GF-Seminar 8.12.25, Complementation of verbs v:VS by nominal objects np:NP

[Hans Leiss]

Hi Aarne,

let me add two comments to your remark that my suggested

Compl2VS : VS -> NP -> VP -- prove a conjecture ; know no secrets ;
-- believe your claims

is overgenerating in that *any* v:VS may be used with a nominal object.

(1) I don't see obvious examples v:VS where we cannot use at least some nominalization like "the fact|claim|concern that S" etc. instead of the object sentence. (That then there are many semantically inadequate np-objects is another, unavoidable cause of overgeneration.)

If the exceptions are few, it would be a lot of lexical work to split most v:VS into two verbs v:VS and v:V2 and omit the V2-variant for the few exceptions.

I just realized that you once had two type conversion rules

UseVS : VS -> V2 ; -- know (a secret)
UseVQ : VQ -> V2 ; -- ask (a question)

in Backward.gf. Clearly, my Compl2VS is a way of using UseVS:

Compl2VS vs np = ComplSlash (SlashV2a (UseVS vs)) np .

But UseVS has the advantage over Compl2VS that it provides trees for questions

what do we believe|claim|confess|know|stipulate ?

In fact, with UseVS we can form questions, for any v:VS and subject np,

(*) QuestSlash whatSg_IP (SlashVP np (SlashV2a (UseVS v)))

using "what" to ask for the *missing sentential complement*. The point is that questions in Grammar (except yes-no-questions QuestCl cl) have an interrogative constituent of type IP, IAdv or IComp, but Grammar has no *interrogative sentential pronoun* (what_IS) and no category of clauses missing a sentential|interrogative|infinitival object (cf. ClSlashS = Cl/S below). Hence,

SlashV2a (UseVS vs) : VPSlash

turns vs into a verb phrase missing a *nominal* object, which is then provided in (*) by the interrogative noun phrase whatSg_IP.

In Lang, there are verbs

fear_VS, hope_VS, know_VS, say_VS, know_VQ, wonder_VQ,

but of these only two have V2-variants, know_V2, fear_V2. So we get

Lang> p -cat=QCl "what do we know"
QuestSlash whatPl_IP (SlashVP (UsePron we_Pron) (SlashV2a know_V2))
QuestSlash whatSg_IP (SlashVP (UsePron we_Pron) (SlashV2a know_V2))

but not

Lang> p -cat=QCl "what do we hope"
The sentence is not complete

Rather than adding rules to combine nominal objects instead of sentential or interrogative (or infinitival) ones with non-V2-verbs, one might use the more general type conversion rules

UseVS : VS -> V2 ; -- know (a secret)
UseVQ : VQ -> V2 ; -- ask (a question)
UseVV : VV -> V2 ; -- try (an alternative method)
UseV2S : V2S -> V3 ; -- answer (them) (many questions what)
UseV2Q : V2Q -> V3 ; -- ask (you) (several questions | what)
UseV2V : V2V -> V3 ; -- promise (you) (my support | what)

and then accept questions like "what did they ask?" or "what did they promise you?". (UseVV and UseV2V should not be used for modal verbs, v:VV,V2V with v.isAux = True: *"what do you must")

To avoid these type conversions and still have "what_IS-questions", Grammar would need a category of clauses missing a sentential object, say ClSlashS (= Cl/S in categorial grammar notation, analog to ClSlash = Cl/NP).

The role of

SlashVP : NP -> VPSlash -> ClSlash = NP -> VP/NP -> Cl/NP

in the above questions "what do we know" in (*) is to "move" the subject np from the initial position of the clause, so that this position can be filled later with an interrogative pronoun by

(*) QuestSlash whatSg_IP (SlashVP np (SlashV2a (UseVS v)))

We would therefore need analog rules

SSlashVP : NP -> VPSlashS -> ClSlashS = NP -> VP/S -> Cl/S
QuestSlashS : IS -> ClSlashS -> QCl
what_IS : IS

to get a tree

QuestSlashS what_IS (SSlashVP np vs)

for "what do we know" etc. (Likewise with Cl/QS, VP/QS for interrogative and Cl/VP, VP/VP for infinitival objects, i.e. questions "what did they ask" and "what did they promise [to do]".) All this would be much work for a small gain, it seems, and a gain hardly needed for application grammars.

