ACL 2023 论文列表

Pre-trained language models (PLM) have made impressive results in a wide range of NLP tasks and it has been revealed that one of the key factors to their success is the parameters of these models implicitly learn various types of knowledge in the pre-training corpus. However, encoding knowledge implicitly in the model parameters has two fundamental drawbacks. First, the knowledge is neither editable nor scalable once the model is trained, which is especially problematic in that knowledge is consistently evolving. Second, it lacks interpretability and prevents us from understanding what kind of knowledge PLM needs to solve a certain task. In this paper, we introduce {pasted macro ‘MODEL’}, a pre-training model with differentiable plug-in memory (DPM). The key intuition behind is to decouple the knowledge storage from model parameters with an editable and scalable key-value memory and leverage knowledge in an explainable manner by knowledge retrieval in the {pasted macro ‘MEMORY’}. We conduct extensive experiments under various settings to justify this design choice. In domain adaptation setting, {pasted macro ‘MODEL’} could be easily adapted to different domains with pluggable in-domain memory—obtaining 3.95 F1 improvements across four domains, without any in-domain training. {pasted macro ‘MODEL’} could also keep absorbing new knowledge after pre-training is done by knowledge updating operation in the {pasted macro ‘MEMORY’} without re-training. Finally, we show that by incorporating training samples into {pasted macro ‘MEMORY’} with knowledge prompting, {pasted macro ‘MODEL’} could further be improved by the instruction of in-task knowledge.

In statistical parsing with CCG, the standard evaluation method is based on predicate-argument structure and evaluates dependencies labelled in part by lexical categories. When a predicate has multiple argument slots that can be filled, the same lexical category is used for the label of multiple dependencies. In this paper, we show that this evaluation can result in disproportionate penalization of supertagging errors and obfuscate the truly erroneous dependencies. Enabled by the compositional nature of CCG lexical categories, we propose *decomposed scoring* based on subcategorial labels to address this. To evaluate our scoring method, we engage fellow categorial grammar researchers in two English-language judgement tasks: (1) directly ranking the outputs of the standard and experimental scoring methods; and (2) determining which of two sentences has the better parse in cases where the two scoring methods disagree on their ranks. Overall, the judges prefer decomposed scoring in each task; but there is substantial disagreement among the judges in 24% of the given cases, pointing to potential issues with parser evaluations in general.

An emerging solution for explaining Transformer-based models is to use vector-based analysis on how the representations are formed. However, providing a faithful vector-based explanation for a multi-layer model could be challenging in three aspects: (1) Incorporating all components into the analysis, (2) Aggregating the layer dynamics to determine the information flow and mixture throughout the entire model, and (3) Identifying the connection between the vector-based analysis and the model’s predictions. In this paper, we present DecompX to tackle these challenges. DecompX is based on the construction of decomposed token representations and their successive propagation throughout the model without mixing them in between layers. Additionally, our proposal provides multiple advantages over existing solutions for its inclusion of all encoder components (especially nonlinear feed-forward networks) and the classification head. The former allows acquiring precise vectors while the latter transforms the decomposition into meaningful prediction-based values, eliminating the need for norm- or summation-based vector aggregation. According to the standard faithfulness evaluations, DecompX consistently outperforms existing gradient-based and vector-based approaches on various datasets. Our code is available at https://github.com/mohsenfayyaz/DecompX.

Existing evaluation metrics for natural language generation (NLG) tasks face the challenges on generalization ability and interpretability. Specifically, most of the well-performed metrics are required to train on evaluation datasets of specific NLG tasks and evaluation dimensions, which may cause over-fitting to task-specific datasets. Furthermore, existing metrics only provide an evaluation score for each dimension without revealing the evidence to interpret how this score is obtained. To deal with these challenges, we propose a simple yet effective metric called DecompEval. This metric formulates NLG evaluation as an instruction-style question answering task and utilizes instruction-tuned pre-trained language models (PLMs) without training on evaluation datasets, aiming to enhance the generalization ability. To make the evaluation process more interpretable, we decompose our devised instruction-style question about the quality of generated texts into the subquestions that measure the quality of each sentence. The subquestions with their answers generated by PLMs are then recomposed as evidence to obtain the evaluation result. Experimental results show that DecompEval achieves state-of-the-art performance in untrained metrics for evaluating text summarization and dialogue generation, which also exhibits strong dimension-level / task-level generalization ability and interpretability.

