NAACL 2022 论文列表

Theoretical work in morphological typology offers the possibility of measuring morphological diversity on a continuous scale. However, literature in Natural Language Processing (NLP) typically labels a whole language with a strict type of morphology, e.g. fusional or agglutinative. In this work, we propose to reduce the rigidity of such claims, by quantifying morphological typology at the word and segment level. We consider Payne (2017)’s approach to classify morphology using two indices: synthesis (e.g. analytic to polysynthetic) and fusion (agglutinative to fusional). For computing synthesis, we test unsupervised and supervised morphological segmentation methods for English, German and Turkish, whereas for fusion, we propose a semi-automatic method using Spanish as a case study. Then, we analyse the relationship between machine translation quality and the degree of synthesis and fusion at word (nouns and verbs for English-Turkish, and verbs in English-Spanish) and segment level (previous language pairs plus English-German in both directions). We complement the word-level analysis with human evaluation, and overall, we observe a consistent impact of both indexes on machine translation quality.

Deep acoustic models represent linguistic information based on massive amounts of data. Unfortunately, for regional languages and dialects such resources are mostly not available. However, deep acoustic models might have learned linguistic information that transfers to low-resource languages. In this study, we evaluate whether this is the case through the task of distinguishing low-resource (Dutch) regional varieties. By extracting embeddings from the hidden layers of various wav2vec 2.0 models (including new models which are pre-trained and/or fine-tuned on Dutch) and using dynamic time warping, we compute pairwise pronunciation differences averaged over 10 words for over 100 individual dialects from four (regional) languages. We then cluster the resulting difference matrix in four groups and compare these to a gold standard, and a partitioning on the basis of comparing phonetic transcriptions. Our results show that acoustic models outperform the (traditional) transcription-based approach without requiring phonetic transcriptions, with the best performance achieved by the multilingual XLSR-53 model fine-tuned on Dutch. On the basis of only six seconds of speech, the resulting clustering closely matches the gold standard.

When a neural language model (LM) is adapted to perform a new task, what aspects of the task predict the eventual performance of the model? In NLP, systematic features of LM generalization to individual examples are well characterized, but systematic aspects of LM adaptability to new tasks are not nearly as well understood. We present a large-scale empirical study of the features and limits of LM adaptability using a new benchmark, TaskBench500, built from 500 procedurally generated sequence modeling tasks. These tasks combine core aspects of language processing, including lexical semantics, sequence processing, memorization, logical reasoning, and world knowledge. Using TaskBench500, we evaluate three facets of adaptability, finding that: (1) adaptation procedures differ dramatically in their ability to memorize small datasets; (2) within a subset of task types, adaptation procedures exhibit compositional adaptability to complex tasks; and (3) failure to match training label distributions is explained by mismatches in the intrinsic difficulty of predicting individual labels. Our experiments show that adaptability to new tasks, like generalization to new examples, can be systematically described and understood, and we conclude with a discussion of additional aspects of adaptability that could be studied using the new benchmark.

Despite the progress in machine translation quality estimation and evaluation in the last years, decoding in neural machine translation (NMT) is mostly oblivious to this and centers around finding the most probable translation according to the model (MAP decoding), approximated with beam search. In this paper, we bring together these two lines of research and propose quality-aware decoding for NMT, by leveraging recent breakthroughs in reference-free and reference-based MT evaluation through various inference methods like N-best reranking and minimum Bayes risk decoding. We perform an extensive comparison of various possible candidate generation and ranking methods across four datasets and two model classes and find that quality-aware decoding consistently outperforms MAP-based decoding according both to state-of-the-art automatic metrics (COMET and BLEURT) and to human assessments.

To enable building and testing models on long-document comprehension, we introduce QuALITY, a multiple-choice QA dataset with context passages in English that have an average length of about 5,000 tokens, much longer than typical current models can process. Unlike in prior work with passages, our questions are written and validated by contributors who have read the entire passage, rather than relying on summaries or excerpts. In addition, only half of the questions are answerable by annotators working under tight time constraints, indicating that skimming and simple search are not enough to consistently perform well. Our baseline models perform poorly on this task (55.4%) and significantly lag behind human performance (93.5%).

Synthetic datasets have successfully been used to probe visual question-answering datasets for their reasoning abilities. CLEVR (John- son et al., 2017), for example, tests a range of visual reasoning abilities. The questions in CLEVR focus on comparisons of shapes, colors, and sizes, numerical reasoning, and existence claims. This paper introduces a minimally biased, diagnostic visual question-answering dataset, QLEVR, that goes beyond existential and numerical quantification and focus on more complex quantifiers and their combinations, e.g., asking whether there are more than two red balls that are smaller than at least three blue balls in an image. We describe how the dataset was created and present a first evaluation of state-of-the-art visual question-answering models, showing that QLEVR presents a formidable challenge to our current models. Code and Dataset are available at https://github.com/zechenli03/QLEVR

Factual consistency is an essential quality of text summarization models in practical settings. Existing work in evaluating this dimension can be broadly categorized into two lines of research, entailment-based and question answering (QA)-based metrics, and different experimental setups often lead to contrasting conclusions as to which paradigm performs the best. In this work, we conduct an extensive comparison of entailment and QA-based metrics, demonstrating that carefully choosing the components of a QA-based metric, especially question generation and answerability classification, is critical to performance. Building on those insights, we propose an optimized metric, which we call QAFactEval, that leads to a 14% average improvement over previous QA-based metrics on the SummaC factual consistency benchmark, and also outperforms the best-performing entailment-based metric. Moreover, we find that QA-based and entailment-based metrics can offer complementary signals and be combined into a single metric for a further performance boost.

