Recent work on the interpretability of deep neural language models has concluded that many properties of natural language syntax are encoded in their representational spaces. However, such studies often suffer from limited scope by focusing on a single language and a single linguistic formalism. In this study, we aim to investigate the extent to which the semblance of syntactic structure captured by language models adheres to a surface-syntactic or deep syntactic style of analysis, and whether the patterns are consistent across different languages. We apply a probe for extracting directed dependency trees to BERT and ELMo models trained on 13 different languages, probing for two different syntactic annotation styles: Universal Dependencies (UD), prioritizing deep syntactic relations, and Surface-Syntactic Universal Dependencies (SUD), focusing on surface structure. We find that both models exhibit a preference for UD over SUD - with interesting variations across languages and layers - and that the strength of this preference is correlated with differences in tree shape.
Previous studies investigating the syntactic abilities of deep learning models have not targeted the relationship between the strength of the grammatical generalization and the amount of evidence to which the model is exposed during training. We address this issue by deploying a novel word-learning paradigm to test BERT's few-shot learning capabilities for two aspects of English verbs: alternations and classes of selectional preferences. For the former, we fine-tune BERT on a single frame in a verbal-alternation pair and ask whether the model expects the novel verb to occur in its sister frame…
RNN language models have achieved state-of-the-art results on various tasks, but what exactly they are representing about syntax is as yet unclear. Here we investigate whether RNN language models learn humanlike word order preferences in syntactic alternations. We collect language model surprisal scores for controlled sentence stimuli exhibiting major syntactic alternations in English: heavy NP shift, particle shift, the dative alternation, and the genitive alternation. We show that RNN language models reproduce human preferences in these alternations based on NP length, animacy, and definiten…
Harvard University（哈佛大学）；Johns Hopkins University（约翰斯·霍普金斯大学）；Google Research（谷歌研究院）；New York University（纽约大学）；Technion – IIT（以色列理工学院）
Targeted syntactic evaluations have demonstrated the ability of language models to perform subject-verb agreement given difficult contexts. To elucidate the mechanisms by which the models accomplish this behavior, this study applies causal mediation analysis to pre-trained neural language models. We investigate the magnitude of models' preferences for grammatical inflections, as well as whether neurons process subject-verb agreement similarly across sentences with different syntactic structures. We uncover similarities and differences across architectures and model sizes -- notably, that large…
Next-word predictions from autoregressive neural language models show remarkable sensitivity to syntax. This work evaluates the extent to which this behavior arises as a result of a learned ability to maintain implicit representations of incremental syntactic structures. We extend work in syntactic probing to the incremental setting and present several probes for extracting incomplete syntactic structure (operationalized through parse states from a stack-based parser) from autoregressive language models. We find that our probes can be used to predict model preferences on ambiguous sentence pre…
Meta AI；MIT（麻省理工学院）；Johns Hopkins（约翰斯·霍普金斯大学）；Purdue University（普渡大学）
Targeted syntactic evaluations of language models ask whether models show stable preferences for syntactically acceptable content over minimal-pair unacceptable inputs. Most targeted syntactic evaluation datasets ask models to make these judgements with just a single context-free sentence as input. This does not match language models' training regime, in which input sentences are always highly contextualized by the surrounding corpus. This mismatch raises an important question: how robust are models' syntactic judgements in different contexts? In this paper, we investigate the stability of lan…
Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）；The Chinese University of Hong Kong（香港中文大学）
Syntactic structure of a sentence text is correlated with the prosodic structure of the speech that is crucial for improving the prosody and naturalness of a text-to-speech (TTS) system. Nowadays TTS systems usually try to incorporate syntactic structure information with manually designed features based on expert knowledge. In this paper, we propose a syntactic representation learning method based on syntactic parse tree traversal to automatically utilize the syntactic structure information. Two constituent label sequences are linearized through left-first and right-first traversals from const…
Both Direct and Indirect Evidence Contribute to Dative Alternation Preferences in Language Models
Department of Linguistics, The University of Texas at Austin（德克萨斯大学奥斯汀分校语言学系）；Department of Linguistics and Philology, Uppsala University（乌普萨拉大学语言学与哲学系）
A standard approach to evaluating language models analyzes how models assign probabilities to valid versus invalid syntactic constructions (i.e. is a grammatical sentence more probable than an ungrammatical sentence). Our work uses ambiguous relative clause attachment to extend such evaluations to cases of multiple simultaneous valid interpretations, where stark grammaticality differences are absent. We compare model performance in English and Spanish to show that non-linguistic biases in RNN LMs advantageously overlap with syntactic structure in English but not Spanish. Thus, English models m…
Fudan University（复旦大学）；Max Planck Institute for Evolutionary Anthropology（马克斯·普朗克进化人类学研究所）
