Talking to a Know-It-All GPT or a Second-Guesser Claude? How Repair reveals unreliable Multi-Turn Behavior in LLMs

Repair, an important resource for resolving trouble in human-human conversation, remains underexplored in human-LLM interaction. In this study, we investigate how LLMs engage in the interactive process of repair in multi-turn dialogues around solvable and unsolvable math questions. We examine whether models initiate repair themselves and how they respond to user-initiated repair. Our results show strong differences across models: reactions range from being almost completely resistant to (appropriate) repair attempts to being highly susceptible and easily manipulated. We further demonstrate that once conversations extend beyond a single turn, model behavior becomes more distinctive and less predictable across systems. Overall, our findings indicate that each tested LLM exhibits its own characteristic form of unreliability in the context of repair.

Reference Games as a Testbed for the Alignment of Model Uncertainty and Clarification Requests

In human conversation, both interlocutors play an active role in maintaining mutual understanding. When addressees are uncertain about what speakers mean, for example, they can request clarification. It is an open question for language models whether they can assume a similar addressee role, recognizing and expressing their own uncertainty through clarification. We argue that reference games are a good testbed to approach this question as they are controlled, self-contained, and make clarification needs explicit and measurable. To test this, we evaluate three vision-language models comparing a baseline reference resolution task to an experiment where the models are instructed to request clarification when uncertain. The results suggest that even in such simple tasks, models often struggle to recognize internal uncertainty and translate it into adequate clarification behavior. This demonstrates the value of reference games as testbeds for interaction qualities of (vision and) language models.

Surprisal and Metaphor Novelty: Moderate Correlations and Divergent Scaling Effects

Novel metaphor comprehension involves complex semantic processes and linguistic creativity, making it an interesting task for studying language models (LMs). This study investigates whether surprisal, a probabilistic measure of predictability in LMs, correlates with different metaphor novelty datasets. We analyse surprisal from 16 LM variants on corpus-based and synthetic metaphor novelty datasets. We explore a cloze-style surprisal method that conditions on full-sentence context. Results show that LMs yield significant moderate correlations with scores/labels of metaphor novelty. We further identify divergent scaling patterns: on corpus-based data, correlation strength decreases with model size (inverse scaling effect), whereas on synthetic data it increases (Quality-Power Hypothesis). We conclude that while surprisal can partially account for annotations of metaphor novelty, it remains a limited metric of linguistic creativity.

Dialogue Is Not Enough to Make a Communicative BabyLM (But Neither Is Developmentally Inspired Reinforcement Learning)

We investigate whether pre-training exclusively on dialogue data results in formally and functionally apt small language models. Based on this pre-trained llamalogue model, we employ a variety of fine-tuning strategies to enforce "more communicative" text generations by our models. Although our models underperform on most standard BabyLM benchmarks, they excel at dialogue continuation prediction in a minimal pair setting. While PPO fine-tuning has mixed to adversarial effects on our models, DPO fine-tuning further improves their performance on our custom dialogue benchmark.

Are BabyLMs Deaf to Gricean Maxims? A Pragmatic Evaluation of Sample-efficient Language Models

Implicit meanings are integral to human communication, making it essential for language models to be capable of identifying and interpreting them. Grice (1975) proposed a set of conversational maxims that guide cooperative dialogue, noting that speakers may deliberately violate these principles to express meanings beyond literal words, and that listeners, in turn, recognize such violations to draw pragmatic inferences. Building on Surian et al. (1996)'s study of children's sensitivity to violations of Gricean maxims, we introduce a novel benchmark to test whether language models pretrained on less than 10M and less than 100M tokens can distinguish maxim-adhering from maxim-violating utterances. We compare these BabyLMs across five maxims and situate their performance relative to children and a Large Language Model (LLM) pretrained on 3T tokens. We find that overall, models trained on less than 100M tokens outperform those trained on less than 10M, yet fall short of child-level and LLM competence. Our results suggest that modest data increases improve some aspects of pragmatic behavior, leading to finer-grained differentiation between pragmatic dimensions.