This work explores the feasibility of eliciting knowledge from language models (LMs) to decode symbolism, recognizing something (e.g.,roses) as a stand-in for another (e.g., love). We present our evaluative framework, Symbolism Analysis (SymbA), which compares LMs (e.g., RoBERTa, GPT-J) on different types of symbolism and analyze the outcomes along multiple metrics. Our findings suggest that conventional symbols are more reliably elicited from LMs while situated symbols are more challenging. Results also reveal the negative impact of the bias in pre-trained corpora. We further demonstrate that a simple re-ranking strategy can mitigate the bias and significantly improve model performances to be on par with human performances in some cases.

With the evergrowing sizes of pre-trained models (PTMs), it has been an emerging practice to only provide the inference APIs for users, namely model-as-a-service (MaaS) setting. To adapt PTMs with model parameters frozen, most current approaches focus on the input side, seeking powerful prompts to stimulate models for correct answers. However, we argue that input-side adaptation could be arduous due to the lack of gradient signals and they usually require thousands of API queries, resulting in high computation and time costs. Specifically, DecT first extracts prompt-stimulated output scores for initial predictions. On top of that, we train an additional decoder network on the output representations to incorporate posterior data knowledge. By gradient-based optimization, DecT can be trained within several seconds and requires only one PTM query per sample. Empirically, we conduct extensive natural language understanding experiments and show that DecT significantly outperforms state-of-the-art algorithms with a 200x speed-up. Our code is available at https://github.com/thunlp/DecT.

Commonsense fact verification, as a challenging branch of commonsense question-answering (QA), aims to verify through facts whether a given commonsense claim is correct or not. Answering commonsense questions necessitates a combination of knowledge from various levels. However, existing studies primarily rest on grasping either unstructured evidence or potential reasoning paths from structured knowledge bases, yet failing to exploit the benefits of heterogeneous knowledge simultaneously. In light of this, we propose Decker, a commonsense fact verification model that is capable of bridging heterogeneous knowledge by uncovering latent relationships between structured and unstructured knowledge. Experimental results on two commonsense fact verification benchmark datasets, CSQA2.0 and CREAK demonstrate the effectiveness of our Decker and further analysis verifies its capability to seize more precious information through reasoning. The official implementation of Decker is available at https://github.com/Anni-Zou/Decker.

Debiasing methods that seek to mitigate the tendency of Language Models (LMs) to occasionally output toxic or inappropriate text have recently gained traction. In this paper, we propose a standardized protocol which distinguishes methods that yield not only desirable results, but are also consistent with their mechanisms and specifications. For example, we ask, given a debiasing method that is developed to reduce toxicity in LMs, if the definition of toxicity used by the debiasing method is reversed, would the debiasing results also be reversed? We used such considerations to devise three criteria for our new protocol: Specification Polarity, Specification Importance, and Domain Transferability. As a case study, we apply our protocol to a popular debiasing method, Self-Debiasing, and compare it to one we propose, called Instructive Debiasing, and demonstrate that consistency is as important an aspect to debiasing viability as is simply a desirable result. We show that our protocol provides essential insights into the generalizability and interpretability of debiasing methods that may otherwise go overlooked.

Recognizing flat, overlapped and discontinuous entities uniformly has been paid increasing attention. Among these works, Seq2Seq formulation prevails for its flexibility and effectiveness. It arranges the output entities into a specific target sequence. However, it introduces bias by assigning all the probability mass to the observed sequence. To alleviate the bias, previous works either augment the data with possible sequences or resort to other formulations. In this paper, we stick to the Seq2Seq formulation and propose a reranking-based approach. It redistributes the likelihood among candidate sequences depending on their performance via a contrastive loss. Extensive experiments show that our simple yet effective method consistently boosts the baseline, and yields competitive or better results compared with the state-of-the-art methods on 8 widely-used datasets for Named Entity Recognition.

