delph-in / matrix

The Grammar Matrix

The Grammar Matrix

The Grammar Matrix is a DELPH-IN project for creating ("customizing") HPSG grammars for deep linguistic analysis. It has been used to teach university courses in grammar engineering as well as to jumpstart development on larger grammars, such as Zhong (Chinese languages) and INDRA (Indonesian).

NOTE: This repository is currently being ported from Subversion and converted from Python 2 to 3, so there may be some portions that do not work as intended.

Links and Resources

• Customization Page – a questionnaire for creating starting grammars
• The Grammar Matrix wiki – documentation for users and developers
• Grammar Engineering FAQ – help for grammar developers

Citation

Please use the following when citing the Grammar Matrix:

@inproceedings{bender-etal-2002-grammar,
    title = "The {G}rammar {M}atrix: An Open-Source Starter-Kit for the Rapid Development of Cross-linguistically Consistent Broad-Coverage Precision Grammars",
    author = "Bender, Emily M.  and
      Flickinger, Dan  and
      Oepen, Stephan",
    booktitle = "{COLING}-02: Grammar Engineering and Evaluation",
    year = "2002",
    url = "https://www.aclweb.org/anthology/W02-1502",
}
@inproceedings{bender-etal-2010-grammar,
    title = "Grammar Prototyping and Testing with the {L}in{GO} {G}rammar {M}atrix {C}ustomization {S}ystem",
    author = "Bender, Emily M.  and
      Drellishak, Scott  and
      Fokkens, Antske  and
      Goodman, Michael Wayne  and
      Mills, Daniel P.  and
      Poulson, Laurie  and
      Saleem, Safiyyah",
    booktitle = "Proceedings of the {ACL} 2010 System Demonstrations",
    month = jul,
    year = "2010",
    address = "Uppsala, Sweden",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/P10-4001",
    pages = "1--6",
}

Acknowledgments

This material is based upon work supported by the National Science Foundation under Grant No. BCS-0644097. Additional support for Grammar Matrix development came from a gift to the Turing Center from the Utilika Foundation.

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

matrix's Issues

request: change to Hierarchy class

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/5
• Reporter: lpoulson
• Created: 10/15/08 15:09:25
• Updated: 11/11/08 22:49:54
• Keywords: Hierarchy class
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

I would like a change to .save so that instead of assuming that the hierarchy members are descendant of top this ultimate supertype can be specified?

E.g., I would like to be able to generate tdl that looks like this:

mark := predsort.
value1 := mark.
value2 := mark.

In addition could that bit be optional? Then I could use the class for tense and aspect hierarchies where 'tense' and 'aspect' already exist as supertypes.

slot code: conflicts with assignment on INFLECTED -/+ values

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/27
• Reporter: lpoulson
• Created: 11/06/08 10:08:05
• Updated: 12/21/10 19:54:25
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Inflection -/+ conflict happened when I used forced and required at the same time (choices file attached). Kelly: A reminder from our discussion: You said that it was because verb-lex started out as inflected - and that your updated code has a fix. In the end the fix will need to allow us to force all verbs through inflection (even if slots are optional) and also the solution needs to work with the variation in verb hierarchies related to auxiliaries-or-not.

Multiple redundant copies of features written out

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/57
• Reporter: sfd
• Created: 02/01/09 00:14:25
• Updated: 03/16/09 03:38:31
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Load the attached choices file. Notice that it has multiple copies of some choices (e.g. noun5_feat1_name=person and noun5_feat1_value=1st). These persist even after the choices file is re-saved.

allow to clear answers on questionnaire

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/38
• Reporter: antske
• Created: 01/14/09 11:16:02
• Updated: 07/01/09 23:00:30
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

It should be possible to clear answers from the questionnaire when they are no longer relevant (for instance, auxiliary properties, when decided to leave the auxiliaries for later)

Customization System Robustness

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/598
• Reporter: goodmami
• Created: 01/25/10 11:46:11
• Updated: 01/25/10 11:47:22
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

We need to make the web form not crash when bogus (or even not-bogus) choices files are entered (uploaded or modified). There are several tasks:

1. Collect known problematic choices files (archived, GoPost, etc).
   
2. Create a semi-random choices-file generator (or update randgram.py)
   
3. Create a test harness that takes choices files and puts them through validation and customization, and recording those that pass validation but fail customization, or any that cause uncaught exceptions.
   
