Skip to content

CapitolHillDefiniteness

EmilyBender edited this page Jan 6, 2017 · 3 revisions

Definiteness

A Discussion at the Capitol Hill meeting, January 2017

CapitolHillTop

FZZ: This one dog cried (in Chinese) — doesn’t parse, because of previous analysis of COG-ST.

SSH: Paper with Joanna on how to express definiteness in Chinese (incl Cantonese & Mandarin). SSH added COG-ST info to her idea. This part differs from the other constraints in Zhong. Other work focused on robust parsing and strict generation. This part is much more inclined to theoretical precision.

EMB: Theoretically oriented?

FCB: Makes claims that are perhaps too strong.

SSH: Definiteness largely dependent on on classifiers and demonstratives. Dem-Cl-N definite, Nume-Cl-N indefinite (in both Mandarin and Cantonese). Cl-N/bare N differ across varieties. Can always be indefinite, but Cl-N can also be interpreted as definite in Cantonese; bare N also as definite in Mandarin.

SSH: map Dem-Cl-N to uniq-or-more, Nume-Cl-N to type-id, Cl-N/N to type-id if strictly indefinite or cog-st if ambiguous between indefinite & definite.

DPF: type-id is type indefinite?

EMB: No, type identifiable. Speaker expects addressee to understand the type of thing being described, but not to know anything about the particular individual.

SSH: Two problems — added one extra subj-head-phrase in rules.tdl. subj-head-nn for non-nominal subject, and subj-head-n for nominal subjects.

FZZ: Extra constraint (requiring subjects to be indefinite) causing some real sentences not to parse because we do find indefinite subjects. The other problem is the hierarchy of the feature value (under cog-st). Problem arises with dem+num+cl NP … one says type-id and one says uniq-or-more.

FZZ: Ex: this one CL dog cried

EMB: What’s the COG-ST there?

FZZ: It’s definite. It’s as if the demonstrative is over-writing what the numeral had to say.

SSH: Do you agree with Joanna’s analysis?

FZZ: Mostly — but Cl-N is very rare in Mandarin.

EMB: When the Num-Cl-N occurs all by itself it’s indefinite, and we don’t want to lose that information.

DPF/GCS: Could introduce a joint subtype between demonstrative (not part of COG-ST) and type-id.

EMB: Alternative: keep Num-Cl underspecified until it either combines with the demonstrative or with nothing (pumping rule).

FZZ: Yes, that’s what I’m planning on.

GCS: Any other possibility?

FZZ: My three books indefinite.

DPF: Really? What about my dog?

FZZ: My dog (no numcl) is definite.

DPF: What’s your test for definiteness?

FZZ: If the listener can ask “which one?”.

LMC: Are you saying that there’s ambiguity between my two dogs two of my dogs for these strings in Chinese?

FZZ: Wo3 de liang2 zhi1 go3 — My two dogs / Liang2 zhi1 wo3 de go3 two of my dogs

DPF: If the possessive is the last thing to connect, it gets to say that it’s definite.

LMC/FZZ: Not a very common thing to say in Chinese all in one NP. Both of those sound quite artificial.

LMC: My two dogs ran away?

FZZ: Wo3 na4 liang2 zhi1 go3 — insert a demonstrative

LMC: I like your two dogs? Are we seeing an interaction between subject position and definiteness here?

FZZ: I would just say I like your dogs. If I really want the number and possessive I’d put the demonstrative in too.

DPF: Okay, let’s not try to bend the grammar to do things it doesn’t need to do. We got into this because we were imagining two cases for num-cl w/ and w/o numeral classifiers. We were just trying to see if we needed a third rule, and apparently we don’t, because it doesn’t occur. We can pretend it does…

FZZ: Another thing to consider is whether relative clauses interact with definiteness.

GCS: So both technical proposals work?

DPF: No, because the bypass proposal fails to make a commitment about definiteness when there’s a demonstrative.

FZZ: Can I tap your linguistic knowledge in other languages?

EMB: Relative clauses are compatible with indefiniteness in English.

DAI: I saw a movie which was very boring.

SSH: Joanna focused on ordering constraints between elements (e.g. possessive, number, demonstratives). Current version of the test suite focuses on that. You probably need more test sentences.

FZZ: Current implementation is slightly different from what’s described in the paper. Classifier expects numeral as specifier. Demonstrative can take a complement which is a classifier. That takes care of the order.

EMB: So maybe you don’t need a pumping rule? The num-cl and the demonstrative can say different things about their SPEC values … or not, because they’re both going to be talking about the same index (the noun’s). So back to the pumping rule.

DPF: Would be surprising if relative clauses constrained definiteness, because we’re just treating them as interactive modifiers, just like adjectives.

EMB: Another reason you wouldn’t expect them to be restricted to definiteness at least is that the canonical use of indefiniteness is introducing new discourse elements, which is exactly where you’d want relative clauses.