We present a system for document retrieval that combines direct classification with standard content-based retrieval approaches to significantly improve the relevance of the retrieved documents. Our system exploits the availability of an imperfect but sizable amount of labeled data from past queries. For domains such as technical support, the proposed approach enhances the system’s ability to retrieve documents that are otherwise ranked very low based on content alone. The system is easy to implement and can make use of existing text ranking methods, augmenting them through the novel Q2R orchestration framework. Q2R has been extensively tested and is in use at IBM.

While neural networks demonstrate a remarkable ability to model linguistic content, capturing contextual information related to a speaker’s conversational role is an open area of research. In this work, we analyze the effect of speaker role on language use through the game of Mafia, in which participants are assigned either an honest or a deceptive role. In addition to building a framework to collect a dataset of Mafia game records, we demonstrate that there are differences in the language produced by players with different roles. We confirm that classification models are able to rank deceptive players as more suspicious than honest ones based only on their use of language. Furthermore, we show that training models on two auxiliary tasks outperforms a standard BERT-based text classification approach. We also present methods for using our trained models to identify features that distinguish between player roles, which could be used to assist players during the Mafia game.

Inspired by human fact checkers, who use different types of evidence (e.g. tables, images, audio) in addition to text, several datasets with tabular evidence data have been released in recent years. Whilst the datasets encourage research on table fact-checking, they rely on information from restricted data sources, such as Wikipedia for creating claims and extracting evidence data, making the fact-checking process different from the real-world process used by fact checkers. In this paper, we introduce PubHealthTab, a table fact-checking dataset based on real world public health claims and noisy evidence tables from sources similar to those used by real fact checkers. We outline our approach for collecting evidence data from various websites and present an in-depth analysis of our dataset. Finally, we evaluate state-of-the-art table representation and pre-trained models fine-tuned on our dataset, achieving an overall F1 score of 0.73.

Pre-trained language models have shown great success in multiple downstream tasks. However, they are computationally expensive to fine-tune. Thus, transfer learning with adapter modules has been introduced to alleviate this problem, helping to extract knowledge of the downstream tasks. Adapterfusion models are an example of the transformers-with-adapter-modules, which merge multiple adapters to incorporate knowledge from different tasks. However, merging multiple adapters will inevitably cause redundancies, increasing the training and inference time massively. Therefore, in this paper, we propose an approach to identify the influence of each adapter module and a novel way to prune adapters based on the prestigious Lottery Ticket Hypothesis. Experiments on GLUE datasets show that the pruned Adapterfusion model with our scheme can achieve state-of-the-art results, reducing sizes significantly while keeping performance intact.

Large language models are shown to memorize privacy information such as social security numbers in training data. Given the sheer scale of the training corpus, it is challenging to screen and filter these privacy data, either manually or automatically. In this paper, we propose Confidentially Redacted Training (CRT), a method to train language generation models while protecting the confidential segments. We borrow ideas from differential privacy (which solves a related but distinct problem) and show that our method is able to provably prevent unintended memorization by randomizing parts of the training process. Moreover, we show that redaction with an approximately correct screening policy amplifies the confidentiality guarantee. We implement the method for both LSTM and GPT language models. Our experimental results show that the models trained by CRT obtain almost the same perplexity while preserving strong confidentiality.

Text clustering methods were traditionally incorporated into multi-document summarization (MDS) as a means for coping with considerable information repetition. Particularly, clusters were leveraged to indicate information saliency as well as to avoid redundancy. Such prior methods focused on clustering sentences, even though closely related sentences usually contain also non-aligned parts. In this work, we revisit the clustering approach, grouping together sub-sentential propositions, aiming at more precise information alignment. Specifically, our method detects salient propositions, clusters them into paraphrastic clusters, and generates a representative sentence for each cluster via text fusion. Our summarization method improves over the previous state-of-the-art MDS method in the DUC 2004 and TAC 2011 datasets, both in automatic ROUGE scores and human preference.

Recently, prompt learning has received significant attention, where the downstream tasks are reformulated to the mask-filling task with the help of a textual prompt. The key point of prompt learning is finding the most appropriate prompt. This paper proposes a novel model PromptGen, which can automatically generate prompts conditional on the input sentence. PromptGen is the first work considering dynamic prompt generation for knowledge probing, based on a pre-trained generative model. To mitigate any label information leaking from the pre-trained generative model, when given a generated prompt, we replace the query input with “None”. We pursue that this perturbed context-free prompt cannot trigger the correct label. We evaluate our model on the knowledge probing LAMA benchmark, and show that PromptGen significantly outperforms other baselines.

