I am working on my light verb construction library for the matrix and have some MRSes that I wanted to get feedback on. Below I have two example of transitive sentences, one where the coverb is verby (Bardi) and one where it is nouny (Persian).
What do you think?
Is there something else you would want in the MRSes that is currently missing?
Are there LVCs in a language that you know of that would need something different in their MRSes (for example, in Japanese the light verb suru is semantically empty so we wouldnât want that showing up in the MRS like it does for these two examples)?
Thank you!
Bardi
daab is the coverb -nya- is the light verb
aamba-nim garrin daab i-n-nya-na
man-ERG hill go.up.to 3-TR-catch-REC.PST
`The man went up the hill.'
Some thoughts and further questions â curious what the MRS mavens around here think:
For the verby coverb (in Bardi), the connection between the light verb and the coverb is in terms of the light verb taking the coverbâs ARG0 (rather than via qeq). I think this makes sense, given that people talk about light verb constructions describing single, cohesive events.
A difference between the two examples is that the verby coverb but not the nouny one takes the subject as an argument. Are we happy with this?
I wonder what should happen in cases where there is an object. Should it be an option provided to the linguist to have the object be a semantic of the argument of the light verb, the coverb, or both?
Inspired by some discussion just now, should take a poll, take a seat or take a bath (English LVCs) look exactly analogous to take a cookie? If not, how should they differ?
Responding to point 4, I donât see any option other than making the light verb MRS in English exactly the same as the normal MRS, because of the possibility of modification of the noun. Iâd also say that I feel that there are different degrees of bleaching of the verb semantics involved, so the light verb / normal verb distinction isnât clear cut - I donât know a good test.
Ideally, one would prevent the overgeneration - give a demo / *make a demo - and we suggested selectional restrictions in the `pain in the neckâ paper. It would also perhaps be nice to make the dialect differences clearer. That said, while âtake a bathâ is still weird for me (BrE âhave a bathâ), I find âtake a showerâ completely normal now. But since we anyway specify the verb in the MRS input to the generator, thatâs arguably not a problem for the grammar. My thoughts these days are that light verbs are not clearly different from the magnitude adjective case anyway.
I agree with what you are saying in the first part about the different degrees of bleaching of verb semantics. As for the second part, what do you mean by the âmagnitude adjective caseâ?
Sorry - I shouldnât just drop things in like that. In English, thereâs a lot of idiosyncracy in the distribution of adjectives like âhighâ, âheavyâ and so on. e.g. âheavyâ goes with precipitation weather nouns like âsnowâ, ârainâ and âshowerâ, but you canât talk about a âheavy showerâ in the washing sense of shower. With âwindâ (in the weather sense!), the normal magnitude adjective is âhighâ - and â*high rainâ, â?heavy windâ (my judgements) - I would call âhigh rainâ an anti-collocation. And so on - semi-regular, with patterns, but if one tries to make it work by giving the adjectives specific semantics, it just ends up looking extremely stipulative. âbig rainâ, âbig windâ donât seem exactly blocked but they are the sort of thing one might expect a non-native speaker to say. (I did a detailed corpus study of these things at some point but Iâm not sure I ever wrote it up properly).
In Meaning Text Theory, if I remember correctly, all these adjectives are associated with a `Magnâ function and nouns stipulate which adjectives they take as the value of the function. But that doesnât really work - for instance, for many nouns, multiple magnitude adjectives are found, but with slightly different meanings - e.g., âhigh taxationâ and âheavy taxationâ.
But Lexical function - Wikipedia need not be f : LU â LU, it can also be f : LU â Set LU, right? In examples in this Wikipedia page we have Syn mapping helicopter to copter, chopper. I never heared about Meaningâtext theory - Wikipedia, thank you @AnnC for mentioned it. But probably this citation to Meaning Text Theory is just a side note.
Actually I never saw a MTT analysis with a set value for Magn, and it would reduce the attraction of the idea, but even if that move were made, it wouldnât allow for the slightly different meanings of âhighâ and âheavyâ. Thatâs not consistent with the adjectives being bleached to Magn. One could say that are not different meanings but connotations - but at that point I would prefer to just say they are distinct adjectives.
I always thought of operators like Magn as being supertypes â the appropriate senses of âhighâ and âheavyâ are subtypes with more detailed properties. So if we can use Magn to give a general outline of what we mean, but it has to be specialized. In the same way, we would have an operator Light-Verb, and it has many children. Of course, this does not actually help us choose which one is the best child in any given case, âŚ
I always thought of operators like Magn as being supertypes â the appropriate senses of âhighâ and âheavyâ are subtypes with more detailed properties.
Yes, absolutely. It would be good to be able to specify supertypes in the generator, in fact. But I donât think that would comparable to the original MTT account.
We have done this for articles (including zero) and things like much_many in the ERG, so I would think it should work, although the more we do the more expensive it would get, of course,
