Back to Slave grammar main page
(Emily Bender and Jeff Good)
The grammar has been ported to the current version of the Matrix. We've hooked it up to Anya Dormer's XFST analyzer for Slave verbs (though the exact code underlying this linkage will probably change as the LKB's new morphology component come online). The XFST analyzer handles the morphophonology of the (quite complex!) Slave verbs, making it possible for our LKB grammar to handle the morphosyntax. First pass lexical rules are now in for most of the 16 prefix positions. More work remains to be done on the semantic representations for incorporated elements (nouns, adverbs, postpositions) and on thematic prefixes. We also need to go back through the lexicon and remove full-form verbs in favor of classifier+root entries.
(Jeff Good)
Basic support for negation has been added to the Slave grammar. The syntax of negation is relatively straightforward, a negative word of form "yi@le", with some possible phonological variation, follows the main verb of the sentence. Adding support for negation entailed adding certain rules for adverbial modifiers generally, and it is hoped this will lead to a treatment of certain other postverbal markers.
(Jeff Good)
After meeting at E-MELD, we discussed an initial plan for dealing with Athabaskan morphology. In its most basic form, the traditional analysis makes use of a number of position classes all based on a single stem. However, this abstracts away from a number of different kinds of "boundaries", generally defined phonologically, within the stem itself.
A simplified version of the position classes, with boundaries represented, taken from Rice (1993) (an article, not the grammar) can be seen below:
| preverb # distributive # iterative # incorporate # direct object | |
| % deictic subject % theme-derivation-aspect | |
| + conjugation/mode + subject | |
| = voice + stem |
The different types of boundaries are associated with different phonological effects with the boundaries closest to the stem showing greater cohesion and those further showing less cohesion. Though the schema above is a simplification in some respects, it reveals the important fact there is justification for dividing the verb into a series of "zones", each of which can be understand as a separate domain for the application of rules.
From the perspective of developing a formal grammar of Slave, our plan is to treat the verb not as a single "word" for parsing and generation but as several words, each built up independently, and then "joined" into one large orthographic word (which would coincide with a prosodic word). While the details of our analysis may be unique to this grammar, the idea of breaking the verb down into these different units has precedent in the syntactic literature.
The advantage of building the system wherein the verb is treated as a series of orthographically/prosodically joined words is that it reduces the complexity of the lexical rules we need to write. In addition, it will allow us to reuse parts of the matrix designed for more analytic languages to the analysis of Slave. An expectation I have (perhaps not shared by others in the project) is that the Slave verb may end up having a comparable analysis to English verb-particle combination---a unit with some clear lexical cohesion but not complete syntactic cohesion. The main difference between English and Slave, in this view, is that the different "pieces" of this complex item in Slave have much more inflection than English.
To make the discussion more concrete, in our present thinking, we would break the verb given below into three word-like units (example taken from page 504 of grammar). This example uses our present ASCII transliteration system described in this changeblog entry.
| ya@i~@h=dla@ | |
| ya@#i+ne+h=dla@ | |
| DIST-SER-2s=hCL-tear |
The three units this verb, with respect the operation of the grammar would be "ya@", "i+ne+h", and "dla@", corresponding to the three types of boundaries.
(Jeff Good)
We (mostly Emily) developed an ASCII transliteration system for the Slave data to facilitate keyboard input. In general, we will shift this system to Rice's transliteration for resources intended for public consumption. (However, unfortunately, even though we can encode characters with multiple diacritics, most systems still seem to have trouble rendering them--in the Slave case, this makes it hard to render vowels like ę́ which should be an "e" with an acute above it and a right-hook below it.
