Indeterminacy by Underspecification Mary Dalrymple (Oxford), Tracy Holloway King (PARC) and Louisa Sadler (Essex) 1 Ambiguity vs Indeterminacy The simple view is that agreement features have atomic values, described by equality: (1) Sie singt/singen. (German) she/they sings/sing She sings, they sing (2) *Sie singt und singen. (German) she/they sings and sing She and they sing (Pullum and Zwicky 1986:765) This gives no obvious account of feature indeterminacy (a form that satisfies conflicting requirements on a feature like ). Groos and van Reimsdijk (1979); Pullum and Zwicky (1986); Zaenen and Karttunen (1984) (3) Er findet Papageien. (German) he finds parrots-acc/dat (4) Er hilft Papageien. (German) he helps parrots-acc/dat (5) Er findet und hilft Papageien. (German) he finds and helps parrots-acc/dat 2 Feature indeterminacy: CASE Dalrymple and Kaplan (2000) propose a treatment of indeterminacy in morphosyntactic features which uses sets as values for such features. (6) Wer nicht gefördert wird, muss klug sein. (German) Who not supported is must clever be nom =nom =nom Who isn t supported must be clever. (7) *Wer nicht geholfen wird, muss klug sein. (German) Who not helped is must clever be nom =dat =nom Who isn t helped must be clever. (8) Ich habe gegessen was übrig war. (German) =acc? =nom 2.1 Disjunction? (9) was: ( ) = nom ( ) = acc (10) was: ( ) = nom was: ( ) = acc (11) Ich habe gegessen was übrig war =acc acc =nom (12) Ich habe gegessen was übrig war =acc nom =nom 2.2 Underspecification? (13) was: no at all (14) Ich habe gegessen was übrig war =acc x =nom (15) x = nom x = acc nom = acc (!) (transitivity of equality) 2.3 Various proposals Ingria (1990): non-distinctness checks, not unification Johnson and Bayer (1995): deduction in Lambek categorial grammar, not equality or consistency Dalrymple and Kaplan (2000): easy solution within description-based framework Daniels (2001); Levy and Pollard (2001); Crysmann (2005): lots of additional types 2.4 Set-based feature representations (16) wer: ( ) = {nom} was: ( ) = {nom,acc} (17) gegessen: acc ( ) übrig: nom ( ) 1 2
(18) Ich habe gegessen was übrig war acc {nom,acc} nom (19) was: (o ) = {nom,acc} gegessen: acc (o ) übrig: nom (s ) rel. clause construction: (o ) = (s ) (20) pred eat, tense past num sg pers 1st pred what {nom,acc} pred left o : adj pred s : (21) *Ich nehme, wem du vertraust. (German) I take who you trust acc {dat} dat I take who(ever) you trust. (22) wem: (a ) = {dat} nehme: acc (a ) [not satisfied] vertraust: dat (d ) rel. clause construction: (a ) = (d ) (23) pred take, num sg pers 1st pred who {dat} pred trust, a : num sg adj pers 2nd pred d : 2.5 Distributed values Dy la (1984) (24) *Co Janek lubi a Jerzy nienawidzi (Polish) what Janek likes and Jerzy hates {nom,acc} acc gen What does Janek like and Jerzy hate? (25) Kogo Janek lubi a Jerzy nienawidzi (Polish) who Janek likes and Jerzy hates {acc,gen} acc gen Who does Janek like and Jerzy hate? focus w : pred who {acc,gen} pred like, [ ] pred Janek focus pred hate, [ ] pred Jerzy (26) kogo (w ) = {acc,gen} lubi nienawidzi acc (w ) gen (w ) 2.6 Indeterminate (noun class) agreement requirements Voeltz (1971), Pullum and Zwicky (1986) (27) umfana nomfazi bayagoduka (Xhosa) young man and-young woman go home 1/2 1/2 class = 1/2 The young man and the young woman are going home. (28) pred go.home pred young.man class 1/2 pred young.woman class 1/2 3 4
