LadUW20161020

Questions for today:

Q1 Working with qeq

When defining a specisl SHI which emphasizes the AUX verb of the clause, we put a qeq in in CONT.HCONS at the node of AUX, linking the LARG to the LTOP of main verb. This works fine, but when the AUX is preceded with negation adverb BU, the adv is supposed to be combined with AUX first, also adding in a qeq. Then how to identify the correct qeq to perform the linking?

shi-aux-dep-lex := shi-other-dep-lex & "linking aux and verb up in qeq. <ex>他 是 会 哭 的 <ex>他 是 不 会 哭 的

• [ SYNSEM.LOCAL.CAT.VAL.COMPS < [ LOCAL [ CAT [ HEAD verb & [ AUX + ],

  • VAL.COMPS < synsem > ],
  • CONT.HCONS <! qeq & [ LARG #larg ] !>] ],
  • [ LOCAL.CONT.HOOK.LTOP #larg ] > ].

Q2. When there are two missing arguments...

Recall this example of alternative object position in SHIDE construction. (106) 他 是 昨天 完成 的 任务 tā shı̀ zuótiàn wánchéng de rènwù he is yesterday finish DE task “It was yesterday that he finished the task” We attempted to define a special DE that takes two complements, one before the head, one after the head and it works fine.

de-4shi-lex := lex-item & raise-sem-lex-item & no-icons-lex-item & norm-zero-arg & "Special DE used in ShiDe structure for alternative location of object of a transitive verb. Feature CH + means this COMP is before the head, CH - means it's after the head. <ex>他 是 昨天 完成 的 任务 Currently using OPT+ to stop this DeP from being treated as an actual modifier for noun. In rule head-adj-int-phrase, constraint [ OPT na-or-- ] is used to exclude this DeP. "

• [ SYNSEM.LOCAL.CAT [ HEAD postp-comp &

  • [ MOD < #subj & [ OPT + ] > ],

  • VAL [ SPR < >,

    • SPEC < >, SUBJ < >,

    • COMPS < [ LOCAL [ CAT [ HEAD verb & [ COP - ],

      • VAL [ SUBJ < #subj >,

        • COMPS < #obj > ] ] ],
      • CH +, OPT - ],

      • #obj & [ LOCAL.CAT.HEAD noun,

        • CH -,

          OPT - ] > ] ] ].

However when this alternated construction is combined with topicalization, the object is shifted to the sentence front. This can be handled with current rule, provided that we can slash the object and pass it on to the level of SHI clause. But in order to let the special DE to be used with emphatic SHI, the DeP also needs to identify the missing subject so that it can be linked to the emphasized noun after SHI. This leads to the need to keep two slashed argument at the level of DeP, somehow. SLASH can only keep one argument at most. Any other creative solution?

• \ex \label{ex:shi-v3-alt} \begingl \gla 任务 是 他 完成 的 \glb \ren4\wu4 \shi4 \ta1 \wan2\cheng2 \de5 \glc task is he finish DE \glft It was him who finished the task \endgl \xe

NOTES on Q1

Emily: You're digging in the HCONS list, which you should not be doing. Zhen Zhen: What is the better way? Emily: That auxiliary is saying something about its complement, so I'm not sure you need to differentiate... perhaps the "shi" should say "whatever the auxiliary's complement should be, is my complement." What this "shi" needs to do is behave like the HEAD-COMP rule, and identify the AUX comp as its own second complement. Beware that it might make the diff-list appends twisty.

A model for this is if you customize a new grammar and use the V2 option -- the verb behaves similarly. What you're having "shi" do here is called argument composition in the literature. You'll find you can have a few fewer entries for "shi" this way.

If you customize a grammar with V2 and auxiliaries you'll see an example of argument composition with auxiliaries.

NOTES ON Q2

Emily: It has its subject missing in the sense that it hasn't found it. That doesn't necessarily mean that it's extracted.

There's a data question here, which is "Can you construct an example with a topicalized subject, and then the object, and then shi (the adverb)"

Zhen Zhen: Yes.

Emily: That's the key that you need for two SLASHes. The Matrix inherited the constraint that there can only be one SLASH from the ERG.

So now you have to see what happens when you remove the 0-1 dlist constraint on SLASH. Which dovetails with something that Laurie and I were talking about with coordination! (long discussion ensues about using diff-lists to do number resolution)

Francis: We would appreciate pointers to people who have tried multiple SLASH implementations of things.

David: Berthold. He gave a presentation on focus in Hausa, and as I recall he had multiple things rattling around in SLASH.

Emily: Another place is multiple-wh movement in Slavic languages. Look in the HPSG bibliography for WH/Slavic. CoreGram might also have a Slavic language in the mix? Stephan Muller? You may also look at his new open access book.

David: Aha! Berthold uses a SLASH-2 list. Don't do that.

Emily: Don't do that.

So try backing off from 0-1dlist and see what happens if you only use dlist. And in the meantime, we might dig into Dan's memory of why he put those things into the ERG that we inherited in the Matrix.

Zhen Zhen: Why is it not advisable to have a separate SLASH list?

Emily: Part of it is theoretical. Once something is extracted, nothing differentiated about where it came from.

Francis: You'd have to have two of a lot of rules.

SLASH acts a bit like a stack here, we can push things on and pop them off as needed.

Zhen Zhen: So the order of extraction matters?

Francis/David: Yes.

Emily: A diff list is a container structure for lists that allows us to muck around with appends. LAST points after LIST, which allows us to append things.

You can't push an indeterminate number things, and you can't pop an indeterminate number of things. Rules need to specify how many.

David: So how do we check that a diff list is empty?

Emily: An empty diff list is one where the LIST and LAST are unified. If you try to impose that constraint, sometimes it can happily unify even if it shouldn't. That's where Dan is doing some magic with these types (e.g. 0-1dlist)

Let's take a look at the root node -- it probably does say 0dlist, doesn't it?

In Slavic languages, the rule says the dlist is now 1 shorter (maybe?)

Francis: I think we can enforce the order by checking the SUBJ and COMPS with the rules.

Emily: Yes, by checking in the COMPS rule that the SUBJ list is non-empty, HEAD-SUBJ can check that COMPS is empty.