NAACL 2022 Papers

Textless spoken language processing is an exciting area of research that promises to extend applicability of the standard NLP toolset onto spoken language and languages with few or no textual resources. Here, we introduce textless-lib, a PyTorch-based library aimed to facilitate research in the area. We describe the building blocks that the library provides and demonstrate its usability by discuss three different use-case examples: (i) speaker probing, (ii) speech resynthesis and compression, and (iii) speech continuation. We believe that textless-lib substantially simplifies research the textless setting and will be handful not only for speech researchers but also for the NLP community at large.

Poetry generation, and creative language generation in general, usually suffers from the lack of large training data. In this paper, we present a novel framework to generate sonnets that does not require training on poems. We design a hierarchical framework which plans the poem sketch before decoding. Specifically, a content planning module is trained on non-poetic texts to obtain discourse-level coherence; then a rhyme module generates rhyme words and a polishing module introduces imagery and similes for aesthetics purposes. Finally, we design a constrained decoding algorithm to impose the meter-and-rhyme constraint of the generated sonnets. Automatic and human evaluation show that our multi-stage approach without training on poem corpora generates more coherent, poetic, and creative sonnets than several strong baselines.

Entity Linking (EL) maps an entity mention in a natural language sentence to an entity in a knowledge base (KB). The Zero-shot Entity Linking (ZEL) extends the scope of EL to unseen entities at the test time without requiring new labeled data. BLINK (BERT-based) is one of the SOTA models for ZEL. Interestingly, we discovered that BLINK exhibits diminishing returns, i.e., it reaches 98% of its performance with just 1% of the training data and the remaining 99% of the data yields only a marginal increase of 2% in the performance. While this extra 2% gain makes a huge difference for downstream tasks, training BLINK on large amounts of data is very resource-intensive and impractical. In this paper, we propose a neuro-symbolic, multi-task learning approach to bridge this gap. Our approach boosts the BLINK’s performance with much less data by exploiting an auxiliary information about entity types. Specifically, we train our model on two tasks simultaneously - entity linking (primary task) and hierarchical entity type prediction (auxiliary task). The auxiliary task exploits the hierarchical structure of entity types. Our approach achieves superior performance on ZEL task with significantly less training data. On four different benchmark datasets, we show that our approach achieves significantly higher performance than SOTA models when they are trained with just 0.01%, 0.1%, or 1% of the original training data. Our code is available at https://github.com/IBM/NeSLET.

While conversational semantic role labeling (CSRL) has shown its usefulness on Chinese conversational tasks, it is still under-explored in non-Chinese languages due to the lack of multilingual CSRL annotations for the parser training. To avoid expensive data collection and error-propagation of translation-based methods, we present a simple but effective approach to perform zero-shot cross-lingual CSRL.Our model implicitly learns language-agnostic, conversational structure-aware and semantically rich representations with the hierarchical encoders and elaborately designed pre-training objectives. Experimental results show that our model outperforms all baselines by large margins on two newly collected English CSRL test sets. More importantly, we confirm the usefulness of CSRL to non-Chinese conversational tasks such as the question-in-context rewriting task in English and the multi-turn dialogue response generation tasks in English, German and Japanese by incorporating the CSRL information into the downstream conversation-based models. We believe this finding is significant and will facilitate the research of non-Chinese dialogue tasks which suffer the problems of ellipsis and anaphora.

Event detection is a classic natural language processing task. However, the constantly emerging new events make supervised methods not applicable to unseen types. Previous zero-shot event detection methods either require predefined event types as heuristic rules or resort to external semantic analyzing tools. To overcome this weakness, we propose an end-to-end framework named Zero-Shot Event Detection Based on Ordered Contrastive Learning and Prompt-Based Prediction (ZEOP). By creatively introducing multiple contrastive samples with ordered similarities, the encoder can learn event representations from both instance-level and class-level, which makes the distinctions between different unseen types more significant. Meanwhile, we utilize the prompt-based prediction to identify trigger words without relying on external resources. Experiments demonstrate that our model detects events more effectively and accurately than state-of-the-art methods.

The current workflow for Information Extraction (IE) analysts involves the definition of the entities/relations of interest and a training corpus with annotated examples. In this demonstration we introduce a new workflow where the analyst directly verbalizes the entities/relations, which are then used by a Textual Entailment model to perform zero-shot IE. We present the design and implementation of a toolkit with a user interface, as well as experiments on four IE tasks that show that the system achieves very good performance at zero-shot learning using only 5–15 minutes per type of a user’s effort. Our demonstration system is open-sourced at https://github.com/BBN-E/ZS4IE. A demonstration video is available at https://vimeo.com/676138340.