4. Fix the problems causing crashes.
   

Add instructions to unit test creation wiki for fragmentation or scope check

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/34
• Reporter: ebender
• Created: 12/04/08 11:40:51
• Updated: 12/04/08 11:40:51
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

We want to systematically check MRSs in new unit tests for well-formedness before checking the new unit tests (or updated gold standards) in. Need to add instructions to relevant bit of Matrix Dev wiki on how to do this. Fragmentation is [incr tsdb()] Trees > Annotate, scope check is probably in LKB Scoped MRS.

UG

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/1
• Reporter: ebender
• Created: 10/02/08 11:28:53
• Updated: 10/02/08 11:40:47
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Build comprehensive, implemented, correct Universal Grammar

Warnings and errors

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/8
• Reporter: ebender
• Created: 10/23/08 12:15:55
• Updated: 05/19/10 18:46:04
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

In addition to the *s indicating errors, we should have ?s indicating warnings (e.g., a lexical type with no subtypes or stems) *s and ?s propagate up to main page, but * and ? on a subpage gives only * on the main page. A choices file with only ?s can be submitted.

question and negation regression tests

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/30
• Reporter: lpoulson
• Created: 11/06/08 10:24:44
• Updated: 12/21/10 20:07:35
• Keywords: question negation
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

We need simple regression tests for questions and negation.

Refactor customize.py and validate.py

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/13
• Reporter: sfd
• Created: 10/23/08 17:30:47
• Updated: 05/31/11 14:00:38
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

customize.py and validate.py are getting really big, and they are currently structured as lists of functions with global variables. Two things to do:

1. (necessary) Split them into smaller modules in a consistent way
2. (desirable) Turn them into classes with members (i.e. object-orient them)

A reason not to do (2) is that it would require replacing "ch." with "self.ch." on a significant percentage (probably a majority in validate.py) of the lines of each file. It might be better to arrange the code in an object-oriented way without actually turning into classes.

Disallow reuse of type names in different verb types

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/47
• Reporter: ebender
• Created: 01/25/09 17:08:00
• Updated: 11/05/11 19:05:25
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Validation needs to check that the user has entered unique type names for each lexical type (etc). Sample problem choices file attached. The current behavior is that the information for all types with the same name is being added to one type with that name.

OSError on customization in questionnaire

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/695
• Reporter: goodmami
• Created: 07/18/11 17:36:53
• Updated: 07/18/11 17:38:18
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The calculation to get the location of matrix-core is incorrect for the questionnaire. It works for local customizations (python matrix.py c ...), but the directory is incorrect for live installations. This is a result of the change in r18751.

sytax errors in matrix.tdl

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/32
• Reporter: goodmami
• Created: 11/11/08 16:38:02
• Updated: 04/15/10 11:44:10
• Keywords: flop
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

There are a few syntax errors in matrix.tdl that the LKB seems to be cool with, but PET's flop doesn't seem to fancy too much.

Line 2055 (needs '&' after supertype)

raise-index-mod-lex-item := norm-ltop-lex-item
  [ SYNSEM.LOCAL [ CAT.HEAD.MOD < [ LOCAL.CONT.HOOK.INDEX #index ] >,
		   CONT.HOOK.INDEX #index ] ].

Line 2126 and 2129 (remove space between ':' and '=')

zero-arg-nonrel : = lex-item &
  [ SYNSEM.NON-LOCAL.REL 0-dlist ].

zero-arg-nonque : = lex-item &
  [ SYNSEM.NON-LOCAL.QUE 0-dlist ].

