CambridgeEducation

Leveraging DELPH-IN grammars to develop educational materials

Discussion at the Cambridge Summit led by Kristen Howell; scribed by Francis Bond (2019).

Some discussion of coverage of various grammars.

Dan: we should also note the work of Norwegian, which has done a lot of work with the Sparrer.

• Even a restricted vocabulary can allow interesting grammatical phenomena, if you work with the grammar. You can try to work within the limits to produce your exercises. You can have a very small grammar and still start doing things with it. Grammar correction over unconstrained text is going to be out of reach for most grammars: although it works with the ERG. Semantic correctness is still an issue.

David: how do we check the semantic correctness --- Dan check against reference answers. David: is the framework there, is it documented --- yes. Francis: Luis's stuff is in github https://github.com/lmorgadodacosta/LCC-AssignmentChecker which may act as some sort of documentation. The Sparrer also has a good documentation of the phenomena.

Emily: how do you fail gracefully. Dan: We tell them what we found, and are quiet if we find nothing.

David: Do you only say something is bad if you parse it and find something and it is wrong.

Dan: yes.

Francis: we have two levels of errors (warning and error): if we can't parse it we just give a vague warning.

Emily: what if you don't have much information about the errors you expect?

David: For Lushotseed I have a fair idea of the errors but am not sure how to write mal-rules.

• e.g. for clitics in the worng place or weird reduplication.

Guy: you normally constrain what the students can do, right?

Dan: yes, we can constrain their input and it gives us more control. So often we will even restrict the part-of-speech. This means we cannot show so much range of variation, which may not be so good for the students as we can't see so many types of errors.

Francis: although we really want them to get it right, so I think this may be a good thing (at least for early learners).

David: how do you restrict the vocab Dan: they have to chose from a list and we reject it otherwise

Emily: picture and words can connect also to the cultural background, ...

One issue can be that different dialects exist and you may have to specify one.

John: lexical collocations are an important problem you have not mentioned here. You can, for example, say this n-gram has not been seen before.

Dan: you may not be able to say what the error was but you can suggest it is strange.

Francis: we would like to use this to keep things in genre

John: keeping people on topic is also difficult --- they may write something good but not relevant https://www.youtube.com/watch?v=tTv5ckMe_2M

Francis: we could also try to do vocabulary games, although the literature suggests that this can backfire if you group things by semantic field.

Dan: if you can generate, then you can produce the correct sentence, and this may be useful. The grammar sparrer does this a lot.

Francis: I would think you need a good statistical model or you could end up being very confusing.