GCS: So what about possessives?

DPF: What is it about possessives that makes them definite?

EMB: I think they’re logically independent properties, but surely linked in English.

DPF/GCS: So why is it easier to start a discourse with a possessive NP than a definite one?

EMB: With reference to “Score keeping in a language game” — it’s not that you already have the shared knowledge, but that you’re asking the hearer to accommodate it, and that’s easier with possessives because you’re provided more information.

DPF: My dog and a large cat just walked into the room. Can I say, Which one?

All: Trying to test the definiteness of the conjoined entity, but it’s hard to use which one that way?

EMB: In Chinese, what form does the which one question take?

FZZ: na3 ge, or na3 NUM ge == which NUM (CL).

DAI: I suspect that coordinate indices will always be definite.

EMB: English doesn’t give us many ways to test this.

DPF: Two dogs and two cats walked into the room. They got into a fight.

DAI: Has to be all four.

EMB: But that doesn’t tell us anything about the definiteness.

DPF: We’re supposed to ask the which one question. Two dogs and two cats got into a fight. Which ones? Can be about the sets of pairs. But that should be impossible if it’s definite.

DAI: So underspecified?

DPF: That would be my guess, but that’s a bit worrisome: My two dogs and my two cats got into a fight. #Which ones? So it seems that the coordinated index is definite if at least one conjunct is.

EMB: Really?

DPF: My dog and two cats got into a fight. Which ones?

EMB: That’s grammatical, no? Which cats?

DAI: But Which ones can’t refer to the group including the dog.

EMB: English isn’t going to have anything sensitive to this. But Farsi for example uses accusative case only on definite direct objects.

DAI: And Inuktitut is split ergative on definiteness.

FZZ: Joanna has provided a test set. I want to buy CL cat should be indefinite. It’s casual, but grammatical. This is an example of numeral omission.

GCS: But then it has to be one.

FZZ: Right.

DPF: But we don’t hallucinate in our grammars.

DAI: But we do with pumping rules.

LMC: I buy CL cat is the informal way of saying I buy one CL cat

DPF/EMB: Pumping rule that introduces the semantics for one.

  • [ NB: In a later discussion, Luis came up with the interesting test of asking How many cats did you buy? and it seems like this form wasn't a suitable answer. So we came around to the idea that I buy CL cat is not in fact a precise paraphrase of I buy one CL cat and probably shouldn't have the same MRS. Might be more suitable to use the feature specification [ NUM sg ] for the effect of this rule. ]

FZZ: Not used in subject position.

LMC: It’s indefinite! I think the definition of indefiniteness is fuzzy.

DPF: We have a precise test. Using it grimly because we don’t have another one.

DAI: So how are you avoiding putting Cl cat in subject position?

FZZ: The grammar right now requires numerals with classifiers.

DAI/EMB: Introduce covert case as a time-honored kludge.

FZZ: How?

EMB: Give the grammar a nominative/accusative case system; where verbs constrain subject to be [ CASE nom ]. Most nouns don’t care, but the pumping rule for classifiers without numbers say [ CASE acc ].

GCS: I have an implementation of that pumping rule for Thai. Could be a test bed.

FZZ: We used to say COG-ST distinguished that.

DPF: But you’ve shown that that’s wrong. We know that something is special about subjects (cf XARG). To distinguish NPs that can show up in that position or not, it’s fine to use the feature CASE. Could call it GLOOP instead if you want to. Seems easier to remember what you want it to be doing if you call it CASE.

FCB: Check interaction with ba3 construction, passives, and other things that change subjects and see if these things show up there. E.g. A boy is bitten by the dog.

FZZ/DAI: With a number, not the Cl+N version (in Mandarin; Cantonese differ).

SSH: So far we’ve avoided using case because it’s been reported that there is no case in Chinese languages.

DPF: A binary distinction in grammatical form between subjects and non-subjects needs a name.

DAI: There are a lot of languages that have this distinction without morphology.

EMB/DPF: If we can call the distinction case in English (when it only appears in the pronouns), why not in Chinese?

MWG: Can you just say that the thing in subject position can’t have an implicit numeral?

EMB: At the NP level, that’s not first-class information. Something in the HEAD value is.

FCB: Need to be able to sell this to our local typologists.

DPF: So spell it ESAC.

GCS: There is a cline that includes subject down to more oblique arguments, right? Is that orthogonal to the cline of definiteness?

EMB: They are analytically independent, but can be correlated. Not as a hard constraint in English, and as we are learning not as a hard constraint in Mandarin either.

GCS: So what’s the other name for it, if not obliqueness?

EMB: It’s CASE.

GCS: I still have some resistance to that.

EMB: Case means variation in the form of NPs that correlates with grammatical function. Here, the variation isn’t morphology, but it’s still variation in form. So it’s case. Anyone who uses that term differently is just wrong.

Clone this wiki locally