These changes will likely need to be copied over to the mmt grammars (can probably just replace matrix.tdl with the new version)

hang stars on labels

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/39
• Reporter: lpoulson
• Created: 01/15/09 19:47:24
• Updated: 08/28/09 17:41:19
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

If has-aux is chosen then at least one auxiliary must be entered. The right behavior is to place the star on the aux section of the Lexicon page if no auxiliary is added. However, there is no place to hang the star. The compromise has been to hang it as a warning on the has-aux choice on the word order page. This is not working and creating confusion. Scott and I decided that the actual fix would entail making it possible to hang stars on labels. Fix this.

Validation of features used

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/44
• Reporter: ebender
• Created: 01/22/09 13:11:36
• Updated: 11/05/11 17:41:36
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file demonstrates how users can currently create inconsistent choices files by defining features, using them in the lexicon section, and then going back and changing the features declared. The answer seems to be more aggressive validation that checks (on each submit) that every feature mentioned is properly defined.

Lexical entries with multiple super types

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/45
• Reporter: ebender
• Created: 01/24/09 23:14:33
• Updated: 03/16/09 03:38:17
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

If someone puts in the same stem in two different verb classes, we are generating (useless) lexical entries with multiple types (one for each class). The expected behavior is to generate unique lex-ids and make separate lexical entries. Sample choices file attached.

Valence-changing rules break grammar creation in matrix.cgi

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/851
• Reporter: cmc3c
• Created: 04/14/17 07:39:08
• Updated: 04/14/17 07:45:39
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

When a questionnaire includes valence-changing rules, matrix.cgi fails on creating a starter grammar. The symptom is Server Error: "malformed header from script. Bad header= [ SYNSEM.LOCAL.CAT.VAL.SUBJ : matrix.cgi"

Comment strings break compatibility with flop

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/31
• Reporter: goodmami
• Created: 11/11/08 16:27:04
• Updated: 04/15/10 11:45:17
• Keywords: flop comment-strings
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

There are a few places in matrix.tdl where comment strings are being used, and these cause problems for flop (in version 0.99.14svn). Comment strings are an uncommon feature and do not seem to be catching on with other groups, so it is probably best to just move these to regular comments.

For example,

anti-synsem := unexpressed &
"A contrasting type to ordinary synsems (expressed
and unexpressed) which is sometimes useful.  No longer
explicitly used in the Matrix.".

should instead be:

;A contrasting type to ordinary synsems (expressed
;and unexpressed) which is sometimes useful.  No longer
;explicitly used in the Matrix.
anti-synsem := unexpressed.

The following places have comment strings:

line 176 (anti-synsem)

line 463 (semarg)

These changes will likely need to be copied over to all mmt grammars.

Multiselect on auxiliaries

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/49
• Reporter: ebender
• Created: 01/25/09 21:11:02
• Updated: 04/08/09 12:34:58
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Allow users to select multiple values for features in complements of auxiliaries (and do the right thing in terms of creating supertypes where necessary), as is done with e.g., verbal inflection.

Restrictions on aux- word order

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/21
• Reporter: antske
• Created: 10/28/08 17:59:39
• Updated: 12/12/08 13:51:34
• Keywords: word order aux-comp
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Bug: when aux-comp order differs from general head-comp restrictions, the customize sets the head of the aux-phrase to 'aux' for head-comp (which is not a head) or 'verb' for comp-head (which does not have the desired effect). I changed 'aux' to 'verb', but more fixing seems required.
Setting constraints to AUX +, is a step in the right direction, but would still not parse
SOauxV and would parse SauxVO in an SOV aux-preceding language.

Clarify verb types

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/59
• Reporter: ebender
• Created: 02/01/09 20:14:06
• Updated: 03/16/09 03:39:57
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Some of the students are defining a case system and then taking the case-less (quirky-case) types for their verbs. I think it might be possible to clarify this with some prose...

