The Ninth FEVER Workshop

Symbol grounding (Harnad, 1990) describes how symbols such as words acquire their meanings by connecting to real-world sensorimotor experiences. Recent work has shown preliminary evidence that grounding may emerge in (vision-)language models trained at scale without using explicit grounding objectives. Yet, the specific loci of this emergence and the mechanisms that drive it remain largely unexplored. In the main part of the talk, I will present our recent work that addresses this problem by introducing a controlled evaluation framework that systematically traces how symbol grounding arises within internal computations through mechanistic and causal analysis. Our findings show that grounding concentrates in middle-layer computations and is implemented through the aggregate mechanism, where attention heads aggregate the environmental ground to support the prediction of linguistic forms. This phenomenon replicates in multimodal dialogue and across architectures (Transformers and state-space models), but not in unidirectional LSTMs. Our results provide behavioral and mechanistic evidence that symbol grounding can emerge in language models, with practical implications for predicting and potentially controlling the reliability of generation. I will conclude the talk by sharing thoughts on how interpretability workflows will lead to novel scientific findings and factual verification.

Fact verification research has made remarkable progress in recent years. However, it remains heavily centered on English-language data, Western knowledge sources, and monocultural assumptions about how claims are expressed and interpreted. In this talk, I will argue that multilinguality and cultural context are not peripheral challenges but central obstacles for trustworthy verification. Drawing on our recent work in multilingual and multicultural NLP, I will discuss how language technologies exhibit substantial disparities across languages (and also dialects) and how meaning is shaped by cultural context.

TBD.

TBD.