ChatGPT, what is a 'Glasanwalt'? - Linguistic strategies in a large language model's interpretation of novel compounds
This study presents a large language model, GPT-4o, with a dataset of artificial German noun-noun compounds that consist of two simplex noun constituents, each associated with a typical interpretation pattern. The task is to derive plausible...
mehr
|
Volltext:
|
|
|
Zitierfähiger Link:
|
|
|
This study presents a large language model, GPT-4o, with a dataset of artificial German noun-noun compounds that consist of two simplex noun constituents, each associated with a typical interpretation pattern. The task is to derive plausible interpretations. We find that GPT-4o performed very well, displaying stable compositional reasoning strategies. As expected from linguistic literature, typical patterns of the constituents were clearly preferred, with a tendency to favor patterns typical for the head constituent.
|
| Export in Literaturverwaltung |
|
Journal for Language Technology and Computational Linguistics. Special issue on LLM fails – failed experiments with generative AI and what we can learn from them
This JLCL special issue focuses on linguistic and NLP experiments with generativeAI that did not yield the desired results. All papers explore the extent in which their failed experiment can contribute to knowledge gain regarding the work with...
mehr
|
Volltext:
|
|
|
Zitierfähiger Link:
|
|
|
This JLCL special issue focuses on linguistic and NLP experiments with generativeAI that did not yield the desired results. All papers explore the extent in which their failed experiment can contribute to knowledge gain regarding the work with generative AI.
|
Editorial
Failed experiments typically have no place in scientific discourse; they are discarded and not published. We believe that this practice results in a loss of potential knowledge gain. A systematic reflection on the causes of failures allows for the...
mehr
|
Volltext:
|
|
|
Zitierfähiger Link:
|
|
|
Failed experiments typically have no place in scientific discourse; they are discarded and not published. We believe that this practice results in a loss of potential knowledge gain. A systematic reflection on the causes of failures allows for the critical examination and/or improvement of methods used. Furthermore, when previously failed experiments are repeated and subsequently succeed, progress can be explicitly determined. From the perspective of methodological reflection, the discussion and documentation of failures thus provide added value for the scientific community. This is particularly true in a field like research on and with generative artificial intelligence (AI), which lacks a long-standing tradition and in which best practices are still in the process of being established. This JLCL special issue focuses on linguistic and NLP experiments with generative AI that did not yield the desired results. All papers explore the extent in which their failed experiment can contribute to knowledge gain regarding the work with generative AI.
|
