Abstract
This article deals with the relationship between human and artificial intelligence in the context of an ongoing Spanish Writing Assistant project, where ChatGPT is used to assist in four key tasks related to either training the underlying language model or preparing future user communication. The project is an interdisciplinary collaboration between lexicographers with experience in language teaching and IT experts from a high-tech company. The article first describes the methodology of the overall project and the specific role of the lexicographers. It then discusses the three tasks in which the latter are directly involved: the construction of a set of two parallel Spanish corpora, one correct and the other with induced errors, the generation of validation material, and the writing of extended grammatical explanations for Spanish learners. Based on a large amount of empirical data, including 35,000 carefully reviewed sentences, the article details the different steps of the interaction between human and chatbot, as well as the experiences and reflections drawn from this process. It concludes that the two parts engage in very different types of relationships depending on the concrete task, and that human knowledge, culture, skills and language intuition are crucial for the chatbot to work properly.
| Original language | English |
|---|---|
| Journal | Lexikos |
| Volume | 34 |
| Pages (from-to) | 21-40 |
| Number of pages | 20 |
| ISSN | 1684-4904 |
| DOIs | |
| Publication status | Published - Mar 2024 |
Keywords
- Generative AI chatbots
- Spanish writing assistants
- corpus building
- human-assisted intelligence
- language didactics
- language learning
- training of language model
- HUMAN-ASSISTED INTELLIGENCE
- TRAINING OF LANGUAGE MODEL
- CORPUS BUILDING
- LANGUAGE LEARNING
- SPANISH WRITING ASSISTANTS
- CHATBOTS