The transliteration is given below:
| ASCII transliteration conventions | |||
| Transliteration | Rice's orthorgraphy | Phonetic description | Example |
| V@ | vowel+acute | high-tone vowel | é |
| V~ | vowel+right-hook | nasal vowel | ę |
| V~@ | vowel+acute+right-hook | high-tone nasal vowel | ę́ |
| ? | glottal stop | glottal stop | ʔ |
| l/ | barred l | voiceless l | ł |
| C! | consonant apostrophe | glottalized consonant | t' |
| V! | vowel apostrophe | glottalized vowel | e' |
We've started looking into the verbal morphology, and found that we're pretty much going to have to work out a holistic solution and then implement. We can't start with the "easy" stuff (morphosyntactically), e.g., verb-subject agreement, since it's buried "inside" other parts of the morphology, like aspect, deictics, etc. Furthermore, the morphophonology is complex enough that we shouldn't really start in without a thorough understanding of the process.
It seems that this aspect of the language will push the boundaries of the Grammar Matrix, but more importantly, of the morphological capabilities of the LKB. This is why we've chosen to work on it, as it should inform the design of improved morphological processing, in the LKB, or in a separate component interfaced with the LKB.
After some discussion, it seems that the best strategy for implementing an analysis within existing technology is to build up three-four separate sub-words around the open-class elements (verb stem, incorporated noun stem, adverbials/postpositions) and then join those sub-words together into orthographic words.
Added simple yes-no questions, per 36.1 of Rice 1989. Sentence-initial question marker ?asi~@/si~@ is treated as a kind of a complementizer, that is, a head that take a sentence as a complement. This entails adding a second, head-initial, head-complement rule, and grouping the head values into two types to ensure the right ordering of head and complement based on the type of the head.
Added vocabularly to test overt possessors, and confirmed that the analysis works as intended (after a couple of bug fixes to intrans-verb-lex and an overdue change to the Matrix). Here's the semantic representation assigned to the (presently sole) parse of "?eyi ?ehkee li~e@ ya@hi@tsee" `The boy's dog barks'.
< h1, e2,
{ h3:distal+dem_q_rel(x4,h6,h5),
h7:_boy_n_rel(x4),
h8:_dog_n_rel(x9),
h10:poss_rel(u11,x4,x9),
h12:def_explicit_q_rel(x9,h14,h13),
h15:_bark_v_rel(x9),
h1:proposition_m_rel(h16) },
{ h5 qeq h7,
h13 qeq h8,
h16 qeq h15 } >
Added spelling-changing lexical rules for possessive prefixes and possessed-form suffix. Sticking with alienable possession for the moment (except for 'child', which is handled by having the possessed-form suffix rule take -aa to -aa; with current technology, we can't list an irregular form and then add further prefixes or suffixes to it).
The possessed-form suffix rule adds a `subject' requirement to a noun, and adds a poss_rel relating the `subject' of the noun to the noun itself. Full NPs as possessors will be thus realized through a non-clausal variant of the head-subject rule. Prononimal NPs are realized as prefixes via a family of lexical rules which all inherit from the type possessive-prefix-lex-rule. These rules cancel the subject requirement and insert the semantic relations for a possessive pronoun and its associated quantifier. In all cases, the quantifier for the whole NP is introduced by a covert determiner phrase. (Since possessed NPs appear to be incompatible with overt determiners, the possessed-form suffix rule makes the specifier requirement of the resulting form incompatible with the range of possible overt determiners.)
To handle test sentences for these pronouns, added types for intransitive verbs. Some transitive verbs with proniminal affixes are currently being treated as intransitives. This means the semantics are wrong. To fix when we get the verb morphology running.
Downloaded LinGO Grammar Matrix. Developed ascii transliteration of Rice's orthography in the absence of good fonts/entry methods for the characters needed. Added lexical types for transitive and ditranstive verbs and nouns, plus collateral types (subtypes of head, semantic relations for types of determiners). Added types and instances for phrase structure rules: head-subject, head-complement, and covert-determiner. Added lexical entries for a few verbs and nouns. Added lexical types for marker postpositions (this one didn't have a good supertype in the matrix -- consider adding one). Can now parse and generate simple transitive and ditransitive sentences with overt NPs.
Word order, per Rice, is: Subject-Oblique-Object-Verb
Back to Slave grammar main page