Negation leads to circular rules

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/62
• Reporter: ebender
• Created: 02/16/09 14:17:05
• Updated: 12/21/10 20:24:09
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Attached choices file creates a grammar that won't load because of bad supertypes on negation lex rules.

Date stamp

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/35
• Reporter: ebender
• Created: 12/04/08 11:45:21
• Updated: 03/16/09 03:32:35
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The code that looks for the date stamp needs to work fine when there is none.

Allow multiple arg-comp-aux-with-pred

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/52
• Reporter: ebender
• Created: 01/26/09 09:24:30
• Updated: 03/31/09 14:08:24
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file illustrates the case where someone wants to have multiple different auxiliaries that are mostly the same except have different FORM requirements. In the current system, this leads to one type (arg-comp-aux-with-pred) with inconsistent constraints on its complement's FORM value. Instead, there should be subtypes to house the FORM constraints.

question code broken (with addition of main-verb hierarchy option)

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/29
• Reporter: lpoulson
• Created: 11/06/08 10:23:18
• Updated: 10/28/11 16:07:40
• Keywords: question
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Question code seems to be broken due to the addition of the main-verb hierarchy added when there are auxiliaries.

slot name tense colliding with feature tense

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/28
• Reporter: lpoulson
• Created: 11/06/08 10:19:58
• Updated: 02/19/09 17:10:57
• Keywords: slot code
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Kelly, per our discussion: ??track feature values can collide with other feature values. This was my note but I wrote another note with the specific example: slot name tense colliding with feature tense - I assume these two notes are related.

Lexical type hierarchy improvements

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/9
• Reporter: ebender
• Created: 10/23/08 12:19:23
• Updated: 01/05/12 12:20:50
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Each type should have a supertypes field (and no longer valence ... these become provided supertypes). Each type should have either subtypes or stems (and if not, get a warning, not an error on validation).

A type can only have subtypes or stems, not both.

Validation will check that each lexical type with stems inherits from all of the required dimensions. By default, this includes valence (for verbs, at least). It can also include others, as specified on the features page:

When someone defines a feature, they get asked: Do all lexical entries have to make a decision on this one (+ or -), or can some be underspecified (+, -, or bool)? (Analogously for non-boolean features).

Validation checks that each of these features is specified: on the type with stems? if not, on its supertype(s) [up the tree]? If not, then *.

Case-marking adpositions can't be optional

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/51
• Reporter: sfd
• Created: 01/25/09 21:53:51
• Updated: 03/16/09 03:34:29
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Unlike case-marking via inflectional morphology, there's no way to describe case-marking adpositions as optional. The questionnaire should support this and the created grammars should behave as described.

Redudnant supertypes from lex rule component

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/48
• Reporter: ebender
• Created: 01/25/09 21:05:53
• Updated: 06/15/11 20:47:44
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Something is amiss in the way that the *-dtr types are assigned. The attached choices file generates a grammar with redundant supertypes (which does not load).

Disallow resuse of matrix type names in type fields

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/46
• Reporter: ebender
• Created: 01/25/09 16:59:16
• Updated: 11/05/11 19:01:35
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Validation needs to check for name-space clashes such as "aspect" as a subtype of "form".

Namespace validation

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/18
• Reporter: sfd
• Created: 10/23/08 17:49:50
• Updated: 11/05/11 19:02:39
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Validation needs to be smarter about names and namespaces.

1. Multiple lexical items with the same orthography should produce multiple lexical entries like "ORTH_1" and "ORTH_2", not merged lexical entries.
2. The names of features and of lexical types must be either validated to prevent illegal characters or stripped of those illegal characters, and they must also be validated to make sure they're unique. [EB: Illegal characters include :, ?, ], ". Some of these cause problems for python, some for the LKB.]

