The flag indicating whether a NP has a negated determiner is now encapsulated in the algebraic parameter Agr, removing the extra field in the NP linearization type.
Extends the previously introduced support for specifying the verb valence (verb-dependent subject and object cases), involving functions up to PredVP and RelVP.
In Latvian, the passive voice is not used if the agent (subject) is known; to preserve the information structure (i.e. the word order), a clause like 'A is <done> by B' is linearized in the active voice ('A <does> B') where A has the object case (e.g. Acc), and B - the subject case (e.g. Nom). Thus, the verb valence patterns are swapped on-the-fly.
This is still a rather quick & dirty implementation: parameters and linearization types have to be optimized (VerbLav), the use of PassV2 in AdvVP is problematic as it doesn't apply VPSlashPrep / ComplSlash, etc.
Agreement with the focus part (the object) works not only for Pers3, but also for Pers1 and Pers2 NPs (if the verb requires non-typical subject/object valences).
Object-dependent double negation works (in addition to the subject-dependent double negation).
The first one is motivated by PhrasebookLav, the second one - by AttemptoLav.
Also a couple of minor fixes.
Double negation: sebject-dependent - works, object-dependent - still has to be fixed (in VerbsLav).
Structural words: everybody, somebody, nobody.
Both motivated by AttemptoLav.
The case of the topic part of a clause (~subject) can now depend on the verb, allowing for less frequent agreement.
E.g. "man[Dat] garšo pica[Nom]" ("I like pizza").
The default case for the topic/subject remains the nominative case.
- Few bug-fixes (agreement between Pron and ComplAP; C1 verb paradigm; how8much_IAdv / how8many_IDet).
- Differentiation between male and female pronouns.
- Irregular verbs now can have prefixes.
- Code refactoring.
