[Linganth] LLMs and Ling Anth Reading Group - April 17 - Maria Erofeeva and Nils Klowait

[Language Machines]

Dear Language Machines Network,


We'd like to invite you to our next reading group meeting. We'll discuss
Maria Erofeeva (Free University of Brussels) and Nils Klowait’s (Paderborn)
draft article "Nonhuman Situational Enmeshment – How Participants Build
Temporal Infrastructures for ChatGPT" on *Monday, March 17th, 2025 from
18:00-20:00 CET*. See below for the schedule of future meetings.


*Abstract: The ability of fine-tuned large-language models (LLMs) to
generate fluent text challenges our understandings of language-in-use. How
can LLMs produce meaningful text without living in the world as humans do?
This paper argues that (1) textually mediated human-machine interaction
requires no shared context beyond what is already indexed by the
collaboratively produced (co-)text, and (2) it is sufficient for the human
to appreciate and advance the resolution of misunderstandings. To
substantiate these claims, the paper contrasts a pragmatic understanding of
meaning with criticism from linguistic anthropology and empirical findings
from conversation analysis. Participants that interact through chat are not
physically co-present and often share a limited personal history.
References to physical presence or experiences not shared thus need to be
made explicit in the (co-)text. Conversation analysis further shows that
many of the potential misinterpretations imagined by semantic analysis in
scenarios of co-presence are not relevant in practice due to the co-text
utterances are embedded in. Participants project understandings through
statements and responses, and the absence of evidence of misunderstanding
is sufficient for the interaction to progress. When there is evidence of a
misunderstanding among humans, i.e. a failure to interpret intentions, this
is dealt with procedurally and step-by-step through the repair
organization. Accordingly, failures by LLMs to fully participate in repair
can be compensated by the human interactant procedurally. In short,
repairing a misguided interaction is an occasional and procedural
collaborative activity where interpretation can remain limited to the human
side of a human-machine interaction.*


RSVP here: languagemachinesnetwork at gmail.com (e.g. reply to this email) to
receive the Zoom link and the PDF of the article when it is available (not
to be circulated, please).


We look forward to seeing you there!


Anna, Siri, Michael


April 21 - Zachary Sheldon (Pittsburgh)

May 19 - Janet Connor (Leiden)

June 16 - Raffaele Buono (UCL)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/linganth/attachments/20250305/10dfa669/attachment.htm>