The InviTE Corpus: Annotating Invectives in Tudor English Texts for Computational Modeling

In this paper, we aim at the application of Natural Language Processing (NLP) techniques to historical research endeavors, particularly addressing the study of religious invectives in the context of the Protestant Reformation in Tudor England. We outline a workflow spanning from raw data, through pre-processing and data selection, to an iterative annotation process. As a result, we introduce the InviTE corpus -- a corpus of almost 2000 Early Modern English (EModE) sentences, which are enriched with expert annotations regarding invective language throughout 16th-century England. Subsequently, we assess and compare the performance of fine-tuned BERT-based models and zero-shot prompted instruction-tuned large language models (LLMs), which highlights the superiority of models pre-trained on historical data and fine-tuned to invective detection.

SceneGram: Conceptualizing and Describing Tangrams in Scene Context

Research on reference and naming suggests that humans can come up with very different ways of conceptualizing and referring to the same object, e.g. the same abstract tangram shape can be a "crab", "sink" or "space ship". Another common assumption in cognitive science is that scene context fundamentally shapes our visual perception of objects and conceptual expectations. This paper contributes SceneGram, a dataset of human references to tangram shapes placed in different scene contexts, allowing for systematic analyses of the effect of scene context on conceptualization. Based on this data, we analyze references to tangram shapes generated by multimodal LLMs, showing that these models do not account for the richness and variability of conceptualizations found in human references.

Are Multimodal Large Language Models Pragmatically Competent Listeners in Simple Reference Resolution Tasks?

We investigate the linguistic abilities of multimodal large language models in reference resolution tasks featuring simple yet abstract visual stimuli, such as color patches and color grids. Although the task may not seem challenging for today's language models, being straightforward for human dyads, we consider it to be a highly relevant probe of the pragmatic capabilities of MLLMs. Our results and analyses indeed suggest that basic pragmatic capabilities, such as context-dependent interpretation of color descriptions, still constitute major challenges for state-of-the-art MLLMs.

Can LLMs Ground when they (Don't) Know: A Study on Direct and Loaded Political Questions

Communication among humans relies on conversational grounding, allowing interlocutors to reach mutual understanding even when they do not have perfect knowledge and must resolve discrepancies in each other's beliefs. This paper investigates how large language models (LLMs) manage common ground in cases where they (don't) possess knowledge, focusing on facts in the political domain where the risk of misinformation and grounding failure is high. We examine the ability of LLMs to answer direct knowledge questions and loaded questions that presuppose misinformation. We evaluate whether loaded questions lead LLMs to engage in active grounding and correct false user beliefs, in connection to their level of knowledge and their political bias. Our findings highlight significant challenges in LLMs' ability to engage in grounding and reject false user beliefs, raising concerns about their role in mitigating misinformation in political discourse.

LLMs Struggle to Reject False Presuppositions when Misinformation Stakes are High

This paper examines how LLMs handle false presuppositions and whether certain linguistic factors influence their responses to falsely presupposed content. Presuppositions subtly introduce information as given, making them highly effective at embedding disputable or false information. This raises concerns about whether LLMs, like humans, may fail to detect and correct misleading assumptions introduced as false presuppositions, even when the stakes of misinformation are high. Using a systematic approach based on linguistic presupposition analysis, we investigate the conditions under which LLMs are more or less sensitive to adopt or reject false presuppositions. Focusing on political contexts, we examine how factors like linguistic construction, political party, and scenario probability impact the recognition of false presuppositions. We conduct experiments with a newly created dataset and examine three LLMs: OpenAI's GPT-4-o, Meta's LLama-3-8B, and MistralAI's Mistral-7B-v03. Our results show that the models struggle to recognize false presuppositions, with performance varying by condition. This study highlights that linguistic presupposition analysis is a valuable tool for uncovering the reinforcement of political misinformation in LLM responses.