Problem Statement

In the ERG meaning representations (and in logic-based semantics, more generally), how to deal with variation in the number and types of semantic arguments?

Survey what other broad-coverage initiatives do, e.g. EngValLex, PropBank, FrameNet, OntoNotes, AMR, etc.

Reading List

Resources to Review

Clear Cases



FrameNet lists two frames for deny, Affirm_or_deny and Prevent_or_allow_possession. On examining the annotated data, we found the following valence frames, and aligned them with the disavow and decline like senses:

disavow (1,3)

decline (2)






With example arguments:

disavow (1,3)

decline (2)




that they cheated


the rumor


the petitioners; the request; Kim a kiss


a kiss to Kim

Note1: In the NP, NP frame, either of the NPs can go missing, but not both at once: *Kim denied. Note2: FrameNet has one instance of something like deny a request classified with the frame that otherwise has the disavow senses. We think this is probably an annotation error.


deny_v_1(e,x,h)                      deny_v_1(e,x,p)
deny_v_2(e,x,x,x)                    deny_v_2(e,x,x,x)

Would like to know how each of these options (as well as always getting the same predicate symbol) look to a logically-inclined semanticist. Which is preferable? Is there yet something different that would be better?

Observations: In this case at least, the analysis is consistent with the approach to having the Sem-I do the refining, since the information needed to distinguish between the predicate symbols is all apparent in the ERS. Exception: Since both ARG2 and ARG3 are syntactically optional in deny_v_2, how would we know which way to specialize NP denies NP? Alternatively, how do we know when to put in ARG3, especially given that in NP denies NP, the second NP could be either ARG2 or ARG3?

This entails two separate lexical entries, which could then have different PRED values.

More data:

Interesting that the addressee is available for CP complement but not the VP one. Necessarily unexpressed in VP frame, but still there in the semantics?

Are gerunds similar in semantic type to CPs and nouns like rumor?

Is rumor something like a proposition in its semantics? But:

The word rumor requires something above the content, that the story had been passed from person to person --- but what deny targets is the content.

Are there predicates that just want the propositional content in their complements and can't take individuals? hope, insist, ... Dan has created a list.

Deny seems to want to take as its complement a proposition that has been asserted to be true --- some nouns carry this for the propositions they package, some don't. So one can deny a rumor, a claim, a conclusion but not a hypothesis, theory, hope or suspicion. The nouns that deny can take denote a communicative event with a truth value assigned. Deny asserts the opposite of that truth value. belief can't be a complement because it's an attitude about truth but not communicative. Deficient in both ways: wager, likelihood.

Deny with a CP complement presupposes that someone believes that the CP is true.

Deny a request similarly involves one actor pushing for one state of affairs (that they request) and the other pushing back the opposite (denying the request). Both rumor and request case involve flipping like this.

Request case also has the communicative act property: Kim denied Sandy a kiss suggests that Kim knew Sandy wanted a kiss.

The supervisor denied the employee a raise in the case where the person who put in the request was HR (and the employee never knew).

The guard denied the prisoner his 10 minutes in the sun. Allowed every day, standing request/expectation. Guard prevents it happening one day; deny is still good here even if prisoner said nothing.

In basketball --- player A goes up to make a shot which player B blocks at the basket: Denied! Cleo was on her way to Oslo. She had even gotten on the plane. And then: Denied! That denied is passive not past tense.

I denied him his rights. I denied her the privilege. Again: Standing expectations?

@The deputy president said for those such as MaMbeki, the struggle against apartheid was inherently a struggle to reclaim their birthright as a people, as successive racist regimes denied them their citizenship, dignity and identity. (source) --- what's being denied is the ability to claim or benefit from citizenship, dignity and identity.

NdT denied the solar system its 9th planet: What's being denied is the community's shared belief in 9 planets, but also the thing we enjoyed which was having 9 planets.

Mom denied my sister her green peas -- not okay if said sister hates peas.

Attempting to get just one sense for request/rumor cases, with all differences following from unsettled v. settled:

request type

rumor type

prop content p through coercion from NP

prop content p directly or through package



for some agent it's a good thing that p (become) +

communicative event asserts p -> +

rights? attempts? states?

flips potential for + to -

flips + to - by asserting

request type takes a much broader range of nouns, but can get that from "p (become) +" being good for agent A.

The only thing that doesn't fall out is the lack of communicative event on the unsettled/request side.

Antonym: grant, seems to work in both cases. I grant you that belief --> only works in the unsettled case, which is predicted. I grant you the ability to hold that belief.

More ex from FrameNet: deny any wrong-doing deny any mistakes ... interpreted as charges/claims that there had been wrong-doing or mistakes. Needs to be coerced into proposition with presupposed communicative event.

In settled case denied the rumor to everyone : everyone can include people who didn't assert the rumor in the first place; addressee of second assertion, not speaker of first. In unsettled case denied x y or denied y to x : y has to be the person for whom "p (become) +" is beneficial.

Would we see the same thing for grant?

Other Wiki Pages

SynSem/PolymorphicVariadicPredicates (last edited 2018-04-20 19:55:47 by StephanOepen)

(The DELPH-IN infrastructure is hosted at the University of Oslo)