Underspecified Values Miriam Butt and Ron Kaplan Par
Underspecified Values Miriam Butt and Ron Kaplan Par. Gram Meeting September 2003, Stuttgart 1
Linguistic notions often confused • • • True underspecification Constrained underspecification Unmarked value Default value Indeterminate value Issues: Grammar-writer notion/intuition: what’s expressed? F-structure configuration: what appears? Grammatical specification: what’s written? F-structure client: who/what? Interpretation? 2
Some distinctions/confusions • True underspecification – Notion: any value is possible (perhaps known from declaration) – – No attribute/value in minimal model (vs. attribute with packed value) Never mentioned in grammar F-structure client (e. g. transfer/generate) can freely instantiate E. g. deer has no value for NUM 3
• Constrained underspecification – Notion: One of several values is possible (restricted subset of declared values) – Attribute and particular value in minimal model – Specified disjunctively { (^ CASE)=ACC | (^ CASE)=NOM } or (^ CASE) $ {ACC NOM} vs (^ CASE) = {ACC NOM} (indeterminate) – Nothing special for f-structure client (packed structure will have a locally contexted value independent of all other contexts) – Note: Wide-scope disjunction, not indeterminate. 4
• Unmarked value – – – (cf. phonology: u Nasal + Nasal) Notion: interpreted as a particular value distinct from others, known by grammar writer, not specified only to avoid f-s clutter No attribute/value in minimal model Checked, if necessary, by negative existential constraint F-structure client must be informed, or have value filled in during hand-off E. g. “Active is unmarked value for Voice” ~(^ Voice) Feature Committee (1997): Avoid marking conventions (clutter f-structure? ) 5
• Default value – Notion: If value not otherwise specified (in a given scope), fill in particular one Equivalent: put explicit fill-in equation, but may be difficult to figure out exactly where – Attribute and particular value in minimal model – Specification: @(DEFAULT (^ CASE) NOM) DEFAULT(D V) = {D | D = V } { (^ CASE) | (^ CASE)=NOM } {D D~=V | D = V } – Nothing special for f-structure client – Global vs. Local context: @(COMPLETE D) Default with respect to whole analysis vs. default with respect to subtree/lexicon On lexical category: @(DEFAULT (^ NUM) PL) @(COMPLETE (^ NUM)) – No new mechanisms, logic (no default unification, defeasible inference…) • But: gobal defaults can be expensive (like existential constraints in general); • Disjunct can’t be chosen until all within-context defining-constraints have been evaluated • Has no bearing on satisfiability, only extends minimal model – Possible extension: (^ CASE) =d NOM • Could be implemented as narrow-scope disjunction w. r. t. coordinate distribution • May not be linguistically appropriate 6
• Indeterminate value (Dalrymple & Kaplan, 2000) – Attribute and set value in minimal model – Notion: Different elements can satisfy different requirements • Transitivity of equality doesn’t apply • Narrow scope disjunction with respect to coordinate distribution – Specification: (^ CASE) = { NOM ACC } Requirements imposed by membership statements: ACC $ (^ CASE) – Nothing special for f-structure client 7
8
- Slides: 8