While many workshops have been devoted to the building of educational software (https://sig-edu.org/bea/2024), the development of intelligent writing assistants (https://in2writing.glitch.me), or the evaluation of written text (Jourdan et al. 2023; Dale & Viethen, 2021; Leacock et al. 2022; Heidorn, 2000; Dale & Kilgarriff, 2010), they were generally only concerned with the formal aspects, that is, the lower end of writing: grammatical error detection, spotting of spelling mistakes, etc.
Yet writing or text composition require much more:
texts must be readable, and to achieve this goal authors must be able to do more than just produce a set of correct forms. Sentences must be organized, point into some direction, and the whole (resulting text) must convey a point.
the hardest and most important part of composition is actually not the very act of writing (conversion of ideas into a set of grammatically correct and well spelled linguistic forms), but the carrying out of the preceding steps, which normally are planning, reasoning, and thinking.
This being so, we need software that assist humans not only at the lower end, but also at the higher levels (conceptual planning: ideation, content organization, etc.) preceding writing. For example, we need to help people
to find ideas,
to evaluate them (truth, relevancy) and
to organize them in such a way that the whole makes sense (coherence) for the reader, allowing her to get the author's point or understand his line of reasoning.
While all this sounds obvious, this seems to be the first workshop
to consider the entire spectrum of writing (planning: ideation, organization; drafting/expression; editing; revision);
to include the higher levels of composition,
to integrate right from the start humans into the loop of the development cycle, and
to consider support and feedback at any moment (before, during, and after writing) and not only at the very end.
Obviously writing is an important skill to survive in our modern world. Yet, writing is a difficult, energy- and time consuming process (cyclic revision, rereading). In addition, it implies a steep learning curve to get from the level of the novice to the expert.
Since nowadays machines 'can' do the job, one may wonder why not leave it to them? Indeed, there are situations where this makes sense (routine work, business letters), but there are also quite a few others where one wouldn't recommend this strategy at all (education).
This being said, one may find a middle ground, a situation where men and machines work together, each one contributing what they are best at. Remains to be seen where in the process machines can help, and where it is better to leave the control to humans. Hence, the main question is not 'shall we use GPT to produce texts,' but 'how and when (what level)' to use LLMs (or other techniques) to help people to produce written text? Obviously, even if humans and machines are both able to produce texts they do this in very different ways (Campbell, 2023; Mahlow, 2023a, 2023b; Bazerman, 2018).
References
Bazerman, C. (2018). What do humans do best? Developing communicative humans in the changing socio-cyborgian landscape. In S. Logan & W. Slater (Eds.), Perspectives on Academic and Professional Writing in an Age of Accountability (pp. 187–203). Southern Illinois University Press.
Campbell, S. H. (2023). What is human about writing? Writing process theory and ChatGPT.
Dale, R. & Kilgarriff, A. (2010). Helping Our Own: Text massaging for computational linguistics as a new shared task. In Proceedings of the 6th International Natural Language Generation Conference.
Dale, R., & Viethen, J. (2021). The automated writing assistance landscape in 2021. Natural Language Engineering,27(4), 511-518.
Heidorn, G. (2000). Intelligent writing assistance. In R. Dale, H. Moisl, and H. Somers, editors, Handbook of Natural Language Processing, pages 181–207. Marcel Dekker, Inc.
Jourdan, L., Boudin, F., Dufour, R. & Hernandez, N. (2023). Text revision in scientific writing assistance: An overview. arXiv preprint arXiv:2303.16726.
Leacock, C., Gamon, M., Mejia, J. A. & Chodorow, M. (2022). Automated grammatical error detection for language learners. Springer Nature.
Mahlow, C. (2023a). Large Language Models and Artificial Intelligence as Tools for Teaching and Learning Writing. Osnabrücker Beiträge zur Sprachtheorie, 101, 175-196.