Neural language models have exhibited outstanding performance in a range of downstream tasks. However, there is limited understanding regarding the extent to which these models internalize syntactic knowledge, so that various datasets have recently been constructed to facilitate syntactic evaluation of language models across languages. In this paper, we introduce JCoLA (Japanese Corpus of Linguistic Acceptability), which consists of 10,020 sentences annotated with binary acceptability judgments. Specifically, those sentences are manually extracted from linguistics textbooks, handbooks and journal articles, and split into in-domain data (86 %; relatively simple acceptability judgments extracted from textbooks and handbooks) and out-of-domain data (14 %; theoretically significant acceptability judgments extracted from journal articles), the latter of which is categorized by 12 linguistic phenomena. We then evaluate the syntactic knowledge of 9 different types of Japanese and multilingual language models on JCoLA. The results demonstrated that several models could surpass human performance for the in-domain data, while no models were able to exceed human performance for the out-of-domain data. Error analyses by linguistic phenomena further revealed that although neural language models are adept at handling local syntactic dependencies like argument structure, their performance wanes when confronted with long-distance syntactic dependencies like verbal agreement and NPI licensing.

In this paper, we propose a novel evaluation paradigm for Targeted Syntactic Evaluations, where we assess how well language models can recognize linguistic phenomena situated at different levels of the Chomsky hierarchy. Specifically, we create formal languages that abstract four syntactic phenomena in natural languages, each identified at a different level of the Chomsky hierarchy, and use these to evaluate the capabilities of language models: (1) (Adj)ˆn NP type, (2) NPˆn VPˆn type, (3) Nested Dependency type, and (4) Cross Serial Dependency type. We first train three different language models (LSTM, Transformer LM, and Stack-RNN) on language modeling tasks and then evaluate them using pairs of a positive and a negative sentence by investigating whether they can assign a higher probability to the positive sentence than the negative one. Our result demonstrated that all language models have the ability to capture the structural patterns of the (Adj)ˆn NP type formal language. However, LSTM and Transformer LM failed to capture NPˆn VPˆn type language and no architectures can recognize nested dependency and Cross Serial dependency correctly. Neural language models, especially Transformer LMs, have exhibited high performance across a multitude of downstream tasks, leading to the perception that they possess an understanding of natural languages. However, our findings suggest that these models may not necessarily comprehend the syntactic structures that underlie natural language phenomena such as dependency. Rather, it appears that they may extend grammatical rules equivalent to regular grammars to approximate the rules governing dependencies.

In this paper, we introduce JBLiMP (Japanese Benchmark of Linguistic Minimal Pairs), a novel dataset for targeted syntactic evaluations of language models in Japanese. JBLiMP consists of 331 minimal pairs, which are created based on acceptability judgments extracted from journal articles in theoretical linguistics. These minimal pairs are grouped into 11 categories, each covering a different linguistic phenomenon. JBLiMP is unique in that it successfully combines two important features independently observed in existing datasets: (i) coverage of complex linguistic phenomena (cf. CoLA) and (ii) presentation of sentences as minimal pairs (cf. BLiMP). In addition, JBLiMP is the first dataset for targeted syntactic evaluations of language models in Japanese, thus allowing the comparison of syntactic knowledge of language models across different languages. We then evaluate the syntactic knowledge of several language models on JBLiMP: GPT-2, LSTM, and n-gram language models. The results demonstrated that all the architectures achieved comparable overall accuracies around 75%. Error analyses by linguistic phenomenon further revealed that these language models successfully captured local dependencies like nominal structures, but not long-distance dependencies such as verbal agreement and binding.