[matrix] Problem regarding Malayalam implementation

[Emily M. Bender]

Hi Anila,

You have turned up a bug in the integration of the coordination
library with the morphotactic infrastructure.  Your n1-left-coord-rule
is a lexical rule, but the customization system is not fully integrating
it.  A solution within your grammar would be to put
[INFLECTED infl-satisfied] on the DTR of that rule.  Meanwhile,
we'll look into a more general solution at the customization level.

Emily


On Tue, Sep 14, 2010 at 10:46 PM, A Varghese <vt.anila at gmail.com> wrote:
> Hi,
>
> n1-left-coord-rule licenses the uninflected noun lexeme:
>
> n1-left-coord-rule := infl-left-coord-rule & n-bottom-coord-phrase &
>   [ SYNSEM.LOCAL [ COORD-STRAT "1",
>                    COORD-REL.PRED "_and_coord_rel" ] ].
>
>
> The value of INFLECTION is specified as "-" in the parse charts of the wrong
> parses:
>
> INFLECTED: [inflected
>
>                         NUM-CASE2-FLAG: -
>
>                         …
>
>                         REST: null]]
>
>             REST: null]
>
> INFLECTED: [infl-satisfied
>
>                         NUM-CASE2-FLAG: na-or-+
>
>                         …]
>
>
> I added n-coord-phrase into malyalam.tdl and modified it as indicated below
> (to ensure that only nouns with like case get coordinated..) Without this
> change, sentences with multiple subjects coordinated+multiple objects
> coordinated were getting wrong parse trees (i.e., nouns with unlike case
> values were getting 'coordinated' in the parse trees).
>
> n-coord-phrase := coord-phrase &
>   [ SYNSEM.LOCAL.CAT.VAL.SPR < [ ] >,
>     SYNSEM.LOCAL.CAT.HEAD noun &
>              [ CASE #case ] ,
>     LCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD noun &
>              [ CASE #case ],
>     RCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD noun &
>              [ CASE #case ] ].
>
>
>
> The illicit parses involving uninflected lexemes exist with and without this
> change.
>
>
> Regards,
>
> Anila
>
>
>
>
> On Wed, Sep 15, 2010 at 4:05 AM, Emily M. Bender <ebender at u.washington.edu>
> wrote:
>>
>> Hi Anila,
>>
>> Those constraints on inflected and infl-satisfied look appropriate,
>> at first glance.
>>
>> What is the rule that licenses the constituent that becomes the
>> daughter of the coordination phrase in the illicit parses?  Look at
>> the parse chart, get the edge that corresponds to that daughter,
>> and look at its local avm (from the pop up menu in the parse chart).
>> What is the value for INFLECTED on that node?
>>
>> Emily
>>
>>
>> On Tue, Sep 14, 2010 at 11:54 AM, A Varghese <vt.anila at gmail.com> wrote:
>> > Dear Prof Bender,
>> > Yes, the uninflected nouns turn up only in coordination constructions.
>> > They
>> > cannot occur as non-coordinated subjects or objects in any of the
>> > sentence
>> > types currently specified.
>> > The noun inflection (number+case) for each class of noun lexemes is
>> > specified as obligatory. Only one noun inflection slot (with multiple
>> > morphemes) per noun lexeme class.
>> > The verbal inflection for each class of verb lexemes is obligatory, and
>> > specifies tense(+neg)/modal on the verb & case on the subject (and
>> > object).
>> > One verbal inflection slot per verb lexeme class, and (non-obligatory)
>> > question suffix as a separate verbal inflection slot that follows all
>> > (and
>> > only) verb inflection slots.
>> >
>> > In malayalam.tdl, "inflected" is specified as
>> > inflected inflected :+ [ NUM-CASE2-FLAG luk,