(29) a. *Igqira nesanuse ayagoduka (Xhosa) doctor and-diviner go home 5/6 7/8 class = 5/6 b. *Igqira nesanuse ziyagoduka (Xhosa) doctor and-diviner go home 5/6 7/8 class = 7/8 The doctor and diviner are going home. (30) pred go.home pred doctor class 5/6 pred diviner class 7/8 (31) Izandla neendlebe zibomvu (Xhosa) hands and-ears are-red 7/8 9/10 class {7/8, 9/10} The hands and the ears are red. (32) pred are.red h : pred hands f : class 7/8 e : pred ears class 9/10 (33) izandla (h class) = 7/8 neendlebe (e class) = 9/10 zibomvu (f class) {7/8, 9/10} 3 The Problem with DK Summary: (34) Er findet und hilft Papageien he finds and helps parrots-acc/dat Nouns have closed sets as values, requirements placed by governing verbs are checked by set membership: (35) finden: acc ( ) hilfen: dat ( ) (36) papageien: ( ) = { acc, dat} 3.1 Transitivity Problem Modifiers and governing predicates must impose compatible agreement requirements, contrary to the predictions of the set-based analysis (Levy 2001). A noun that is indeterminately accusative or dative must take a dative modifier if the predicate requires dative, and an accusative modifier for an accusative predicate; other patterns are disallowed, even when the noun is indeterminately specified for. (37) Er hilft *die/den Papageien. (German) he helps *the-acc/the-dat parrots-acc/dat (38) Er hilft *alte/alten Papageien. (German) he helps *old-acc/old-dat parrots-acc/dat The set-based analysis permits the impossible combinations: (39) *Er hilft dat die acc Papageien. (German) ={dat acc} Note that it does not help in any way to give the nominal and the determiner each their own intrinsic features - this does not pass the constraints introduced by the Verb on to the nominal modifiers. The issue is to get the right degree of indeterminacy in the right place. 3.2 Second Order Indeterminacy This concerns the interaction of indeterminate verbs with indeterminate nouns - where a predicate places indeterminate requirements on an indeterminate feature such as. For example, some Russian verbs require ects that are either genitive or accusative: (40) On he (41) On he proždal svoju podrugu Irinu. (Russian) waited-for self s girlfriend-acc Irina proždal zvonka ot svoego brata Grigorija. (Russian) waited-for call-gen from self s brother Gregory Coordinated ects with one genitive and one accusative conjunct are also possible - showing that this is indeterminacy not ambiguity: (42) Včera ves den on proždal svoju podrugu Irinu i zvonka ot yesterday all day he waited-for self s girlfriend-acc Irina and call-gen from svoego brata Grigorija. (Russian) self s brother Gregory. Yesterday he waited all day for his girlfriend Irina and for a call from his brother Gregory (Levy 2001) (43) pred wait.for, [ pred he ] f : h : pred girlfriend {acc} e : pred call {gen} 5 6
(44) proždat ( ) { acc, gen} podrugu ( ) = { acc} zvonka ( ) = { gen} Such interactions are problematic for the set-based account, since they require a non-null intersection between the set of values specified by the noun and the set required by the verb, a requirement that is not possible to impose within the standard formal assumptions of LFG. 4 Complex Feature: Atomic Value We treat as a complex feature (f-structure), with attributes corresponding to each (core). Nouns and their modifiers specify negative values for the s they do not express. Verbs specify positive values for the (s) they require to be realized. (45a) is a (fully) determinate noun-acc and (45b) is a partially indeterminate nounacc/dat (45) a. acc gen - b. acc gen - dat A verb requiring an which is dat will combine with (45b) but not (45a) giving (47): (46) verb-dat: ( dat) = + (47) acc gen - dat + Indeterminate nouns have fewer negative specifications. A noun that is indeterminately accusative and dative has the following specification, ruling out the other options: (48) Papageien: ( nom) = - ( gen) = - Such nouns are compatible with a positive specification for both acc and dat. Our account straightforwardly covers (34), since no clash results from simultaneously specifying positive values for both acc