Ryan Georgi Ling 567 Lab 9 Writeup: * NOTE: I'm afraid I've been running into a bunch of translation problems trying to get this to work, and it's fast approaching midnight. I've tried to note the interesting things I was able to get to transfer, as well as hopefully some info that should help debugging. ----====== GRAMMAR CLEAN-UP =====----- --Replaced Matrix, removed comps <> constraint from HEAD-DTR on basic-head-subj phrase -- Added FIXING BARE AFFIXES: -- I noticed for some reason I'd forgot to put %suffix instead of %prefix for some of my circumfix affixes. (Copy and paste problem). -- Changed the DTR type on all second-applying circumfix rules to not just be the parent circumfix_rule type, but rather the type of only the previously applying rule. -- Refined the input/output constraints on the pronoun incorporation rules. I had previously only constrained them to apply to a type uninfl_verb (slightly different from INFLECTED +, as I wanted to distinguish between object incorporation and verb/noun agreement). To reduce spurious chart edges, I instead made two subtypes of uninfl_verb (so as to not break the ordering of other rules), can_incorp and dont_incorp. I constrained the object incorporation rule to only take can_incorp, and added dont_incorp as a supertype to some verb-lex items such as modal_verb_lex and intransitive-verb-lex. This didn't completely fix my problem with reaching too many chart edges, but it did greatly reduce it. ---======= DEFLECTED AGREEMENT ======------- To recap, deflected agreement behaves the following way in Arabic: Inflection is changed only on the verbs. In SVO word order, nouns and all verbs agree in both person/gender and number. In VSO word order, agreement is for person/gender, but number agreement *on the verb* is always singular, regardless of the number on the noun. This appears to be the case for all verbs. Examples: (SVO - Full agreement) ?al-?awlad-u y-i?ra-u: ?al-dars-a the-boys-NOM 3MPL-read-3MPL the-lesson-ACC 'the boys read the lesson' VSO (Partial Agreement) y-i?ra ?al-?awlad-u ?al-dars-a 3MSG-read the-boys-NOM the-lesson-ACC 'the boys read the lesson' * SVO (Partial agreement) ?al-?awlad-u y-i?ra: ?al-dars-a the-boys-NOM 3MSG-read the-lesson-ACC '* the boys read the lesson' * VSO (Full Agreement) y-i?ra-u: ?al-?awlad-u ?al-dars-a 3MPL-read-3MPL the-boys-NOM the-lesson-ACC '* the boys read the lesson' -------- HOW IT'S HANDLED: I changed the head-subj-phrase to inherit from head-final rather than head-initial. This made SVO word order the default, and kept the constraints for verb-noun agreement I had made on my subj-agreement-lex-rule types. For VSO word order to be handled, I made a new set of lexical rules that inherit from defl-subj-agreement-lex-rule, a infl-val-change-only rule that moves constrains the verb's SUBJ to be null and identifies the elements of the ARG-ST with the COMPS. The subtypes of this rule all inflect with the singular form, but only constrain the GEND and PER of the first COMPS element. (since this is a ltol-rule, I also added a const-ltow-rule that takes only this type of rule as input) ================== SOME INTERESTING EXAMPLES FOR MT ============= Embedded pronouns:----------------- t-atay-ni: ?al-mara-u ?al-taam-a "The woman gave me food" Embedded Clause with pro-drop:----------------- y-aftikir-na ?anna rajul-u-n t-ishtara ?al-taam-a "They (fem) think that the a man buys the food" Deflected agreement:--------------- ?al-?awlad-u y-i?ra-u: ?al-dars-a (SVO + FULL) y-i?ra ?al-?awlad-u ?al-dars-a (VSO + PARTIAL) * y-i?ra-u: ?al-?awlad-u ?al-dars-a (VSO + FULL) * ?al-?awlad-u y-i?ra: ?al-dars-a (SVO + PARTIAL) "The children read the lesson" Imperative + Demonstrative:----------- i?ra ?al-kita:b-a da: "read this book!" ================== TRANSLATION RESULTS ========================= I only got a few examples working, but I'll try to show what I got! "The woman hit the food" (whatever that means) t-adribu ?al-mara-u ?al-taam-a yields: hitta-ar kona mat hitta-ar kona mat-inn hitta-ar kona-in mat hitta-ar kona-in mat-inn kona hitta-ar mat kona hitta-ar mat-innmadur hyggja-ar ad kona kaupa-ar mat kona-in hitta-ar mat kona-in hitta-ar mat-inn But, as I've noted in ePost, though 'food' and 'cat' have exactly the same distribution in icelandic, I wasn't able to get "The woman hit the cat" to parse (I get an invalid pred