Description
Large Language Models have become mainstream raising high expectations of AI. This talk will start with a brief overview of the state-of-the-art in language technology, leading to the question of what the impact of recent developments can be on applications in digital humanities. The short answer to what this technology can do in this complex domain is: `we don’t know’. It is however clear that despite impressive looking results, automatically processing natural language is far from a solved problem. Research on Natural Language Processing (NLP) has a tradition of evaluating tools on benchmarks, but it is well-known that such results do not necessarily transfer to other datasets, in particular, when these come from another domain. Moreover, typical evaluation metrics such as accuracy and f-score are often not precise enough to know whether the output of a tool can be used for a specific study. I will therefore argue that it remains necessary to carefully evaluate NLP tools in the context of their application. Such an evaluation should be tailored identifying potential problems that can infer with the research question of the scholars using the tool. I will present some recent ideas on how to raise awareness and how to support researchers when searching for appropriate tools for their research.| Period | 23 May 2024 |
|---|---|
| Event title | Flows & Frictions Symposium |
| Event type | Conference |
| Location | Gothenburg, SwedenShow on map |
| Degree of Recognition | International |