Lab 4 Writeup

I've made a couple changes to my orthography/representation.  I decided to not use a period to mark the retroflex r, and use a tilde instead: r~.  Also, I decided to use a grave accent ` instead of the single quote ' to mark the implosives and ejective (now: d`, b` and `k).

For the Word Order page, I chose SVO for Hausa.  I marked that Hausa had determiners preceding the noun, and used the demonstrative "that" ('wata', f.) for the sample setences.  Definiteness in Hausa is marked by an affix: a low tone on the ending vowel followed by -n or -r(retroflex) (m/f).  Pronouns, consonant-final nouns, or loan words are followed by the connector word d`i- and this connector gets the definiteness suffix.  I don't think I'll be dealing too much with these examples, but I might have to tweak the grammar later on if I need to include this d`i- form.   

I skipped Setential Negation, because while Hausa has adverbs which are used to negate the whole sentence, none of the options seemed to fit the positioning of these adverbs. The options given were to the right, left or either side of the category it modifies. Hause has both adverbs appearing at the start and finish of the sentence they're negating.  I thought that the option for the adverb appearing on either side might work, but the sentence can't just have one negative particle, it has to have both.

For Coordination, I choose Polysyndeton, though Hausa also allows optional N-Polysyndeton (and occuring in front of the list of conjoined items).  Are we going to consider 'or' (disjunction)? It behaves differently, in that it IS allowed to connect sentences (where 'and', 'dà' isn't)).

I left blank the section on Yes-No Matrix Questions. Hausa uses four different methods of marking questions, from intonation, lengthening the last vowel of the last word, and adding various words (like the word or, 'koo')either sentence-initially or sentence-finally.  I'm not sure which is used most often, as the reference grammar just describes all four.

For the basic lexicon, I added two nouns('girl' and 'soup'), a demonstrative ('that'), and two verbs ('came','cooked') with the appropriate-inflection PAC attached as a prefix.  Auxiliaries would be difficult with a limited vocabulary since they take a inflectional PAC, but their complement VPs don't (I'm counting these verbs as having a 'zero' form PAC). 

When I first tried to parsing one of the sample sentences from the starter-grammar, it didn't work because of the format I used to list the verbs.  My testsuite was set up to separate the PAC from the verb with a dash and didn't accept that as one word.  The basic sentence I was trying (that girl came, 'wata yaarinyàa taazoo') was covered by the head-complement rule, and parsed as I expected it to.  It had a NP-VP tree structure, with the NP expanded into a D and N.  I'm not sure why it produced two parses, and I can't see the difference from the parse chart, because both parses have a HEAD-SPEC being the mother of DETERMINER-LEX and NOUN-LEX.

Coverage, as measured by itsdb:
My test suite had 75 examples, with 33 positive (grammatical) and 42 negative (ungrammatical).  When I ran it through itsdb, there were parses generated for 5 sentences.  These were 'the man built a house.', 'the house built the man', 'the girl cooks soup', 'the waterpot broke' and 'some old woman came'.  I just realized that 'the house built the man' is ungrammatical because house is feminine, and it can't replace man and keep the same verb inflection.  I Out of 33 possible positive sentences, this is 15.2%.  Of the negative examples, none parsed.  When I change the grammaticality judgment on the sentence I mentioned ('the house built the man'), there will be one parse on a negative example.  This will be fixed when I put in verb-noun agreement rules.  

I realized that most of the errors came from pronouns not being in the lexicon, so the biggest improvement I could make in coverage is to add them.  To do this, I'd have to add the full case system information to the feature structures of the pronouns, so that the object pronouns are not allowed in subject positions, or as direct objects when an indirect object is present (without an indirect object, pronouns acting as direct objects take the object form).  An alternative would be to restructure a lot of my examples to not include pronouns.  I think this would be a smarter way to improve coverage, and would help reveal more about what grammar rules need to be written.  It might change the ratio of grammatical to ungrammatical sentences, since regular nouns can occur as subjects or direct objects (indirect objects are marked by a particle).

One of the main rules I need to add is a lexical/morphological rule ensure that the form of the PAC is in agreement with the subject.  This would also improve coverage, as I can't just add all the forms of the PAC to all the verb 'stems' and add them lexicon.  Adding agreement rules for adjectives would also help, since I have a lot of examples illustrating which positions the adjective can appear in (before and after the noun), and the feminine and masculine forms of the adjective.  Besides changing my examples to include fewer pronouns, I need to change some examples to spread out phenomena more.  One sentence has the noun 'the ring', 'zooben', with a determiner affix -n.  This is unnecessary, and determiners are not what I'm highlighting in the example.  Moreover, that's the only place the noun appears in the test it's not contrasted anywhere with the base form 'zoobee.'  Cutting back on the nouns I use in the examples might make it easier to see how the phenomena is being handled by the grammar rules, without concentrating too much on the lexicon.