[matrix] problems with some linguistic issues

[Emily M. Bender]

Dear Altaf,

Thank you for posting questions here.  I will try to provide
useful answers.

> 1. In Bangla, a transitive or ditransitive verb can take another
> 'non-finite-verb' as its complements which could be corresponding to PP
> complement in English. For example a sentence: "naafid-ke jete balbe"
> 
> "Naafid-acc
> to-go tell"
>                                                                        "Tell
> Naafid to go"
> 
> For this example, the ditransitive verb 'balbe' takes a verbal complement
> which is in non-finite form 'jete', while in English the ditransitive verb
> takes a PP complement 'to go'. But in the matrix customization script, the
> transitive verb can only takes NP or PP as complements. Do I need to include
> seperate rules for transitive verbs to take non-finite verbs as its
> complements?

You will need to define additional types in bagla.tdl for verbs like
'balbe'.  Fortunately, there are useful types defined already in matrix.tdl
that you can inherit from, they are just called on (yet) by the
customization script.  Just as transitive-verb-lex in bangla.tdl
inherits from transitive-lex-item and verb-lex, you'll want to
define a new subtype (maybe subj-obj-raising-verb-lex) in bangla.tdl,
and have it inherit from verb-lex and ditrans-second-arg-raising-lex-item.
(For the definition of those types, please inspect  matrix.tdl.)

Once you've defined the type, you'll need to put additional constraints
on its arguments to ensure that first complement is an NP (and
not an N-bar, say) and that the second is a VP (and not an S or a V).

Note that 'tell' as it is used in "Tell Naafid to go" in English
is also of this type:  that 'to' is the infinitival marker 'to' and
not actually a preposition.

> 2. As for english, modal verb 'can' is called the subject raising auxiliary.
> 'Bangla' has a auxiliary 'hay', but this auxiliary doesn't raise the subject
> just like English. For example: "naafid-ke balte hay"
>                                                         "Naafid-acc to-tell
> be"
>                                                         "Naafid has to tell
> / Naafid has to be told"
> 
> This example shows that: there could be two interpretations for the same
> sentence (Naafid could be a 'subject' or 'object', regardless of the 'acc'
> case marker). Actually, in 'Bangla' the actual interpretation is
> disambiguated by the context. But, if there is any other main verb, instead
> of that auxiliary, there will be no ambiguities because the noun 'Naafid'
> will be an object (accusative).
> For example: "naafid-ke balte jaay"
>                    "Naafid-acc to-tell go"
>                    "(Somebody) goes to tell Naafid"
> 
> Here, a finite main verb 'jaay' is used instead of auxiliary 'hay'. The
> point here is that, when auxliary 'hay' is used, the NP in that sentence
> 'could be' a subject while ignoring the context, not 'must be' (because in
> English, auxiliary 'can' ensures a subject to be exist in the sentence).
> >From this point of view, can I select 'hay' as a "subject raising 
> >auxiliary"
> in the matrix customization script?

>From a quick read of this description, it seems like to me that 'hay'
is a subject raising auxiliary, and that Bangla also has (at least)
subject pro-drop.  The customization script doesn't (yet) provide
analyses of argument drop, but again, there is support within the
matrix itself.  For details on how to add it to our grammar, see
this lab from Ling 567:

http://courses.washington.edu/ling567/lab6.html

> 3. Why "neg-add-lr" rule used for? For a testing purpose, I have downloaded
> a matrix for free word order and I tried to parse some test sentences. When
> I tried to parse the sentence "naafid maare" (Naafid beats), it couldn't
> parse. As I saw from the parse chart: the verb 'maare' (beats) goes to the
> rule 'neg-add-lr', but that type didn't participate into some of the root
> nodes (such as: subj-head, head-comp) . For the convenience, I have attached
> a copy of the file 'choices'.

You got neg-add-lr in your grammar because you specified that one of
the negation strategies is a selected adverb.  The neg-add-lr is a lexical
rule which adds a complement to the beginning of the verb's COMPS list
and changes the semantics appropriately.  The rule will show up in the chart
whenever there is a verb which could be its daughter, but if there is
no negative adverb around for that (COMPS-augmented) verb to pick up as
a complement, that edge won't spawn any further edges.

Please let me know if you have any further questions!

Emily