Updated by EB:

3. Slots and morphemes should not be allowed to have the same names (current system, takes attached choice file and creates a lexical rule type which inherits from itself).

FORM requirement for auxiliary complements

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/43
• Reporter: anonymous
• Created: 01/21/09 13:18:26
• Updated: 04/16/09 09:48:30
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Research this issue and if no compelling argument against, change the FORM value requirement for auxiliary complements to be optional (i.e., no requirement as an option) and remove finite/nonfinite default for languages with auxiliaries.

Move repeated constraints to rule supertypes

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/61
• Reporter: ebender
• Created: 02/16/09 13:43:59
• Updated: 02/16/09 13:43:59
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file leads to a grammar with lots of repeated constraints on the lexical rules. The problem is that there are portmanteau morphemes which mark tense and (redundantly) subject/object properties. It would be nice if we could either allow the user to define and cross-classify types for each of these dimensions, or if we could automatically notice the redundancy and posit the appropriate supertypes.

Colons in input not validated, lead to python error

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/42
• Reporter: ebender
• Created: 01/19/09 10:50:49
• Updated: 08/28/09 17:42:03
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Students in 567 tried using colons as part of their word spellings, and this led to python errors. We need to either validate out this choice or somehow escape the colons in the strings. Sample choices file attached.

Bug in choices.py old choices files don't load

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/623
• Reporter: antske
• Created: 04/13/10 12:58:13
• Updated: 04/14/10 09:15:52
• Keywords: choices files not loading
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Some older choices files cannot be loaded by the customization system anymore.
Most likely due to a bug in choices.py.

[INFLECTED -] over applied

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/58
• Reporter: ebender
• Created: 02/01/09 20:03:28
• Updated: 12/21/10 13:30:35
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file ends up with [INFLECTED -] on verb-lex as well as several of its subtypes, but not all. I think that the intention of the grammar writer was to have auxiliaries not go through any lexical rules. I can't tell just now if this is a customization bug or a lack of clarity in the UI.

run-unit-tests.sh should call install

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/36
• Reporter: ebender
• Created: 12/04/08 11:47:44
• Updated: 03/16/09 03:32:58
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

to make sure that local copy of matrix-core is up to date.

i-length of seed strings

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/19
• Reporter: anonymous
• Created: 10/23/08 21:08:19
• Updated: 10/23/08 21:08:19
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The i-length of seed strings needs to be entered into MatrixTDB so we can manipulate strings by length in [incr tsdb()]... this probably involves both entering a length for seed strings and modifying that length with stringmods.

update to unit-testing (cvs->svn)

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/6
• Reporter: lpoulson
• Created: 10/17/08 10:58:13
• Updated: 11/11/08 22:47:39
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The piece of code written to add tests to cvs needs to be changed to svn. In addition the wiki directions need updating to reflect that.

"=" in text fields

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/17
• Reporter: sfd
• Created: 10/23/08 17:45:34
• Updated: 11/11/08 22:48:09
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Go to the questionnaire and enter any string containing "=" into any of the form fields, then watch the web questionnaire go CHOKE-CHOKE.

run-unit-tests automatically refreshes matrix-core

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/60
• Reporter: ebender
• Created: 02/13/09 10:47:08
• Updated: 03/20/09 14:41:13
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The unit test scripts should ensure that the files in matrix-core are up-to-date with the files checked in to svn.

Update: What we want is for the matrix-core to match what's in the local working copy of matrix/ not svn. Scott did this change.

negation and infl-ltow-rule interaction broken

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/56
• Reporter: ebender
• Created: 01/28/09 17:38:58
• Updated: 07/01/09 13:18:36
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file leads to negation being an infl-ltow-rule, which suggests that my integration of negation into the current lexical rule scheme is not fully correct. I think the real solution here is to finally admit that ltow doesn't entail add-only-no-ccont and make those independent.

