C-LARA

At the last Zoom call, we talked about possible publications. Based on that, and further developments during the week, I’d like to suggest concentrating on the four topics below. They are all things I think we could get done reasonably quickly, and which have a good chance of producing interesting results.

Multi-Word Expressions

This is coming along well (most recent update here). There is a clear plan for making further progress. A good result would be both theoretically interesting and extremely useful. Submitting a paper to ALTA 2024 would be a natural continuation of last year’s paper. I have been discussing with Francis.

Images

We are close to having the enhanced image-generation scheme working in C-LARA. This will add two key capabilities:

• Creating a set of images with a coherent style based on an initial image.
• Revising an already created image.

Both of these are straightforward to do, exploiting the powerful image understanding functionality offered by GPT-4o. In the first case, we ask GPT-4o to create a verbal description of the first image’s style, and include that in the request used for subsequent images; in the second case, we ask for a description of the current image, and then combine that with the change request. I have been experimenting using a combination of the ChatGPT-4o web interface and API calls from the Python buffer, and it works well.

I am not yet sure where we could submit a paper, but it seems that it should open up exciting possibilities for pedagogical use. I am currently discussing with with Christèle and Ivana.

Using C-LARA for Oceanic languages

We have been making good progress on using C-LARA with the Kanak languages Iaai and Drehu, and we have several more things in the pipeline to enhance its capabilities here. It would be natural to think of submitting something to COOL 2025. We should carry on discussing with Anne-Laure, Stéphanie and Pauline.

AI authorship/phonetic lexica

Following our rather frustrating interaction with the ISCA ethics chair about AI authorship, it seems to me that there is only one way to make the point clearly enough: submit a paper where the AI is the first/only author, and can produce documentary evidence that it has done the greater part of both the research and the writing. I have been discussing with Chat, and we have what we think is a promising idea. Details here.

Next Zoom call

The next call will be at:

Thu May 23 2024, 18:00 Adelaide (= 08.30 Iceland = 09.30 Ireland/Faroe Islands = 10.30 Europe = 11.30 Israel = 12.00 Iran = 16.30 China = 18:30 Melbourne = 19.30 New Caledonia)