HarmonyTop

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)"