Rem. Because Grammar doesn't have categories like Cl/QS, I could not do what I wanted with interrogative sentential correlates as in "woran glaubt ihr" for glauben_VS. ("an was glaubt ihr" works for glauben_V2.)

(2) The second comment is that adding new categories VX, ... of (binary) verbs besides V2, VS, VQ, VV seems to afford either subtyping on the level of categories (like VX =< V2, VX =< VS) to inherit clause construction rules or a duplication of clause construction rules that use VX instead of V2 or VS etc.

But this seems a more drastic change of Grammar than adding complementation rules like Compl2VS : VS -> NP -> VP or type connversion rules UseVS : VS -> V2.

So I would prefer the overgeneration that results from UseVS etc. (It seems less overgenerating to me than the combination of sentential, interrogative and infinitival complements in the existing category SC, for example.)

Best,

Hans

[Aarne Ranta]

Hello Hans (and everyone interested in this topic),

Thanks for your thorough analysis of using VS verbs with NP complements! I have discussed this with colleagues and thought about it myself. My current conclusion is that, yes, indeed, VS verbs can be used with certain NP complements, in particular "it": "I know it", "I say it", "I believe it", etc. That some objects seem impossible, like "I say the house", need not be worse than the fact that the RGL does not implement selectional restrictions (domain restrictions) on NPs or any other category. Some other objects are possible, like "I say nothing", "I believe this story", etc. 

A subtlety comes from the fact that some similar-looking uses of verbs are more like different verbs. For instance, "I believe you", "I believe in science", are not instances of believe_VS. One way to see this is the multilingual perspective: in Finnish, for instance, all of them take different complement cases: "I believe it" accusative, "I believe you" partitive, "I believe in science" illative. Another, probably more familiar example is "I know you" with V2 vs. "I know it" with VS, where even the verbs are different in German (and Finnish): kennen vs. wissen. This makes me wonder if we need to introduce any inherent object case to VS even if we decide to use it with NP objects, or if these non-accusative cases actually belong to separate V2 verbs.

The most natural and powerful way to implement this seems to be

  SlashVS : VS -> VPSlash

in analogy with V2. Unfortunately, this name has already been taken by another function, so maybe SlashVSa or Slash2VS

  Compl2VS : VS -> NP -> VP

would follow from this, and also the wh-movement use would be made possible. The old suggestion

  UseVS : VS -> V2

now looks quite strange to me, even though it does the same job. The strangeness lies in that it is converting between lexical categories, and we expect lexical categories only to be populated by 0-argument functions.

As you write, also V2S, VQ, and V2Q seem to permit a limited range of NP arguments. I am less sure about VV and V2V, especially for modal VVs: ?"I can it".

In conclusion, I am growingly positive to most of your suggestions and would like them to be tested more widely in different languages.

With best regards

  Aarne

  

 

--

---
You received this message because you are subscribed to the Google Groups "Grammatical Framework" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gf-dev+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/gf-dev/20251210140204.E99087E1053%40marmolata.cis.uni-muenchen.de.

[Hans Leiss]

Hello Aarne,

thanks for your comments. I am currently extending the implementation I have (to V2S, wh-extraction and left-extraction of sentential objects, for example) and explaining it in the documentation on LangGer I‘m working on.

When this is done, I make it available and we can discuss it in detail.

Hans

per Mōbilé

Am 19.12.2025 um 14:14 schrieb Aarne Ranta <aa...@chalmers.se>:



To view this discussion visit https://groups.google.com/d/msgid/gf-dev/CAH881EGs_DttaMhLk78Y3ggD8PS0JTnObWfoOdriFK9V26eU1A%40mail.gmail.com.