Mixed Conjunctions

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/50
• Reporter: sfd
• Created: 01/25/09 21:48:23
• Updated: 08/28/09 17:43:10
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Create a grammar with two coordination strategies, one for N and NP marked with KA, one for VP and S marked with WA (this is pseudo-Ainu). The resulting grammar will accept KA anywhere WA is expected and vice-versa, because although the coordination rules are marked with COORD-STRAT to prevent them from interfering with each other, the conjunctions aren't distinguished in any way.

Coordination bug (CFORM)

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/55
• Reporter: ebender
• Created: 01/27/09 21:54:11
• Updated: 08/28/09 17:43:25
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file specifies two coordination strategies, and everything looks fine except that both of the conjunctions end up with CFORM "1", so only one of them ends up working.

old choices file cannot be uploaded

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/655
• Reporter: antske
• Created: 11/24/10 09:34:18
• Updated: 12/21/10 20:02:43
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Old choices file (leading to grammar in older version) can currently not be uploaded. I have no recollection of manually editing the choices file, but could not confirm this with 100% certainty...

Upper case on TRAC feature names

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/53
• Reporter: ebender
• Created: 01/27/09 16:36:27
• Updated: 12/21/10 20:34:42
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The TRAC features that get added by the customization system appear to take their spelling orthography from the slot names, and are thus ending up as lower case. The convention in tdl is for feature names to be in upper case, so the strings should be modified appropriately. In addition, we need to be sure to validate these names so that they don't end up colliding with any other feature names.

Coordination bug (INFLECTED)

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/63
• Reporter: ebender
• Created: 02/20/09 21:29:59
• Updated: 06/16/11 18:59:52
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file leads to infl-left-coord-rule and infl-bottom rules which require [INFLECTED -] daughters. However the coordination markers attach in this language to forms that are perfectly fine as stand-alone words (i.e., [INFLECTED +]).
This constraint appears to be on the matrix.tdl coordination types.

different adpositions marking the same case

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/20
• Reporter: antske
• Created: 10/28/08 17:47:11
• Updated: 03/16/09 03:31:21
• Keywords: case marking adp
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

when adding several case marking adpositions to the lexicon that mark the same case, the output in the lexicon becomes for instance:

acc-marker := case-marker-p-lex &
[ STEM < "add_1" & "add_2" >,
SYNSEM.LOCAL.CAT.HEAD.CASE acc ].

with the consequence that the adposition is ignored altogether.
Maybe change this to
acc-marker-1 :=
acc-marker-2 :=

Counter-intuitive notion of slot optionality

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/54
• Reporter: ebender
• Created: 01/27/09 16:45:49
• Updated: 06/16/11 00:39:54
• Keywords:
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

The attached choices file shows what I think is a problem with our notion of optional lexical rules. The grammar that is generated is supposed to be able to parse sentences like the following:

naan naDa-kka-lle
I.NOM walk-INF-NEG
'I did not walk.'

However, the verb form is ending up as [INFLECTED -], and only going through ltol rules. I think what's going on is that the negative forms don't involve some slot that is marked as non-optional on other forms. The solution I've suggested to the student whose grammar this is is to make the negative morphemes (counterintuitively) non-optional, but maybe we can make this clearer somehow?

Enhance creation of verb type hierarchies

Migrated from Trac:

• Original: http://lemur.ling.washington.edu/trac/matrix/ticket/24
• Reporter: lpoulson
• Created: 11/06/08 09:55:02
• Updated: 10/28/11 16:01:59
• Keywords: verb hierarchies
• Notes:
  • There may be formatting problems from the conversion
  • Comments on tickets are not migrated! See the original issue.

Currently all verb types defined must specify trans or intrans - this
means that most verb classes will have to be defined twice. For each feature added the number of types that need to be added (roughly) doubles. This gets old fast. Change the code to "intuit" supertypes or otherwise assist in the automatic creation of deeper hierarchies.