StanfordCrowgeyIntrinsicVariableNotes

Joshua Crowgey

Part I

• Lushootseed updates
  • Climb-like metagrammar
  • numerous new features (causatives, applicatives, passives, possessive affixes, reduplication, ...)
  • now parsing ~50% of testsuite from Lushootseed Reader
• Valence Changing Morphology
  • All transitives are derived (from monovalent unaccusatives)
  • We could write alternate lexical entries for causativized (transitivized) words...
    • but that's not satisfying
  • Instead treat them like ERG's periphrastic causatives (two events: original and causative)
• An agent is already like a causer
  • I don't see why, if that's the case, why we need separate cause events
  • Why don't we unify ARG1 and cause_rel, in some sense
  • So I build an MRS where the ARG0 is shared by the verb and cause_rel
  • And this is where I'm doing violence to the intrinsic-variable property

Dan:

• so you have one ARG0 and two ARG1s?

JC:

• that's right, but two predicates
• I could go with RMRS, but that would require a rewrite of matrix.tdl

Part II

• variable types and derivation
• nominals can take subjects, sometimes need TENSE and MOOD
• I could say nouns have a predicative lexical rule

Questions

Antske:

• [some group, were at NAACL] want to put all events on a timeline (timeML or Red)

JC:

• I don't think I lose information; I have two predications, so their respective ARG1s are distinct

Antske:

• But I think it becomes unclear

Guy:

• What not just add an ARG2? Leave it unspecified in the default case.

JC:

• I hadn't really thought about that, but ok.

EMB:

• If we are coming from English, we'd demote the original ARG1 to ARG2, then the new thing becomes the new ARG1?

Antske:

• Do all verbs do this?

Guy:

• Ann has said that we don't explicitly say ARG1 is the agent, so maybe in Lushootseed it's more like a patient?

David:

• On the second issue (noun phrases).. Do you have to have a determiner for all of these?

JC:

• I think yes..

David:

• names?

JC:

• Not usually..

David:

• So what happens when we put the determiner thing in front of a causativized thing?

JC:

• that's a really good question.

David:

• So you're saying that noun phrases are basically verbal.. I think we should discuss this, because the same thing happens in (...)
• Do you get possessive marking only on nouns, or other things?

JC:

• only nouns

David:

• Even nominalized things?

JC:

• right, so that's one thing to distinguish nouns and verbs; verbs can't take possessives

Woodley:

• ...prepositions?

JC:

• ...(scribe missed it)

Dan:

• I'm getting used to the idea of one ARG0, but I'm still surprised about using the same ARG1, why not splurge and give an ARG2

JC:

• It's because it's predictable; i know what the ARG1 of a cause rel is

Dan:

• I thought there'd be TENSE or ASPECT or some things that can happen to one of the predications but not the other
• As soon as you do the ARG2, it seems you don't need the cause_rel

Francis:

• It seems what would make this stronger is a way of telling the difference.
• The intuition that this describes it better is often hard to justify.
• Are there any tests where sharing an ARG0 makes a difference? So far I haven't seen this.

JC:

• The only thing I've seen so far would argue for more than one ARG0

Guy:

• But you've said that there are still 2 predicates

JC:

• but I'm still unclear what a predicate is