Skip to content

HarmonyTop

FrancisBond edited this page Jun 23, 2020 · 6 revisions

Semantic Harmonization

In order to make processing using MRSs from different grammars more compatible, we try to do stick to some naming conventions for predicates. In addition we propose a set of shared universal super types. These are used by all grammars in the LOGON machine translation architecture and are kept in the file predicates.tdl.

Predicate Names

These follow the RMRS naming convention.

_lemma_pos_sense_rel  (lexically introduced predicates)
sense_rel            (abstract predicates introduced by constructions)

Universal Supertypes

There are some predicate types used in the decomposition of common words, that we propose could be shared among all DELPH-IN grammars. Currently (2008-08-13) there is some variation among grammars:

ERG Example JACY SRG
manner_rel how manner_rel manner_n_rel
named_rel Dan named_rel named_rel
person_rel who, someone person_n_rel person_rel
place_n_rel where, here place_rel place_n_rel
reason_rel why reason_rel ???
thing_rel this, something thing_rel thing_rel
time_n_rel when, now
which_q_rel what, when, ...
nominalization_rel nominalization_rel nominalization_rel
  • DanFlickinger and FrancisBond will prpose a set of univerals and

    • send to the matrix-list
  • After consensus is reached, these will be included as part of the matrix

Syntactic Harmonization

There seems to be less to gain from trying to synchronize type-names across grammars, although it is an interesting research question.

  • Try to adopt Matrix types and names where possible
  • Anything used in post-processing should be the same if possible
    • e.g., generic types

Files and Directory Structure

  • Writing scripts is made harder by small differences in naming.

    • Version.lsp (ERG, Jacy, SRG); Version.lisp (GG)

    • script, ascript, _script, ...

  • Solution: (any volunteers?)

    • Try and define a conventional directory structure for grammars
    • Exemplify in the matrix
    • Document in the wiki
    • Only support conventional directory structure in scripts

Version.l(i)sp structure

Try to go for something like "Grammar Variant (date)", where date is YYYY-MM-DD

  • e.g. "ERG mal (2020-10-16)"

ERG will probably shorten to just YYYY for stable releases:

  • e.g. "ERG mal (2020)"
Clone this wiki locally