Fine-tuning continuous prompts for target tasks has recently emerged as a compact alternative to full model fine-tuning. Motivated by these promising results, we investigate the feasibility of extracting a discrete (textual) interpretation of continuous prompts that is faithful to the problem they solve. In practice, we observe a “wayward” behavior between the task solved by continuous prompts and their nearest neighbor discrete projections: We can find continuous prompts that solve a task while being projected to an arbitrary text (e.g., definition of a different or even a contradictory task), while being within a very small (2%) margin of the best continuous prompt of the same size for the task. We provide intuitions behind this odd and surprising behavior, as well as extensive empirical analyses quantifying the effect of various parameters. For instance, for larger model sizes we observe higher waywardness, i.e, we can find prompts that more closely map to any arbitrary text with a smaller drop in accuracy. These findings have important implications relating to the difficulty of faithfully interpreting continuous prompts and their generalization across models and tasks, providing guidance for future progress in prompting language models.

Identifying all possible user intents for a dialog system at design time is challenging even for skilled domain experts. For practical applications, novel intents may have to be inferred incrementally on the fly. This typically entails repeated retraining of the intent detector on both the existing and novel intents which can be expensive and would require storage of all past data corresponding to prior intents. In this paper, the objective is to continually train an intent detector on new intents while maintaining performance on prior intents without mandating access to prior intent data. Several data replay-based approaches have been introduced to avoid catastrophic forgetting during continual learning, including exemplar and generative replay. Current generative replay approaches struggle to generate representative samples because the generation is conditioned solely on the class/task label. Motivated by the recent work around prompt-based generation via pre-trained language models (PLMs), we employ generative replay using PLMs for incremental intent detection. Unlike exemplar replay, we only store the relevant contexts per intent in memory and use these stored contexts (with the class label) as prompts for generating intent-specific utterances. We use a common model for both generation and classification to promote optimal sharing of knowledge across both tasks. To further improve generation, we employ supervised contrastive fine-tuning of the PLM. Our proposed approach achieves state-of-the-art (SOTA) for lifelong intent detection on four public datasets and even outperforms exemplar replay-based approaches. The technique also achieves SOTA on a lifelong relation extraction task, suggesting that the approach is extendable to other continual learning tasks beyond intent detection.

Semi-supervised learning is a promising way to reduce the annotation cost for text-classification. Combining with pre-trained language models (PLMs), e.g., BERT, recent semi-supervised learning methods achieved impressive performance. In this work, we further investigate the marriage between semi-supervised learning and a pre-trained language model. Unlike existing approaches that utilize PLMs only for model parameter initialization, we explore the inherent topic matching capability inside PLMs for building a more powerful semi-supervised learning approach. Specifically, we propose a joint semi-supervised learning process that can progressively build a standard K-way classifier and a matching network for the input text and the Class Semantic Representation (CSR). The CSR will be initialized from the given labeled sentences and progressively updated through the training process. By means of extensive experiments, we show that our method can not only bring remarkable improvement to baselines, but also overall be more stable, and achieves state-of-the-art performance in semi-supervised text classification.

Probing is a popular approach to understand what linguistic information is contained in the representations of pre-trained language models. However, the mechanism of selecting the probe model has recently been subject to intense debate, as it is not clear if the probes are merely extracting information or modelling the linguistic property themselves. To address this challenge, this paper introduces a novel model-free approach to probing via prompting, which formulates probing as a prompting task. We conduct experiments on five probing tasks and show that PP is comparable or better at extracting information than diagnostic probes while learning much less on its own. We further combine the probing via prompting approach with pruning to analyze where the model stores the linguistic information in its architecture. Finally, we apply the probing via prompting approach to examine the usefulness of a linguistic property for pre-training by removing the heads that are essential to it and evaluating the resulting model’s performance on language modeling.

In most Vision-Language models (VL), the understanding of the image structure is enabled by injecting the position information (PI) about objects in the image. In our case study of LXMERT, a state-of-the-art VL model, we probe the use of the PI in the representation and study its effect on Visual Question Answering. We show that the model is not capable of leveraging the PI for the image-text matching task on a challenge set where only position differs. Yet, our experiments with probing confirm that the PI is indeed present in the representation. We introduce two strategies to tackle this: (i) Positional Information Pre-training and (ii) Contrastive Learning on PI using Cross-Modality Matching. Doing so, the model can correctly classify if images with detailed PI statements match. Additionally to the 2D information from bounding boxes, we introduce the object’s depth as new feature for a better object localization in the space. Even though we were able to improve the model properties as defined by our probes, it only has a negligible effect on the downstream performance. Our results thus highlight an important issue of multimodal modeling: the mere presence of information detectable by a probing classifier is not a guarantee that the information is available in a cross-modal setup.