Dear ChatGPT, Please Study Sociolinguistics

Recently, my husband started using ChatGPT voice mode to practice English, so I asked him how it was going. I was surprised to hear that he doesn’t actually use it very often. When I asked why, he said

人間でそういうふうに喋ってたら、その人が嫌われる。
If a person talked like that, everyone would hate them.

He was talking about the way that AI responses are always very long, which for language learners can feel overwhelming (something I’ve written about here with regard to Japanese learners). Instead of being 会話のキャッチボール (the back and forth of a natural conversation), he said talking to ChatGPT feels like being in a one-sided debate, or being on the receiving end of a lecture by a slightly annoyed teacher.

I think what he finds off-putting is the fact that ChatGPT doesn’t match his conversational style. This means that even when the response matches the topic (which it generally does), it feels contextually inappropriate. It’s as if ChatGPT is that person you don’t want to get stuck talking to at a party.

ChatGPT is that person you don’t want to get stuck talking to at a party.

So Dear ChatGPT, please study sociolinguistics and try to develop more interactionally appropriate conversation patterns. Or if sociolinguistics is too broad, maybe just start with the Maxim of Quantity, the first of Grice’s Maxims of Conversation.

The maxim of quantity:

relates to the quantity of information to be provided, and under it fall the following maxims:

1. Make your contribution as informative as is required (for the current purposes of the exchange).

2. Do not make your contribution more informative than is required.
   (Grice, 1975, p. 45, emphasis added)

In other words, Enough Already! AI’s verbosity detracts from its naturalness. And for the record, I’ve now used ChatGPT 5 voice a few times, including study mode, but I haven’t noticed any improvements yet in terms of the length or volume of output.

Shortening AI’s responses would make the output feel not only more natural but also more conversational. And, for language learners, especially in voice mode, shorter responses would make it much easier for learners to use generative AI for language practice.

I’ll leave the remaining maxims (quality, anyone?) for another day. But fixing just this one issue would be a huge improvement. Maybe the algorithm could be coded something like “match the length of the response to the length of the prompt.”

I would also prefer that Japanese ChatGPT not default to あなた (you), which takes us into the realm of linguistic politeness. But I’ll leave that for a future post.