Social chatbots, also known as chit-chat chatbots, evolve rapidly with large pretrained language models. Despite the huge progress, privacy concerns have arisen recently: training data of large language models can be extracted via model inversion attacks. On the other hand, the datasets used for training chatbots contain many private conversations between two individuals. In this work, we further investigate the privacy leakage of the hidden states of chatbots trained by language modeling which has not been well studied yet. We show that speakers’ personas can be inferred through a simple neural network with high accuracy. To this end, we propose effective defense objectives to protect persona leakage from hidden states. We conduct extensive experiments to demonstrate that our proposed defense objectives can greatly reduce the attack accuracy from 37.6% to 0.5%. Meanwhile, the proposed objectives preserve language models’ powerful generation ability.

The Yes/No QA task (Clark et al., 2019) consists of “Yes” or “No” questions about a given context. However, in realistic scenarios, the information provided in the context is not always sufficient in order to answer the question. For example, given the context “She married a lawyer from New-York.”, we don’t know whether the answer to the question “Did she marry in New York?” is “Yes” or “No”. In this paper, we extend the Yes/No QA task, adding questions with an IDK answer, and show its considerable difficulty compared to the original 2-label task. For this purpose, we (i) enrich the BoolQ dataset (Clark et al., 2019) to include unanswerable questions and (ii) create out-of-domain test sets for the Yes/No/IDK QA task. We study the contribution of training on other Natural Language Understanding tasks. We focus in particular on Extractive QA (Rajpurkar et al., 2018) and Recognizing Textual Entailments (RTE; Dagan et al., 2013), analyzing the differences between 2 and 3 labels using the new data.

Temporal Expression Extraction (TEE) is essential for understanding time in natural language. It has applications in Natural Language Processing (NLP) tasks such as question answering, information retrieval, and causal inference. To date, work in this area has mostly focused on English as there is a scarcity of labeled data for other languages. We propose XLTime, a novel framework for multilingual TEE. XLTime works on top of pre-trained language models and leverages multi-task learning to prompt cross-language knowledge transfer both from English and within the non-English languages. XLTime alleviates problems caused by a shortage of data in the target language. We apply XLTime with different language models and show that it outperforms the previous automatic SOTA methods on French, Spanish, Portuguese, and Basque, by large margins. XLTime also closes the gap considerably on the handcrafted HeidelTime method.

Word embeddings are one of the most fundamental technologies used in natural language processing. Existing word embeddings are high-dimensional and consume considerable computational resources. In this study, we propose WordTour, unsupervised one-dimensional word embeddings. To achieve the challenging goal, we propose a decomposition of the desiderata of word embeddings into two parts, completeness and soundness, and focus on soundness in this paper. Owing to the single dimensionality, WordTour is extremely efficient and provides a minimal means to handle word embeddings. We experimentally confirmed the effectiveness of the proposed method via user study and document classification.

The Word-in-Context (WiC) task has attracted considerable attention in the NLP community, as demonstrated by the popularity of the recent MCL-WiC SemEval shared task. Systems and lexical resources from word sense disambiguation (WSD) are often used for the WiC task and WiC dataset construction. In this paper, we establish the exact relationship between WiC and WSD, as well as the related task of target sense verification (TSV). Building upon a novel hypothesis on the equivalence of sense and meaning distinctions, we demonstrate through the application of tools from theoretical computer science that these three semantic classification problems can be pairwise reduced to each other, and therefore are equivalent. The results of experiments that involve systems and datasets for both WiC and WSD provide strong empirical evidence that our problem reductions work in practice.

Recent literature has seen growing interest in using black-box strategies like for testing the behavior of NLP models. Research on white-box testing has developed a number of methods for evaluatinghow thoroughly the internal behavior of deep models is tested, but they are not applicableto NLP models. We propose a set of white-box testing methods that are customized for transformer-based NLP models. These include MASK NEURON COVERAGE (MNCOVER) that measures how thoroughlythe attention layers in models are exercised during testing. We show that MNCOVER can refine testing suites generated by CheckList by substantiallyreduce them in size, for more than 60% on average, while retaining failing tests – thereby concentrating the faultdetection power of the test suite. Further we show how can be used to guide CheckList input generation,evaluate alternative NLP testing methods, and drive data augmentation to improve accuracy.

