representations directly often results in poor performance on the semantic textual similarity task. Recently, several contrastive learning methods have been proposed for learning sentence representations and have shown promising results. However, most of them focus on the constitution of positive and negative representation pairs and pay little attention to the training objective like NT-Xent, which is not sufficient enough to acquire the discriminating power and is unable to model the partial order of semantics between sentences. So in this paper, we propose a new method ArcCSE, with training objectives designed to enhance the pairwise discriminative power and model the entailment relation of triplet sentences. We conduct extensive experiments which demonstrate that our approach outperforms the previous state-of-the-art on diverse sentence related tasks, including STS and SentEval.

Data sharing restrictions are common in NLP, especially in the clinical domain, but there is limited research on adapting models to new domains without access to the original training data, a setting known as source-free domain adaptation. We take algorithms that traditionally assume access to the source-domain training data—active learning, self-training, and data augmentation—and adapt them for source free domain adaptation. Then we systematically compare these different strategies across multiple tasks and domains. We find that active learning yields consistent gains across all SemEval 2021 Task 10 tasks and domains, but though the shared task saw successful self-trained and data augmented models, our systematic comparison finds these strategies to be unreliable for source-free domain adaptation.

Interpretable methods to reveal the internal reasoning processes behind machine learning models have attracted increasing attention in recent years. To quantify the extent to which the identified interpretations truly reflect the intrinsic decision-making mechanisms, various faithfulness evaluation metrics have been proposed. However, we find that different faithfulness metrics show conflicting preferences when comparing different interpretations. Motivated by this observation, we aim to conduct a comprehensive and comparative study of the widely adopted faithfulness metrics. In particular, we introduce two assessment dimensions, namely diagnosticity and complexity. Diagnosticity refers to the degree to which the faithfulness metric favors relatively faithful interpretations over randomly generated ones, and complexity is measured by the average number of model forward passes. According to the experimental results, we find that sufficiency and comprehensiveness metrics have higher diagnosticity and lower complexity than the other faithfulness metrics.

Given the prevalence of pre-trained contextualized representations in today’s NLP, there have been many efforts to understand what information they contain, and why they seem to be universally successful. The most common approach to use these representations involves fine-tuning them for an end task. Yet, how fine-tuning changes the underlying embedding space is less studied. In this work, we study the English BERT family and use two probing techniques to analyze how fine-tuning changes the space. We hypothesize that fine-tuning affects classification performance by increasing the distances between examples associated with different labels. We confirm this hypothesis with carefully designed experiments on five different NLP tasks. Via these experiments, we also discover an exception to the prevailing wisdom that “fine-tuning always improves performance”. Finally, by comparing the representations before and after fine-tuning, we discover that fine-tuning does not introduce arbitrary changes to representations; instead, it adjusts the representations to downstream tasks while largely preserving the original spatial structure of the data points.

In the domain of Morphology, Inflection is a fundamental and important task that gained a lot of traction in recent years, mostly via SIGMORPHON’s shared-tasks. With average accuracy above 0.9 over the scores of all languages, the task is considered mostly solved using relatively generic neural seq2seq models, even with little data provided. In this work, we propose to re-evaluate morphological inflection models by employing harder train-test splits that will challenge the generalization capacity of the models. In particular, as opposed to the naïve split-by-form, we propose a split-by-lemma method to challenge the performance on existing benchmarks. Our experiments with the three top-ranked systems on the SIGMORPHON’s 2020 shared-task show that the lemma-split presents an average drop of 30 percentage points in macro-average for the 90 languages included. The effect is most significant for low-resourced languages with a drop as high as 95 points, but even high-resourced languages lose about 10 points on average. Our results clearly show that generalizing inflection to unseen lemmas is far from being solved, presenting a simple yet effective means to promote more sophisticated models.