Lab 3
Curtis/McHugh
Lakota [lkt]
2013.04.19

# 1. New and revised choices

## Revisions of prior choices

We revised our grammar to use the "no case" choice. This is because
Lakota NPs are not marked for case. Subject and object roles are
determined by NP order, which is generally SOV; speakers will often
invert order for focus, in which case the ambiguity must be resolved by
context.

For example:

  šúŋka kiŋ igmú kiŋ khuwá
  dog DEF cat DEF chase
  'The dog chases/chased the cat.'

could also be read as 'the DOG is chased by the cat.'

There are agentive and patientive forms of pronomial verb affixes, but
this is not within the domain of the case library of the customization
system.

We have also corrected the analysis and implementation of the pronomial
affixes. Lakota pronomial affixes are attached to the verb via pre-
and/or infixing, depending on the verb, person, and combination of
affixes. They are also obligatory when an NP subject and/or object is
not present, and forbidden when the relevant NP is present.

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: pn
  igmú kiŋ wakhúwa
  igmú kiŋ wa-khuwá
  igmú kiŋ khuwá-1SgAgt
  cat DEF chase-1SG.AGT
  'I chase the cat'

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: pn
  igmú kiŋ makhúwa
  igmú kiŋ ma-khuwá
  igmú kiŋ khuwá-1SgPat
  cat DEF  chase-1SG.PAT
  'The dog chases me'

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: pn
  makhúwa
  ma-Ø-khúwa
  khuwá-1SgPat-3SgAgt
  chase-1SG.PAT-3SG.AGT
  'It chases me'

Where we had previously implemented them as pronouns in the lexicon,
they are now implemented as affixes using the morphology library. We
created three verb position classes, one for each valid combination:

- itrs-pat-subj: intransitive-stative verb + patientive pronomial subject

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: c
  káŋ
  Ø-káŋ
  káŋ-3SgPat
  be_old-3SG.PAT
  'he/she/it is old'

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: c
  nikáŋ
  ni-káŋ
  káŋ-2SgPat
  be_old-2SG.PAT
  'you (sg) are old'

- itra-agt-subj: intransitive-active verb + agentive pronomial subject

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: c
  lowáŋ
  Ø-lowáŋ
  lowáŋ-3SgAgt
  sing-3SG.AGT
  'he/she/it sings'

- tra-pat-obj: transitive-active verb + patientive pronomial object + agentive pronomial subject

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: pn
  uŋkhúwapi
  uŋ-khuwá-pi
  khuwá-1PlPat-3PlAgt
  chase-1PL.PAT-3PL.AGT
  'They chase us'

Our initial analysis noted that when both pronomial affixes are present,
they are generally in object-subject (patient-agent) order, with certain
exceptions. Since we have assumed a finite-state morphological analyzer
as input to the grammar which already maps the correct pre- and
infixation to canonically-ordered suffixes, our grammar needs no
additional position classes beyond these three to properly model the
pronomial affixes. The morphological analyzer also maps the surface
forms of the affixes, which are subject to a variety of
morphophonological changes, into artificial syntactic morphemes (e.g.,
the 2SG.PAT affix 'ni-' is mapped to -2SgPat) and the surface verb form
into the lexical stem.

Although this choice makes the upper-side FST language less readable to
Lakota speakers, we feel this is a reasonable tradeoff, given that the
conversion of pre- and infixes to suffixes necessary in any case already
imposes a readability burden.

To simplify construction of the lexical rules for these affixes, we
created a hierarchy of types to treat each subject/object feature
(PER/NUM) separately; each lexical rule '-morph' type inherits the
appropriate combination of feature-specifying supertypes. For example,
'itrs-subj-3sg-pat-morph' inherits the following constraints:

- SYNSEM.LOCAL.CAT.VAL.SUBJ.FIRST.OPT + (itrs-subj-pat-lex-rule)

- SYNSEM.LOCAL.CAT.VAL.SUBJ.FIRST.LOCAL.CONT.HOOK.INDEX.PNG.PER 3rd (itrs-subj-3per-lex-rule)

- SYNSEM.LOCAL.CAT.VAL.SUBJ.FIRST.LOCAL.CONT.HOOK.INDEX.PNG.NUM sg
  (itrs-subj-sg-lex-rule)

As part of the reimplementation of pronomials into affixes, we also
implemented some latent restrictions on their combination. For example:

  Source: author
  Vetted: f
  Judgment: u
  Phenomena: pn
  mawákhuwa
  ma-wá-khuwá
  khuwá-1SgPat-1SgAgt
  chase-1SG.PAT-1SG.AGT
  'I chase me'

  Source: author
  Vetted: f
  Judgment: u
  Phenomena: pn
  uŋwákhuwapi
  uŋ-wá-khuwá-pi
  khuwá-1PlPat-1SgAgt
  chase-1PL.PAT-1SG.AGT
  'I chase us'

Although Lakota does have other means of expressing the intended
semantics of the first example (i.e., reflexive verb forms) neither of
these constructions are valid syntax. A similar restriction exists for
combinations of the dual agentive pronomial with 1pers patientives.