error; though I even cut and pasted the pred from Katie's lexicon into my entry for cat). Also, I haven't been able to find a reason why this isn't working, but I get 0 MRSes for trying to get "The woman hit the small food" (t-adribu ?al-mara-u ?al-taam-a ?al-sagheer-a) I've looked at both MRSes, they are at the very bottom. ------------ "The woman gave food" First, I tried my incorporated pronoun: t-atay-ni: ?al-mara-u ?al-taam-a (Where "_give_v_rel" has the following: ARG0: e2 ARG1: x5 [...PNG.PER: THIRD ... PNG.NUM: sg] (x5 is mapped to woman through a qeq) ARG2: x3 [...PNG.PER: FIRST ... PNG.NUM: sg] (x3 does not appear elsewhere... I think that's right) ARG3: x4 [...PNG.PER: THIRD ... PNG.NUM: sg] (x4 is mapped to food through a qeq) ...this got translated to Katie's system as kona gefa-ar mat-i kona gefa-ar mat-i-num kona-in gefa-ar mat-i kona-in gefa-ar mat-i-num ...which she unfortunately tells me is simply "The woman gives food" I tried an explicit direct object instead: ("The woman gives the man food") t-atay ?al-mara-u ?al-rajul-a ?al-taam-a ..which yielded: kona gefa-ar mann mat-i kona gefa-ar mann mat-i-num kona-in gefa-ar mann mat-i kona-in gefa-ar mann mat-i-num ..which katie said were correct translations, yay! ------- IMPERATIVES adribu rajul-a-n ("hit a man!") --> hitta mann hitta-id mann hitta-um mann ..which katie again says are correct. ----------- DEMONSTRATIVES I tried doing: adribu ?al-rajul-a da: ("hit that man!") But got the error on katie's side: "invalid predicates: |predsort|" ..I have the pronoun transfer rules as posted in the appropraite places (triple checked), so I'm not sure I know what to do about this. ---------- EMBEDDED CLAUSES: y-aftikir-na ?anna ?al-mara-u t-ishtara ?al-taam-a "They (fem) think that the woman buys the food" and also y-aftikir ?al-rajul-u... "The man thinks that..." ...I got zero results, and from what I can tell, it looks like somehow the qeq for the ARG2 to "think" is winding up qeq'ed to a label not in the MRS; I'm not sure what this means. ======= WRAP-UP ======================================== For this lab, not only did I get the deflected agreement working, I also managed to get rid of the bare endings so that even the "y-aftikir ?al-rajul-u ?anna ?al-mara-u t-ishtara ?al-taam-a" above generates in very little time. As I mentioned in class, I'd mistakenly written much of my test suite without thinking about deflected agreement. I went back and tried to find places where I'd put false positive or negatives, but I may have missed a few. Thankfully it seems I stuck to mostly singular subjects, so although I imagine this isn't the best coverage I can get, it's still better than it has been, and with very little overgeneration! My coverage is now at 49.2%, up from 42% in lab8. My overgeneration analysis from tsdb is looking very good at 6.6% down from around 7.9% , though, again, this may be falsely high. APPENDIX: ================================================= ;;; ;;; Transfer Output --- ARABIC ;;; [ LTOP: h1 INDEX: e2 [ e SORT: SEMSORT E.TENSE: TENSE E.ASPECT: ASPECT E.MOOD: MOOD SF: IFORCE ] RELS: < [ "_hit_v_rel" LBL: h1 ARG0: e2 ARG1: x4 [ x SORT: SEMSORT COG-ST.SPECI: BOOL PNG.PER: THIRD PNG.GEND: FEM PNG.NUM: SG ] ARG2: x3 [ x SORT: SEMSORT PNG.PER: THIRD PNG.NUM: SG PNG.GEND: MASC COG-ST.SPECI: BOOL ] ] [ "_woman_n_rel" LBL: h5 ARG0: x4 ] [ "_exist_q_rel" LBL: h6 ARG0: x4 RSTR: h7 BODY: h8 ] [ "_food_n_rel" LBL: h9 ARG0: x3 ] [ "_exist_q_rel" LBL: h10 ARG0: x3 RSTR: h11 BODY: h12 ] [ "_small_a_rel" LBL: h10 ARG0: e13 [ e E.TENSE: TENSE E.ASPECT: ASPECT E.MOOD: MOOD SF: IFORCE SORT: SEMSORT ] ARG1: x3 ] > HCONS: < h7 qeq h5 h11 qeq h9 > ] ;;; ;;; Transfer Output --- ICELANDIC ;;; [ LTOP: h1 INDEX: e2 [ e SF: PROP-OR-QUES E.TENSE: TENSE E.ASPECT: ASPECT E.MOOD: MOOD SORT: SEMSORT ] RELS: < [ "_woman_n_rel" LBL: h3 ARG0: x4 [ x SORT: SEMSORT PNG.PER: THIRD PNG.NUM: SG PNG.GEND: FEM COG-ST.SPECI: BOOL ] ] [ "_exist_q_rel" LBL: h5 ARG0: x4 RSTR: h6 BODY: h7 ] [ "_hit_v_rel" LBL: h1 ARG0: e2 ARG1: x4 ARG2: x8 [ x SORT: SEMSORT PNG.PER: THIRD PNG.NUM: SG PNG.GEND: MASC COG-ST.SPECI: BOOL ] ] [ "_small_a_rel" LBL: h9 ARG0: e10 [ e E.TENSE: TENSE E.ASPECT: ASPECT E.MOOD: MOOD SF: IFORCE SORT: SEMSORT ] ARG1: x8 ] [ "_food_n_rel" LBL: h9 ARG0: x8 ] [ "_exist_q_rel" LBL: h11 ARG0: x8 RSTR: h12 BODY: h13 ] > HCONS: < h6 qeq h3 h12 qeq h9 > ]