In human speakers’ daily conversations, what we do not say matters. We not only compute the literal semantics but also go beyond and draw inferences from what we could have said but chose not to. How well is this pragmatic reasoning process represented in pre-trained large language models (LLM)? In this study, we attempt to address this question through the lens of manner implicature, a pragmatic inference triggered by a violation of the Grice manner maxim. Manner implicature is a central member of the class of context-sensitive phenomena. The current work investigates to what extent pre-trained LLMs are able to identify and tease apart different shades of meaning in manner implicature. We constructed three metrics to explain LLMs’ behavior, including LLMs-surprisals, embedding vectors’ similarities, and natural language prompting. Results showed no striking evidence that LLMs have explainable representations of meaning. First, the LLMs-surprisal findings suggest that some LLMs showed above chance accuracy in capturing different dimensions of meaning, and they were able to differentiate neutral relations from entailment or implications, but they did not show consistent and robust sensitivities to more nuanced comparisons, such as entailment versus implications and equivalence versus entailment. Second, the similarity findings suggest that the perceived advantage of contextual over static embeddings was minimal, and contextual LLMs did not notably outperform static GloVe embeddings. LLMs and GloVe showed no significant difference, though distinctions between entailment and implication were slightly more observable in LLMs. Third, the prompting findings suggest no further supportive evidence indicating LLM’s competence in fully representing different shades of meaning. Overall, our study suggests that current dominant pre-training paradigms do not seem to lead to significant competence in manner implicature within our models. Our investigation sheds light on the design of datasets and benchmark metrics driven by formal and distributional linguistic theories.

The rapid development of large language models (LLMs) motivates us to explore how such state-of-the-art natural language processing systems can inform aphasia research. What kind of language indices can we derive from a pre-trained LLM? How do they differ from or relate to the existing language features in aphasia? To what extent can LLMs serve as an interpretable and effective diagnostic and measurement tool in a clinical context? To investigate these questions, we constructed predictive and correlational models, which utilize mean surprisals from LLMs as predictor variables. Using AphasiaBank archived data, we validated our models’ efficacy in aphasia diagnosis, measurement, and prediction. Our finding is that LLMs-surprisals can effectively detect the presence of aphasia and different natures of the disorder, LLMs in conjunction with the existing language indices improve models’ efficacy in subtyping aphasia, and LLMs-surprisals can capture common agrammatic deficits at both word and sentence level. Overall, LLMs have potential to advance automatic and precise aphasia prediction. A natural language processing pipeline can be greatly benefitted from integrating LLMs, enabling us to refine models of existing language disorders, such as aphasia.

Graphical representations of speech generate powerful computational measures related to psychosis. Previous studies have mostly relied on structural relations between words as the basis of graph formation, i.e., connecting each word to the next in a sequence of words. Here, we introduced a method of graph formation grounded in semantic relationships by identifying elements that act upon each other (action relation) and the contents of those actions (predication relation). Speech from picture descriptions and open-ended narrative tasks were collected from a cross-diagnostic group of healthy volunteers and people with psychotic or non-psychotic disorders. Recordings were transcribed and underwent automated language processing, including semantic role labeling to identify action and predication relations. Structural and semantic graph features were computed using static and dynamic (moving-window) techniques. Compared to structural graphs, semantic graphs were more strongly correlated with dimensional psychosis symptoms. Dynamic features also outperformed static features, and samples from picture descriptions yielded larger effect sizes than narrative responses for psychosis diagnoses and symptom dimensions. Overall, semantic graphs captured unique and clinically meaningful information about psychosis and related symptom dimensions. These features, particularly when derived from semi-structured tasks using dynamic measurement, are meaningful additions to the repertoire of computational linguistic methods in psychiatry.

AI language models are increasingly transforming language research in various ways. How can language educators and researchers respond to the challenge posed by these AI models? Specifically, how can we embrace this technology to inform and enhance second language learning and teaching? In order to quantitatively characterize and index second language writing, the current work proposes the use of similarities derived from contextualized meaning representations in AI language models. The computational analysis in this work is hypothesis-driven. The current work predicts how similarities should be distributed in a second language learning setting. The results suggest that similarity metrics are informative of writing proficiency assessment and interlanguage development. Statistically significant effects were found across multiple AI models. Most of the metrics could distinguish language learners’ proficiency levels. Significant correlations were also found between similarity metrics and learners’ writing test scores provided by human experts in the domain. However, not all such effects were strong or interpretable. Several results could not be consistently explained under the proposed second language learning hypotheses. Overall, the current investigation indicates that with careful configuration and systematic metrics design, AI language models can be promising tools in advancing language education.

Yan Cong, Emmanuele Chersoni, Yu-Yin Hsu, Philippe Blache. Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP. 2023.