To implement these restrictions we added forbid constraints on the 1pers
and 2pers patientive lexical rule types to prohibit incompatible
agentive lexical rule types. This has the side effect, due to the
implementation of the suffixes, of making these verb forms invalid
tokens (and errors) rather than instead of merely resulting in no valid
parses.

We also removed noncountable from the gender system and replaced it with
a COUNT feature (uncountable nouns are marked COUNT -).


## New Choices

### Additional Verb Valences

We have added two new verb types to our analysis: plain verbs and
impersonal verbs. Impersonal verbs generally refer to topics such as the
weather or temporal/spatial situations. They take no arguments or
(pronomial affixes) and so have a zero valence.

  Source: b:760
  Vetted: s
  Judgment: g
  Phenomena: wo
  maǧážu
  maǧážu
  maǧážu
  be_raining
  'It is raining'

The customization system does not allow verbs to be specified with a
zero valence, so the impersonal verbs are not implemented in our grammar yet.

We have also added 'plain verbs'. Plain verbs are similar to the
intransitive-stative verbs we previously analyzed, but can only
take inanimate subjects as overt NPs (no pronomial affixes are allowed):

  Source: b:760
  Vetted: s
  Judgment: g
  Phenomena: agr
  mní kiŋ šmé
  mní kiŋ šmÁ
  mní kiŋ šmÁ
  water DEF be_deep
  'the water is deep'

In order to implement plain verbs, we created a new class of verbs,
plain, which are intransitive and specify [ GENDER inanimate
] on the subject NP. Nouns inherit the GENDER feature from their noun type.

### Argument Optionality

To properly account for the complementary distribution of pronomial
affixes and NPs, we applied the argument optionality library. We made
both subject- and object-dropping always allowed, requiring a marker
when dropped, and not permitted with overt NPs.

### Tense

Lakota does not express tense. Past and present are not distinguished,
and the role played by future tense is subsumed under the more general
irrealis mood, discussed below.

### Aspect

Aspect in Lakota is expressed via verbal enclitics. They are generally
analyzed and written as separate words, although some speakers treat
them as suffixes. Ingham (2003) identifies two aspectual
suffixes/enclitics: hAŋ (continuative) and what he refers to as 'recent
past', with the complex forms 'laka čhaŋ' or 'pilaka tkȟa'.

We were unable to cross-reference Ingham's 'recent past' aspect with any
of our other resources. On further investigation into the original
source corpus, it appears his analysis is faulty. One of the examples he
cites from Deloria's (1939) Dakota Texts is:

  íčhimani k'uŋ él ílaka čhaŋ ayáštaŋ iyeye
  'as soon as the traveller reached him he stopped shouting'

Ingham's analysis would gloss 'í-{laka čhaŋ}' as arrive_there-PST.RECENT.
Cross-referencing with other materials, however, list čhaŋ as a
conjunction, meaning 'when....then', which would indicate a better gloss
for ílaka is arrive_there-INTENS 'exactly when he arrived there'. We
thus only analyzed continuative as an aspectual marker. Continuative is
also more commonly considered as a suffix than a separate word, and is
usually written connected to the verb with a hyphen. For example:

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: tam
  igmú kiŋ makhuwá-hiŋ
  igmú kiŋ ma-khuwá-hAŋ
  igmú kiŋ khuwá-1SgPat-Cont
  cat DEF chase-1SG.PAT-CONT
  'The cat keeps chasing me'

We have chosen to follow this analysis in our implementation. The
continuative aspect is implemented as a verbal suffix, in the
non-obligatory position class cont-pos, which follows any pronomial
affix position classes. The affix adds [ ASPECT continuative ] to the
verb. It is also modeled using the FST upper-side artificial morpheme
'-Cont' to retain consistent usage among all morphemes which are mapped
into their proper surface forms and position by the morphological
analyzer.

### Mood

As opposed to the sparse aspectual markers, Lakota has a rich assortment
of mood enclitics. We selected three of these to analyze and implement:
'ktA' irrealis, 'séčA' possible, and 'škȟÁ' quotative. The relative
ordering of the postverbal enclitics is fixed by implied position
classes, and some of the enclitics are mutually exclusive (séčA and
škȟÁ, for example).

The mood enclitics we analyzed are only permitted in this order (with
škȟÁ permitted instead of séčA):

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: neg,tam
  igmú kiŋ sápa kta šni séče
  igmú kiŋ sápA ktA šni séčA
  igmú kiŋ sápA ktA šni séčA
  cat DEF be_black IRR NEG POSSBL
  'The cat probably won't be black'

We decided that analyzing the enclitics as AUX verbs was the most
natural approach, since we also chose to model sentential negation (also
expressed via a postverbal enclitic) as a negative auxiliary (see
below).