Multilingual BERT (mBERT) has demonstrated considerable cross-lingual syntactic ability, whereby it enables effective zero-shot cross-lingual transfer of syntactic knowledge. The transfer is more successful between some languages, but it is not well understood what leads to this variation and whether it fairly reflects difference between languages. In this work, we investigate the distributions of grammatical relations induced from mBERT in the context of 24 typologically different languages. We demonstrate that the distance between the distributions of different languages is highly consistent…
The Edmond and Lily Safra Center for Brain Sciences（埃德蒙与莉莉·萨夫拉脑科学中心）；School of Computer Science and Engineering（计算机科学与工程学院）；Hebrew University of Jerusalem（希伯来大学）
Syntactic analysis plays an important role in semantic parsing, but the nature of this role remains a topic of ongoing debate. The debate has been constrained by the scarcity of empirical comparative studies between syntactic and semantic schemes, which hinders the development of parsing methods informed by the details of target schemes and constructions. We target this gap, and take Universal Dependencies (UD) and UCCA as a test case. After abstracting away from differences of convention or formalism, we find that most content divergences can be ascribed to: (1) UCCA's distinction between a S…
Natural language exhibits patterns of hierarchically governed dependencies, in which relations between words are sensitive to syntactic structure rather than linear ordering. While re-current network models often fail to generalize in a hierarchically sensitive way (McCoy et al.,2020) when trained on ambiguous data, the improvement in performance of newer Trans-former language models (Vaswani et al., 2017)on a range of syntactic benchmarks trained on large data sets (Goldberg, 2019; Warstadtet al., 2019) opens the question of whether these models might exhibit hierarchical generalization in th…
The Hong Kong University of Science and Technology（香港科技大学）；Sinovation Ventures（创新工场）；Tencent AI Lab（腾讯人工智能实验室）
Conventional word embeddings represent words with fixed vectors, which are usually trained based on co-occurrence patterns among words. In doing so, however, the power of such representations is limited, where the same word might be functionalized separately under different syntactic relations. To address this limitation, one solution is to incorporate relational dependencies of different words into their embeddings. Therefore, in this paper, we propose a multiplex word embedding model, which can be easily extended according to various relations among words. As a result, each word has a center…
How to define a context-free backbone for DGs: Implementing a DG in the LFG formalism
本文提出了一种多维依存文法（DG），它将依存树从词序中抽象出来，使得表面排序不是通过遍历依存树来确定的。我们开发了一个\emph{word order domain structure}的概念，它是链接的，但在结构上不同的语法依赖树。然后，我们讨论了这样一个DG的实现使用构造基于统一的短语结构的方法，即词汇功能语法（LFG）。特别注意的是在DG的LFG的功能不确定性方面的不连续性分析。
We introduce the syntactic formalism of the \textit{Cambridge Grammar of the English Language} (CGEL) to the world of treebanking through the CGELBank project. We discuss some issues in linguistic analysis that arose in adapting the formalism to corpus annotation, followed by quantitative and qualitative comparisons with parallel UD and PTB treebanks. We argue that CGEL provides a good tradeoff between comprehensiveness of analysis and usability for annotation, which motivates expanding the treebank with automatic conversion in the future.
Rutgers, The State University of New Jersey（罗格斯大学，新泽西州立大学）
In this thesis we present a semantic representation formalism based on directed graphs and explore its linguistic adequacy and explanatory benefits in the semantics of plurality and quantification. Our graph language covers the essentials of natural language semantics using only monadic second-order variables. We define its model-theoretical interpretation in terms of graph traversal, where the relative scope of variables arises from their order of valuation. We present a unification-based mechanism for constructing semantic graphs at a simple syntax-semantics interface, where syntax as a part…
The University of Tokyo（东京大学）；Artificial Intelligence Research Center, AIST（产业技术综合研究所人工智能研究中心）；LeapMind Inc.（LeapMind公司）
Incorporating stronger syntactic biases into neural language models (LMs) is a long-standing goal, but research in this area often focuses on modeling English text, where constituent treebanks are readily available. Extending constituent tree-based LMs to the multilingual setting, where dependency treebanks are more common, is possible via dependency-to-constituency conversion methods. However, this raises the question of which tree formats are best for learning the model, and for which languages. We investigate this question by training recurrent neural network grammars (RNNGs) using various …
Formal languages theory is useful for the study of natural language. In particular, it is of interest to study the adequacy of the grammatical formalisms to express syntactic phenomena present in natural language. First, it helps to draw hypothesis about the nature and complexity of the speaker-hearer linguistic competence, a fundamental question in linguistics and other cognitive sciences. Moreover, from an engineering point of view, it allows the knowledge of practical limitations of applications based on those formalisms. In this article I introduce the adequacy problem of grammatical forma…
When Discourse Pressures Conflict: Information Structure in Vision-Language Model Outputs
Department of Computer Science, Aalborg University（奥尔堡大学计算机科学系）；Department of Psycholinguistics and Neurolinguistics, ELTE Research Centre for Linguistics（ELTE语言研究中心心理学语言学与神经语言学系）；ELTE Bárczi Gusztáv Faculty of Special Needs Education（ELTE巴尔茨吉斯塔夫特殊教育学院）
Search has for a long time been an important tool for users to retrieve information. Syntactic search is matching documents or objects containing specific keywords like user-history, location, preference etc. to improve the results. However, it is often possible that the query and the best answer have no term or very less number of terms in common and syntactic search can not perform properly in such cases. Semantic search, on the other hand, resolves these issues but suffers from lack of annotation, absence of WordNet in case of low resource languages. In this work, we have demonstrated an en…
