Kelly O'Hara Ling 567 Lab 8 Writeup 24 February 2007 Basic sentential negation in Zulu is marked by a prefix (a-) on the verb. In pretty much every tense except future, the final vowel also changes, but this was one of my motivations for choosing future tense for my test suite. Therefore, the first thing I activly tried to parse was a-ku-zo-ngi-limaz-a NEG-SC15-FUT(IMM)-1SG-hurt-FV 'It will not hurt me.' If you parse this, you'll notice that it gets 3 parses, but that's just because ku- is the subject concord for classes 15, 16, and 17, and the parser has no way of distinguishing them (the pronoun is the same, too). For the the purposes of this lab, I did not worry about final vowel changes, so the present tense version of the above ('It does not hurt me') can be formed by removing the tense marker (zo-). The customization script gave me an inflecting-lex-rule & c-cont-change-only-lex-rule for the negation rule. I noticed most of the constraints are the same as the ones suggested for the 'can' morpheme rule in the lab instructions. The C-CONT.HCONS is a diff-list with one qeq; the HARG of the qeq is identified with the ARG1 of the relation on the C-CONT.RELS list, and its LARG with the daughter's LTOP. RELS is a diff-list with the LBL identified with the C-CONT.HOOK.LTOP and the ARG0 with the C-CONT.HOOK.INDEX. The differences are that the thing on the RELS list is an event-relation, and its PRED value is "_neg_r_rel". I made some modifications to the rule in previous labs, such as making the DTR value a subj-concord-lex-rule, to help position this rule in the right place in my morphological hierarchy. Unfortunately for me, Zulu has separate negation strategies for imperatives and 'can'. The imperative negation is done by an inherently-negative auxiliary verb, 'musa'. Its complement is in the infinitive: musa uku-gijim-a don't C15-run-FV 'Don't run!' See that C15 there? The infinitive is actually a noun class. However, there is some ambiguity: uku-dl-a `C15-eat-FV' can be translated as 'to eat' or 'food'. I decided the easiest way to implement this was to make them homonyms, and have a separate inf-verb-lex-rule that added the same prefix. This was trickier than I was expecting. I also spent a lot of time figuring out how to implement the auxiliary verb. The auxiliary takes the plural ending from the imperative: musa-ni uku-gijim-a don't-PL INF-run-FV 'Don't run!' The auxiliary does not have overt tense marking (as it is almost certainly present tense, which has a zero marker), and cannot take a subject concord (because it is an imperative, which do not take subject concords). I have no data for whether the auxiliary can take an object concord. I decided to play it safe and say no. So this parses: musa uku-shay-a in-ja don't INF-hit-FV C9-dog 'Don't hit the dog' and this doesn't: yi-musa uku-shay-a in-ja OC9-don't INF-hit-FV C9-dog 'Don't hit the dog' The other separate form of negation has to do with 'can'. Zulu has two prefixes dealing with ability: nga- which means "can", and nge- which means "can't". When nge- is used, the final vowel changes to -i (the same as with the non-future tenses, as discussed above). I have not currently implemented any final-vowel changes, as doing so means I would have to get them working for every kind of sentence. This looks like it will take a bit of pulling things apart and putting them back together, and I didn't want to break everything. That sounds like a project more suited to the next lab. On the other hand, given that the final vowel is not itself contributing any semantic information (that is, there is always a separate negation marker) perhaps simply ignoring it is an acceptable solution. AAt any rate, I implemented the "can't" morpheme, but it only works with the standard -a final vowel. So this parses: ngi-nge-dl-a in-gilazi 1SG-can't-eat-FV C9-glass 'I can't eat glass' and this doesn't: ngi-nge-dl-i in-gilazi 1SG-can't-eat-FV C9-glass 'I can't eat glass' As far as semantics go, the semantics of "It does not hurt me" look like the example given in the lab instructions, after taking into account the lack of pronouns. The PRED value on 'musa' is "_neg_v_rel" and sentence is COMM rather than PROP-OR-QUES, but the semantics otherwise look the same. For nge-, I made the PRED value "_can't_v_rel" because I wasn't sure how else to capture both the "can" semantics and the negation semantics. Please let me know if the MRS looks wrong. I had some problems with overgeneration with the auxiliary verb, but I think they are under control now. If parse (as given above) "musa uku-gijim-a", and then generate, you should get only that string back. The rest give the combinations of questions and concords that I have come to expect. As for positive "can", I mentioned above that it is marked as an affix on the verb: ngi-nga-dl-a in-gilazi 1SG-can-eat-FV C9-glass 'I can eat glass' This was relatively straightforward to implement using the directions given in the lab. As I mentioned on epost, I had to add a constraint on the type of the DTR value, to get the morphemes to appear in the right order. I made a new type, can-dtr that has the possible inputs to the ability-lex-rule as its subtypes. I also had ability-lex-rule inherit from verb-tense-rule-dtr. This was again just to get morpheme ordering correct. The other thing that I changed from the lab instructions was the PRED value, which I left underspecified in the basic ability-lex-rule. I then created two subtypes: pos-ability-lex-rule and neg-ability-lex-rule, to account for the different prefixes discussed above. I had you look at my MRS on epost, so I'm sure the semantics are right. Not all the sentences from my test suite are parsing, because some of them I took from one of my grammar books, and they have locative expressions in them. Since I haven't dealt with locatives at all, naturally they fail. However, I have done a lot of testing on my own, So I think everything parses that should. Both of these should parse: umu-ntwana u-nga-gijim-a C1-child SC1-can-run-FV 'The child can run' umu-ntwana u-zo-nga-yi-bon-a in-ja C1-child SC1-FUT(IMM)can-OC9-see-FV C9-dog 'The child will be able to see the dog' I experimented with moving the affix around; as long as it appears between the slots where the tense marker and the object concord go (they might both be zeros) it parses, and everything else fails. Hooray! For next week, I am not sure where to focus my attention. The gaping hole in my test suite is coordination. I could get NP coordination working without too much difficulty, but getting VP coordination working would involve getting final vowels working. However, getting final vowels working would be nice in itself, so I don't have the -a hard-coded onto all of my verbs anymore. Another thing that I have in my test suite is the aformentioned locative constructions. I'm not sure if those would be hard or not; I would need some help with the analysis I think. But most of what I have worked on so far is working more or less how I want it to. I'm pretty pleased with how things are turning out.