>> >     NUM-CASE3-FLAG luk,
>> >     NUM-CASE1-FLAG luk,
>> >     TENSE-NEG-MODAL3-FLAG luk,
>> >  ...   etc.. ]
>> > and "infl-satisfied" is specified as
>> > infl-satisfied :+ [ NUM-CASE2-FLAG na-or-+,
>> >     NUM-CASE3-FLAG na-or-+,
>> >     NUM-CASE1-FLAG na-or-+,
>> >     TENSE-NEG-MODAL3-FLAG na-or-+,
>> >   ...  etc.. ]
>> >
>> > Do I need to change these specifications in malayalam.tdl?
>> >
>> > Thanks and regards,
>> > Anila
>> >
>> >
>> > On Tue, Sep 14, 2010 at 9:47 PM, Emily M. Bender
>> > <ebender at u.washington.edu>
>> > wrote:
>> >>
>> >> Dear Anila,
>> >>
>> >> Do the uninflected nouns only show up in coordination
>> >> constructions, or can they show up elsewhere, too?
>> >>
>> >> There should be constraints on the feature INFLECTED
>> >> high in the phrase structure rule hierarchy that prevent
>> >> uninflected forms from appearing as daughters to
>> >> syntax rules, provided the matching/contrasting constraints
>> >> are on lexical types and any non-final lexical rules.
>> >>
>> >> The first thing to check would be whether your case
>> >> rules are specified to be obligatory.
>> >>
>> >> Emily
>> >>
>> >> On Tue, Sep 14, 2010 at 3:10 AM, A Varghese <vt.anila at gmail.com> wrote:
>> >> > Dear Prof Bender,
>> >> >
>> >> > My question is: How do I ensure that only fully inflected nouns (and
>> >> > not uninflected noun lexemes) get coordinated?
>> >> >
>> >> > Here's the problem in brief..
>> >> >
>> >> > The coordination rule allows uninflected noun lexemes to be
>> >> > coordinated (in subject/object position), often with multiple parses
>> >> > for these ungrammatical constructions. In the lexicon, the noun
>> >> > lexemes are specified only for person and gender; case and number are
>> >> > specified as noun inflections. (I use a hyphen as the "null affix"
>> >> > for
>> >> > singular nominative nouns.)
>> >> >
>> >> > I modified the coordination rules in malayalam.tdl to ensure that the
>> >> > CASE values of the mother node and the daughter are identical (in
>> >> > coordinate constructions). This eliminated the problem of multiple
>> >> > parses that were showing up earlier even for the grammatical
>> >> > sentences
>> >> > involving coordination. (The sentences with coordinated uninflected
>> >> > noun lexemes were being parsed even before the rules were changed.)
>> >> >
>> >> > The changes that were made to the rules won't appear in the choices
>> >> > file, so I'm not sending it. I've included some data below (in case
>> >> > you'd like to see the patterns that I'm referring to).
>> >> >
>> >> >
>> >> > Thanks once again!
>> >> >
>> >> > Regards,
>> >> >
>> >> > Anila
>> >> >
>> >> >
>> >> >
>> >> >
>> >> > kaLLan (thief): uninflected 3p, masc lexeme
>> >> > kaLLan-  : (3p, masc) + sing + nom
>> >> > kaLLan-maar: + pl + nom
>> >> > kaLLan-e : + sing + acc
>> >> > kaLLan-maar-e  : + pl + acc
>> >> >
>> >> > vaidyan (doctor)
>> >> >
>> >> > -umm : coordinating conjunction.
>> >> >
>> >> >
>> >> > Hari- kaLLan-e nookk-um  (Hari-nom thief-acc look-fut) is
>> >> > grammatical,
>> >> > and receives 1 parse.
>> >> >
>> >> > *Hari- kaLLan- nookk-um  (Hari-nom thief-nom look-fut) is
>> >> > ungrammatical, and is not parsed.
>> >> >
>> >> > *Hari- kaLLan nookk-um (Hari-nom thief-uninflected look-fut) is
>> >> > ungrammatical, and is not parsed.
>> >> >
>> >> > Hari- kaLLan-e-umm vaidyan-e-umm nookk-um (Hari-nom thief-acc-and
>> >> > doctor-acc-and look-fut) is grammatical; 1 parse.
>> >> >
>> >> > *Hari- kaLLan-umm vaidyan-umm nookk-um (Hari-nom