While recent work on multilingual language models has demonstrated their capacity for cross-lingual zero-shot transfer on downstream tasks, there is a lack of consensus in the community as to what shared properties between languages enable such transfer. Analyses involving pairs of natural languages are often inconclusive and contradictory since languages simultaneously differ in many linguistic aspects. In this paper, we perform a large-scale empirical study to isolate the effects of various linguistic properties by measuring zero-shot transfer between four diverse natural languages and their counterparts constructed by modifying aspects such as the script, word order, and syntax. Among other things, our experiments show that the absence of sub-word overlap significantly affects zero-shot transfer when languages differ in their word order, and there is a strong correlation between transfer performance and word embedding alignment between languages (e.g., 𝜌s=0.94 on the task of NLI). Our results call for focus in multilingual models on explicitly improving word embedding alignment between languages rather than relying on its implicit emergence.

Word alignment has proven to benefit many-to-many neural machine translation (NMT). However, high-quality ground-truth bilingual dictionaries were used for pre-editing in previous methods, which are unavailable for most language pairs. Meanwhile, the contrastive objective can implicitly utilize automatically learned word alignment, which has not been explored in many-to-many NMT. This work proposes a word-level contrastive objective to leverage word alignments for many-to-many NMT. Empirical results show that this leads to 0.8 BLEU gains for several language pairs. Analyses reveal that in many-to-many NMT, the encoder’s sentence retrieval performance highly correlates with the translation quality, which explains when the proposed method impacts translation. This motivates future exploration for many-to-many NMT to improve the encoder’s sentence retrieval performance.

Understanding longer narratives or participating in conversations requires tracking of discourse entities that have been mentioned. Indefinite noun phrases (NPs), such as ‘a dog’, frequently introduce discourse entities but this behavior is modulated by sentential operators such as negation. For example, ‘a dog’ in ‘Arthur doesn’t own a dog’ does not introduce a discourse entity due to the presence of negation. In this work, we adapt the psycholinguistic assessment of language models paradigm to higher-level linguistic phenomena and introduce an English evaluation suite that targets the knowledge of the interactions between sentential operators and indefinite NPs. We use this evaluation suite for a fine-grained investigation of the entity tracking abilities of the Transformer-based models GPT-2 and GPT-3. We find that while the models are to a certain extent sensitive to the interactions we investigate, they are all challenged by the presence of multiple NPs and their behavior is not systematic, which suggests that even models at the scale of GPT-3 do not fully acquire basic entity tracking abilities.

Recent causal probing literature reveals when language models and syntactic probes use similar representations. Such techniques may yield “false negative” causality results: models may use representations of syntax, but probes may have learned to use redundant encodings of the same syntactic information. We demonstrate that models do encode syntactic information redundantly and introduce a new probe design that guides probes to consider all syntactic information present in embeddings. Using these probes, we find evidence for the use of syntax in models where prior methods did not, allowing us to boost model performance by injecting syntactic information into representations.

Referring resolution is the task of identifying the referent of a natural language expression, for example “the woman behind the other woman getting a massage”. In this paper we investigate which are the kinds of referring expressions on which current transformer based models fail. Motivated by this analysis we identify the weakening of the spatial natural constraints as one of its causes and propose a model that aims to restore it. We evaluate our proposed model on different datasets for the task showing improved performance on the most challenging kinds of referring expressions. Finally we present a thorough analysis of the kinds errors that are improved by the new model and those that are not and remain future challenges for the task.

Pre-trained language models (PLMs) that use subword tokenization schemes can succeed at a variety of language tasks that require character-level information, despite lacking explicit access to the character composition of tokens. Here, studying a range of models (e.g., GPT- J, BERT, RoBERTa, GloVe), we probe what word pieces encode about character-level information by training classifiers to predict the presence or absence of a particular alphabetical character in a token, based on its embedding (e.g., probing whether the model embedding for “cat” encodes that it contains the character “a”). We find that these models robustly encode character-level information and, in general, larger models perform better at the task. We show that these results generalize to characters from non-Latin alphabets (Arabic, Devanagari, and Cyrillic). Then, through a series of experiments and analyses, we investigate the mechanisms through which PLMs acquire English-language character information during training and argue that this knowledge is acquired through multiple phenomena, including a systematic relationship between particular characters and particular parts of speech, as well as natural variability in the tokenization of related strings.

Transformer-based models are now predominant in NLP.They outperform approaches based on static models in many respects. This success has in turn prompted research that reveals a number of biases in the language models generated by transformers. In this paper we utilize this research on biases to investigate to what extent transformer-based language models allow for extracting knowledge about object relations (X occurs in Y; X consists of Z; action A involves using X).To this end, we compare contextualized models with their static counterparts. We make this comparison dependent on the application of a number of similarity measures and classifiers. Our results are threefold:Firstly, we show that the models combined with the different similarity measures differ greatly in terms of the amount of knowledge they allow for extracting. Secondly, our results suggest that similarity measures perform much worse than classifier-based approaches. Thirdly, we show that, surprisingly, static models perform almost as well as contextualized models – in some cases even better.