[Prev][Next][Index]

[XTAG Summary]: feature errors


Minutes from the XTAG meeting on 22nd July 1999
-----------------------------------------------

Errors found in XTAG feature structures after type-checking against a
signature. 

There were four main types of errors found:

1. Grammar-based: ambiguous names, or ambiguous functions for
   features. 
2. Typos.
3. Undocumented.
4. Incorrect or obsolete feature.

The errors discussed during the meeting were the following. There are
a few others in the list, but they are of the same types as the ones
below. 

1. <assign-comp> has some non-overlap in feature values compared to <comp>
   <assign-comp> extras = { ecm, ppart_nil, none }
   <comp> extras = { nil }

   - type of error: possibly ambiguous, grammar-based
   - fix: possible duplication of effort for <assign-comp>
   - cannot fix without detailed knowledge of <assign-comp>

2. <case> = nom ==> nominative
   <mode> = nom ==> nominal/adjectival predicate

   - type of error: typo
   - fix: change <mode> = nom to <mode> = pred

3. <rel-pron> has overlap with <mode> but has extra values
   <rel-pron> extra values = { ppart, ger, adj_clause }

   - type of error: possibly ambiguous, grammar-based
   - fix: dunno

4. <relpron> exists alongside <rel-pron>

   - type of error: typo
   - change <relpron> to <rel-pron>

5. <disc-conj> not documented

   - type of error: documentation
   - fix: add to tech-report
   - prolly stands for discourse conjunction

6. <predet> occurs in noun and det signs

   - type of error: incorrect feature
   - fix: should be removed since Carlos thinks it was widely removed
   - from the grammar at some point and remainder were skipped

7. <gen> occurs ambiguously, stands for 'gender' as well as 'genitive'
   
   - type of error: ambiguous, grammar-based
   - fix: should change all occurences of <gen> in noun signs to
   - another name, say <gendr>

8. In one tree, NP.b:<assign-case> = N.t:<case>

   - type of error: typo
   - fix: change <assign-case> to <case>

9. <punct> is defined as a complex feature, e.g. <punct contains
   comma> but instances of <punct> are found in the grammar.

   D_r.b:<punct bal> = Punct_1.t:<punct>

   - type of error: incorrect feature
   - fix: dunno

10. In tree PsPu:

   S_f.t:<assign-comp> = ind/...

   - type of error: unknown
   - check lexical features to see if 'ind' is allowed for
   - <assign-comp>

11. Should we have case on sentential subjects?

   - type of error: grammar-based
   - fix: remove all instance of <case> = nom or <assign-case> = nom
   - in sentential subject tree families.