To implement the relative ordering of the enclitics, we created
additional FORM subtypes which are specified on the auxiliary, and each
auxiliary specifies the enclitics it can appear to the right of as
complement features. However, due to a bug in the customization system
AUX verbs other than neg-aux are generated as specifying

  [ ARG-ST < [ ],
             [ LOCAL.CAT.HEAD.AUX - ] > ].

thus preventing auxiliaries from taking auxiliaries as complements. We
plan to revisit this implementation when we begin manually editing the
grammar files.


### Agreement

As overt subject and object NPs are unmarked, there is very little role
for agreement in the phenomena we have analyzed thus far. Agreement is
exhibited in the 'plain verbs', which select for inanimate NP subjects
only:

  Source: b:760
  Vetted: s
  Judgment: g
  Phenomena: agr
  mní kiŋ šmé
  mní kiŋ šmÁ
  mní kiŋ šmÁ
  water DEF be_deep
  'the water is deep'

We implemented this as described above in the discussion of plain verbs.

### Negation

Sentential negation in Lakota is indicated via the postverbal enclitic
šni:

  Source: a:157
  Vetted: f
  Judgment: g
  Phenomena: neg
  tuwé kiŋ waŋbláke šni
  tuwé kiŋ waŋ-bl-yáŋkA šni
  tuwé kiŋ waŋyáŋkA-1SgAgt šni
  be_who DEF see-1SgAgt NEG
  'I did not see who it was'

In this example tuwé 'be_who' is a nominalized verb acting as the object
of waŋyaŋkA 'to see smth'. We implemented sentential negation using the
auxiliary verb option in the customization system. This option provides
the neg-aux type which contributes a neg_rel semantic relation.  The
negative šni participates in the ordering restriction of enclitics
discussed above.

Lakota also exhibits a form of negative concord, where the negation is
meant to apply to a nominal rather than the entire sentence:

  Source: a:157
  Vetted: f
  Judgment: g
  Phenomena: neg
  tuwéni waŋbláke šni
  tuwéni waŋ-bl-yáŋkA šni
  tuwéni waŋyáŋkA-1SgAgt šni
  be_who.NEG see-1SgtAgt NEG
  'I did not see anyone' (I saw no-one being.)'

However, this is a complex form of concord that is not supported by the
customization system, and so we have not implemented it.

We also noted that this analysis also affects the treatment of tuwéni
and related forms as negative indefinite articles; however, in the
absence of any other phenomenon prompting a deeper analysis, we elected
to retain them in their current forms.

### Demonstratives/Definiteness

Lakota has several demonstratives, which differentiate three numbers
(singular, dual, and plural) and three relative positions (near the
speaker, near the listener, and distant). These demonstratives frequently
co-occur with determiners, and can precede or follow the NP:

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: adj
  ká igmú kiŋ wakhúwa
  ká igmú kiŋ wa-khúwa
  ká igmú kiŋ khuwá-1SgAgt
  DEM cat DEF chase-1SG.AGT
  'I chase that (yonder) cat'

  Source: author
  Vetted: f
  Judgment: g
  Phenomena: adj
  hé igmú kiŋ wakhúwa
  hé igmú kiŋ wa-khúwa
  hé igmú kiŋ khuwá-1SgAgt
  DEM cat DEF chase-1SG.AGT
  'I chase that cat'

Although demonstratives can occur by themselves, their co-occurrence
with determiners make them better analyzed as adjectives, which are not
specifiable in the customization system. We therefore have not
implemented them so far.

# 2. Properties illustrated in testsuite but not covered by grammar or
  customization system

The largest set of testsuite items not covered by the grammr or
customization system is the demonstratives. They will not be implemented
until we get to adjectives.

Our grammar also doesn't fully cover the ordering of AUX enclitics, due
in part to the application of AUX - constraints.

Finally, our testsuite includes examples of 0-valence (impersonal)
verbs, which are not implementable in the customization system.

# 3. Coverage of grammar over the testsuite

Our overall coverage of the testsuite is now 86.4%, with 17 parse
errors. Of the parse errors, 8 represent prohibited combinations of
verb affixes and so are correct; the remaining 9 are accounted for by
unimplemented grammar elements (7 demonstratives, 2 impersonal
verbs). There is one negative example which overgenerates; this is
related to the limited support for multiple auxiliaries.

# 4. Where does the customization system seem unable to cope with the
  properties of your language?

The most complex area where the customization system was in the AUX
implementation of multiple enclitics. The rules it generates
overconstrain in such a way that they cannot be easily combined.

The customization system also does not allow 0-valence verbs, which are
needed for impersonal verbs in Lakota.

Our analysis of demonstratives as adjectives is also outside the scope
of the customization system.
