Legal citations require correctly recalling the law references of complex law article names and article numbering, which large language models typically treat as multi-token sequences. Motivated by the form-meaning pair of constructionist approaches, we explore treating these multi-token law references as a single holistic law token and examining the implications for legal citation accuracy and differences in model interpretability. We train and compare two types of models: LawToken models, which encode the legal citations as a single law token, and LawBase models, which treat them as multi-token compounds. The results show that LawToken models outperform LawBase models on legal citation tasks, primarily due to fewer errors in the article numbering components. Further model representation analysis reveals that, while both models achieve comparable semantic representation quality, the multi-token-based LawBase suffers from degraded representations in multistep decoding, leading to more errors. Taken together, these findings suggest that form-meaning pairing can operate in a larger context, and this larger unit may offer advantages in future modeling of legal reasoning. In practice, this approach can significantly reduce the likelihood of hallucinations by anchoring legal citations as discrete, holistic tokens, thereby minimizing the risk of generating nonexistent or incorrect legal references.

Constructions are direct form-meaning pairs with possible schematic slots. These slots are simultaneously constrained by the embedded construction itself and the sentential context. We propose that the constraint could be described by a conditional probability distribution. However, as this conditional probability is inevitably complex, we utilize language models to capture this distribution. Therefore, we build CxLM, a deep learning-based masked language model explicitly tuned to constructions’ schematic slots. We first compile a construction dataset consisting of over ten thousand constructions in Taiwan Mandarin. Next, an experiment is conducted on the dataset to examine to what extent a pretrained masked language model is aware of the constructions. We then fine-tune the model specifically to perform a cloze task on the opening slots. We find that the fine-tuned model predicts masked slots more accurately than baselines and generates both structurally and semantically plausible word samples. Finally, we release CxLM and its dataset as publicly available resources and hope to serve as new quantitative tools in studying construction grammar.

Ever-expanding evaluative texts on online forums have become an important source of sentiment analysis. This paper proposes an aspect-based annotated dataset consisting of telecom reviews on social media. We introduce a category, implicit evaluative texts, impevals for short, to investigate how the deep learning model works on these implicit reviews. We first compare two models, BertSimple and BertImpvl, and find that while both models are competent to learn simple evaluative texts, they are confused when classifying impevals. To investigate the factors underlying the correctness of the model’s predictions, we conduct a series of analyses, including qualitative error analysis and quantitative analysis of linguistic features with logistic regressions. The results show that local features that affect the overall sentential sentiment confuse the model: multiple target entities, transitional words, sarcasm, and rhetorical questions. Crucially, these linguistic features are independent of the model’s confidence measured by the classifier’s softmax probabilities. Interestingly, the sentence complexity indicated by syntax-tree depth is not correlated with the model’s correctness. In sum, this paper sheds light on the characteristics of the modern deep learning model and when it might need more supervision through linguistic evaluations.