Large Language Models for phraseology, MWUs and CNPs. Some theoretical considerations and a practical experiment in the field of specialised language.

Colson, Jean-Pierre
(2025) International Workshop Beyond single words: the Interplay of phraseology, Multi-Word Units (MWU) and Complex Noun Phrases (CNP) — Location: Université Paris Cité, Paris (11.December.2025)

Files

JPParis25.pdf
  • Open Access
  • Adobe PDF
  • 4.29 MB
Download

Details

Authors
Abstract
Artificial intelligence is now partly seen as a hype that might not last for so long. It has also received a lot of criticism, among others for its hybrid theoretical foundations. Among these, linguistics played a key role, that is often ignored by engineers and computer scientists. In particular, the Transformer architecture is in many ways compatible with distributional semantics, corpus linguistics, and phraseology. However, the vector space model underlying Transformer can only account for part of the contextual linguistic information, due to linguistic factors but also to the limits of the underlying mathematical model. For these reasons, fine-tuning an LLM of intermediate size will in most cases yield better results than those provided by LLMs such as ChatGPT or Gemini. Corpus-based results will still be valuable in most cases to complement AI.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Colson, J.-P. (2025). Large Language Models for phraseology, MWUs and CNPs. Some theoretical considerations and a practical experiment in the field of specialised language. International Workshop Beyond single words: the Interplay of phraseology, Multi-Word Units (MWU) and Complex Noun Phrases (CNP), Université Paris Cité, Paris. https://hdl.handle.net/2078.5/271075