Intelligent systems that aim at mastering language as humans do must deal with its semantic underspecification, namely, the possibility for a linguistic signal to convey only part of the information needed for communication to succeed. Consider the usages of the pronoun they, which can leave the gender and number of its referent(s) underspecified. Semantic underspecification is not a bug but a crucial language feature that boosts its storage and processing efficiency. Indeed, human speakers can quickly and effortlessly integrate semantically-underspecified linguistic signals with a wide range of non-linguistic information, e.g., the multimodal context, social or cultural conventions, and shared knowledge. Standard NLP models have, in principle, no or limited access to such extra information, while multimodal systems grounding language into other modalities, such as vision, are naturally equipped to account for this phenomenon. However, we show that they struggle with it, which could negatively affect their performance and lead to harmful consequences when used for applications. In this position paper, we argue that our community should be aware of semantic underspecification if it aims to develop language technology that can successfully interact with human users. We discuss some applications where mastering it is crucial and outline a few directions toward achieving this goal.

Visual language such as charts and plots is ubiquitous in the human world. Comprehending plots and charts requires strong reasoning skills. Prior state-of-the-art (SOTA) models require at least tens of thousands of training examples and their reasoning capabilities are still much limited, especially on complex human-written queries. This paper presents the first one-shot solution to visual language reasoning. We decompose the challenge of visual language reasoning into two steps: (1) plot-to-text translation, and (2) reasoning over the translated text. The key in this method is a modality conversion module, named as DePlot, which translates the image of a plot or chart to a linearized table. The output of DePlot can then be directly used to prompt a pretrained large language model (LLM), exploiting the few-shot reasoning capabilities of LLMs. To obtain DePlot, we standardize the plot-to-table task by establishing unified task formats and metrics, and train DePlot end-to-end on this task. DePlot can then be used off-the-shelf together with LLMs in a plug-and-play fashion. Compared with a SOTA model finetuned on more than thousands of data points, DePlot+LLM with just one-shot prompting achieves a 29.4% improvement over finetuned SOTA on human-written queries from the task of chart QA.

Dating Greek papyri accurately is crucial not only to edit their texts but also to understand numerous other aspects of ancient writing, document and book production and circulation, as well as various other aspects of administration, everyday life and intellectual history of antiquity. Although a substantial number of Greek papyri documents bear a date or other conclusive data as to their chronological placement, an even larger number can only be dated tentatively or in approximation, due to the lack of decisive evidence. By creating a dataset of 389 transcriptions of documentary Greek papyri, we train 389 regression models and we predict a date for the papyri with an average MAE of 54 years and an MSE of 1.17, outperforming image classifiers and other baselines. Last, we release date estimations for 159 manuscripts, for which only the upper limit is known.

Dataset distillation aims to create a small dataset of informative synthetic samples to rapidly train neural networks that retain the performance of the original dataset. In this paper, we focus on constructing distilled few-shot datasets for natural language processing (NLP) tasks to fine-tune pre-trained transformers. Specifically, we propose to introduce attention labels, which can efficiently distill the knowledge from the original dataset and transfer it to the transformer models via attention probabilities. We evaluated our dataset distillation methods in four various NLP tasks and demonstrated that it is possible to create distilled few-shot datasets with the attention labels, yielding impressive performances for fine-tuning BERT. Specifically, in AGNews, a four-class news classification task, our distilled few-shot dataset achieved up to 93.2% accuracy, which is 98.5% performance of the original dataset even with only one sample per class and only one gradient step.

Modern machine learning relies on datasets to develop and validate research ideas. Given the growth of publicly available data, finding the right dataset to use is increasingly difficult. Any research question imposes explicit and implicit constraints on how well a given dataset will enable researchers to answer this question, such as dataset size, modality, and domain. We operationalize the task of recommending datasets given a short natural language description of a research idea, to help people find relevant datasets for their needs. Dataset recommendation poses unique challenges as an information retrieval problem; datasets are hard to directly index for search and there are no corpora readily available for this task. To facilitate this task, we build the DataFinder Dataset which consists of a larger automatically-constructed training set (17.5K queries) and a smaller expert-annotated evaluation set (392 queries). Using this data, we compare various information retrieval algorithms on our test set and present a superior bi-encoder retriever for text-based dataset recommendation. This system, trained on the DataFinder Dataset, finds more relevant search results than existing third-party dataset search engines. To encourage progress on dataset recommendation, we release our dataset and models to the public.

