On this page you can download the discussion paper that was submitted for publication in the Journal for Media Linguistics. The blogstract summarises the submission in a comprehensible manner. You can comment on the discussion paper and the blogstract below this post. Please use your real name for this purpose. For detailed comments on the discussion paper please refer to the line numbering of the PDF.
Blogstract of
Semantic Coherence and Topic Continuity in Interactions with AI
by Netaya Lotze & Anna Greilich
Human-Computer Interaction (HCI) has evolved dramatically with advances in artificial intelligence, particularly with the advent of large language models (LLMs) and generative transformer models (GPTs). Despite these technological strides, the linguistic quality of AI dialogue systems remains inconsistent, often exhibiting surface-level cohesion without genuine semantic coherence. This discrepancy poses challenges for the usability and reliability of AI systems, especially in conversational contexts where maintaining topic continuity is paramount.
In human-human communication (HHC), interlocutors effortlessly draw on shared knowledge as common ground (Stalnaker 2002) and complex semantic relations to ensure coherent interaction (Brinker/Hagemann 2001). In contrast, AI systems frequently produce responses that appear linguistically cohesive but lack deeper coherence, a phenomenon I have previously termed quasi-coherence (Lotze 2016, 2025). The recent AI discourse labels similar system behaviours as hallucinations, highlighting a persistent gap between linguistic form and semantic function. We do not define hallucinations and quasi-coherence as synonymous, but rather as partially overlapping concepts that share certain intentions while diverging in their analytical scope — a distinction we elaborate on in detail throughout this article.
Therefore, we to this day propose three forms of semantic coherence in HCI: (1) user-maintained logical coherence, (2) system-generated illusions of coherence through successful keyword parsing or LLM generating, and (3) quasi-coherence marked by surface cohesion without meaningful semantic coherence (Lotze 2016, 289).
This article aims to explore semantic coherence (Hallyday/Hasan 1976) and topic continuity (Givón 1983) within HCI by applying a multidisciplinary methodological framework (following Lotze 2020). We contrast classic rule- and plan-based chatbot architectures with modern voice user interfaces (VUIs), such as Amazon Alexa, and earlier generative models exemplified by ChatGPT3. By investigating these systems through qualitative and quantitative lenses, we seek to elucidate the linguistic mechanisms underlying coherence and its breakdowns in AI dialogue.
Our analysis draws upon three empirical studies from our research group, encompassing written and oral HCI across different interaction scenarios. Through this comprehensive examination, we aim to advance theoretical understanding of semantic continuity in HCI and inform the development of more coherent and user-responsive dialogue systems (following Lotze 2020).