>> >> > thief-uninflected-and
>> >> > doctor-uninflected-and look-fut) is ungrammatical, but receives 1
>> >> > parse. (Some such constructions receive multiple parses.)
>> >> >
>> >> >
>> >> >
>> >> >
>> >> >
>> >> > On 9/14/10, Emily M. Bender <ebender at u.washington.edu> wrote:
>> >> >> Dear Anila,
>> >> >>
>> >> >> I hope to be able to help you, but don't have time to read long
>> >> >> documents.  Please distill your question down to the most pointed
>> >> >> query you can create (short and sweet).  Please send the query
>> >> >> in the body of the email.  I typically don't have time to look at
>> >> >> grammars,
>> >> >> so no need to send them.  (If you are working directly with the
>> >> >> customization
>> >> >> system, it can be useful to have the choices file.)
>> >> >>
>> >> >> Emily
>> >> >>
>> >> >>
>> >> >> On Mon, Sep 13, 2010 at 12:36 PM, A Varghese <vt.anila at gmail.com>
>> >> >> wrote:
>> >> >>> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>> >> >>>
>> >> >>>            Warning text added to this message by
>> >> >>>                  UW Information Technology
>> >> >>>                        help at uw.edu
>> >> >>>
>> >> >>>  ATTACHMENTS RENAMED
>> >> >>>
>> >> >>>  This message came to the UW with an attached file with a
>> >> >>>  name that ended in .zip or .exe.  Because files of this
>> >> >>>  type can automatically infect computers with a virus, the
>> >> >>>  attachment has been renamed.
>> >> >>>
>> >> >>>  o If the sender of the message is known to you, and you
>> >> >>>    were expecting the message, you need simply save the
>> >> >>>    attachment using the original name or save it as is and
>> >> >>>    rename it back to the original name on your computer.
>> >> >>>
>> >> >>>  o If the sender is not known to you, it is possible that
>> >> >>>    the attachment contains a virus and you may simply
>> >> >>>    delete the message.
>> >> >>>
>> >> >>>  o If this message claims to be official and
>> >> >>>    instructs you to open the attachment to get
>> >> >>>    important information, it is likely to be fake.
>> >> >>>    Virus writers are increasingly using sophisticated
>> >> >>>    social engineering techniques to mislead people.
>> >> >>>
>> >> >>>  UW-IT never sends important information about your
>> >> >>>  account or password in an email attachment.  Instead
>> >> >>>  you will be directed to a web page on a UW-IT site.
>> >> >>>
>> >> >>>  If you have further questions, please contact your
>> >> >>>  local computing support or
>> >> >>>    UW Information Technology
>> >> >>>    email: help at uw.edu
>> >> >>>    phone: 206.221.5000
>> >> >>>
>> >> >>>           Warning text added to this message by
>> >> >>>                 UW Information Technology
>> >> >>> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>> >> >>>
>> >> >>> Dear Prof Bender,
>> >> >>>
>> >> >>> I've not been able to resolve the problem related to coordination
>> >> >>> on
>> >> >>> my own, and I have to ask you for help.
>> >> >>>
>> >> >>> The details regarding the problem are in the attached MS Word
>> >> >>> document. I'm also sending you the entire Malayalam grammar (the
>> >> >>> choices file does not contain the changes I've made to the
>> >> >>> Malayalam.tdl file).
>> >> >>>
>> >> >>> Thanks in advance!
>> >> >>>
>> >> >>> (I hope I'm not bothering you with something really really trivial
>> >> >>> that I should have spotted on my own..)
>> >> >>>
>> >> >>> Regards,
>> >> >>>
>> >> >>> Anila
>> >> >>>
>> >> >>
>> >> >
>> >
>> >
>
>