Recent advances in NLP have significantly improved the performance of language models on a variety of tasks. While these advances are largely driven by the availability of large amounts of data and computational power, they also benefit from the development of better training methods and architectures. In this paper, we introduce CamemBERTa, a French DeBERTa model that builds upon the DeBERTaV3 architecture and training objective. We evaluate our model’s performance on a variety of French downstream tasks and datasets, including question answering, part-of-speech tagging, dependency parsing, named entity recognition, and the FLUE benchmark, and compare against CamemBERT, the state-of-the-art monolingual model for French. Our results show that, given the same amount of training tokens, our model outperforms BERT-based models trained with MLM on most tasks. Furthermore, our new model reaches similar or superior performance on downstream tasks compared to CamemBERT, despite being trained on only 30% of its total number of input tokens. In addition to our experimental results, we also publicly release the weights and code implementation of CamemBERTa, making it the first publicly available DeBERTaV3 model outside of the original paper and the first openly available implementation of a DeBERTaV3 training objective.

Obtaining labeled data to train a model for a task of interest is often expensive. Prior work shows training models on multitask data augmented with task descriptions (prompts) effectively transfers knowledge to new tasks. Towards efficiently building task-specific models, we assume access to a small number (32-1000) of unlabeled target-task examples and use those to retrieve the most similar labeled examples from a large pool of multitask data augmented with prompts. Compared to the current practice of finetuning models on uniformly sampled prompted multitask data (e.g.: FLAN, T0), our approach of finetuning on cross-task nearest neighbors is significantly more data-efficient. Using only 2% of the data from the P3 pool without any labeled target-task data, our models outperform strong baselines trained on all available data by 3-30% on 12 out of 14 datasets representing held-out tasks including legal and scientific document QA. Similarly, models trained on cross-task nearest neighbors from SuperNaturalInstructions, representing about 5% of the pool, obtain comparable performance to state-of-the-art models on 12 held-out tasks from that pool. Moreover, the models produced by our approach also provide a better initialization than single multitask finetuned models for few-shot finetuning on target-task data, as shown by a 2-23% relative improvement over few-shot finetuned T0-3B models on 8 datasets.

We analyze the different data sampling approaches used in selecting data for human evaluation and ranking of machine translation systems at the highly influential Conference on Machine Translation (WMT). By using automatic evaluation metrics, we are able to focus on the impact of the data sampling procedure as separate from questions about human annotator consistency. We provide evidence that the latest data sampling approach used at WMT skews the annotated data toward shorter documents, not necessarily representative of the full test set. Lastly, we examine a new data sampling method that uses the available labour budget to sample data in a more representative manner, with the goals of improving representation of various document lengths in the sample and producing more stable rankings of system translation quality.

In-context learning (ICL) enables large language models (LLMs) to perform new tasks by prompting them with a sequence of training examples. However, it is known that ICL is very sensitive to the choice of training examples: randomly sampling examples from a training set leads to high variance in performance. In this paper, we show that carefully curating a subset of training data greatly stabilizes ICL performance without any other changes to the ICL algorithm (e.g., prompt retrieval or calibration). We introduce two methods to choose training subsets—both score training examples individually, then select the highest-scoring ones. CondAcc scores a training example by its average dev-set ICL accuracy when combined with random training examples, while Datamodels learns linear regressors that estimate how the presence of each training example influences LLM outputs. Across five tasks and two LLMs, sampling from stable subsets selected by CondAcc and Datamodels improves average accuracy over sampling from the entire training set by 7.7% and 6.3%, respectively. Surprisingly, the stable subset examples are not especially diverse in content or low in perplexity, in contrast with other work suggesting that diversity and perplexity are important when prompting LLMs.

Keyphrase generation is the task of summarizing the contents of any given article into a few salient phrases (or keyphrases). Existing works for the task mostly rely on large-scale annotated datasets, which are not easy to acquire. Very few works address the problem of keyphrase generation in low-resource settings, but they still rely on a lot of additional unlabeled data for pretraining and on automatic methods for pseudo-annotations. In this paper, we present data augmentation strategies specifically to address keyphrase generation in purely resource-constrained domains. We design techniques that use the full text of the articles to improve both present and absent keyphrase generation. We test our approach comprehensively on three datasets and show that the data augmentation strategies consistently improve the state-of-the-art performance. We release our source code at https://github.com/kgarg8/kpgen-lowres-data-aug.