and dat. (49) hilft: ( dat) = + (50) findet: ( acc) = + (51) [ ] pred he pred find, pred parrots acc + dat + c : gen - pred help, This accounts for the s of indeterminacy covered by the original DK proposal: German Free Relatives (8): (52) was: ( dat) = - ( gen) = - (53) übrig: ( nom) = + (54) gegessen: ( acc) = + Xhosa Indeterminate Agreement Requirements (31) (55) zibomzu: ( class) = 7/8 9/10 (56) izandla: ( class) = 7/8 (57) neendelebe: ( class) = 9/10 4.1 Transitivity Problem Adjectives and Determiners state negative values for features, placing further constraints on the features of the noun they modify. (58) alten: ((adj ) nom) = - ((adj ) acc) = - ((adj ) gen) = - Fully indeterminate rosa pink : (59) rosa: (no specifications) (60) Er findet und hilft rosa Papageien. (German) he finds and helps pink-nom/acc/dat/gen parrots-acc/dat He finds and helps pink parrots. (61) Er hilft *alte/alten Papageien. (German) he helps *old-acc/old-dat parrots-acc/dat He finds and helps old parrots. 7 8
4.2 Second order Problem The verb proždat places indeterminate requirements: (62) proždat : ( {acc gen}) = + (63) podrugu: ( nom) = - ( gen) = - ( dat) = - (64) zvonka: ( acc) = - ( nom) = - ( dat) = - (65) pred wait.for, [ pred he ] pred girlfriend acc + p : gen - inst - w : prep - pred call acc - z : gen + inst - prep - This allows the verb to govern coordinated ects with different features, as long as each conjunct is compatible with a positive specification for either acc or gen. 5 Conclusion and Further Issues Our approach uses underspecification rather than set values to handle indeterminacy. It is formally simple, and correctly allows for incremental and monotonic refinement of requirements in particular contexts. the treatment of multi-featural agreement indeterminacy in DK appears to carry over to our account: (66) weil [Ihr das Haus] und [Franz den Garten] kauft because you.2pl the house and Franz.3pl the garden buy because you buy the house and Franz buys the garden (German) We have not addressed the issue of which forms are indeterminate, and which are ambiguous Blevins (2000) We have focussed on grammatical, and have not said anything about semantic Seemingly indeterminate s with num and gen are still to be investigated within this set of assumptions: DK suggest that num and gen are never indeterminate and that these data follow from the interaction between coordination and feature distribution References Blevins, James P. 2000. Markedness and agreement. Transactions of the Philological Society 98(2):233 262. Crysmann, Berthold. 2005. Syncretism in german: a unified approach to underspecification, indeterminacy, and likeness of. In S. Müller, ed., Proceedings of the HPSG05 Conference. Stanford, CA: CSLI Publications: http://www-csli.stanford.edu/publications. Dalrymple, Mary and Ronald M. Kaplan. 2000. Feature indeterminacy and feature resolution. Language 76(4):759 798. Daniels, Michael W. 2001. On a type-based analysis of feature neutrality and the coordination of unlikes. In F. van Eynde, L. Hellan, and D. Beermann, eds., On-line Proceedings of the HPSG 01 Conference. Dy la, Stefan. 1984. Across-the-board dependencies and in Polish. Linguistic Inquiry 15(4):701 705. Groos, Anneke and Henk van Reimsdijk. 1979. Matching effects in free relatives: A parameter of core grammar. In A. Belletti, L. Brandi, and L. Rizzi, eds., Theory of Markedness in Generative Grammar: Proceedings of the 1979 GLOW Conference, pages 171 216. Pisa: Scuola Normale Superiore de Pisa. Ingria, Robert J. P. 1990. The limits of unification. In Proceedings of the 28th Annual Meeting of the ACL, Pittsburgh, pages 194 204. Association for Computational Linguistics. Johnson, Mark and Samuel Bayer. 1995. Features and agreement in Lambek categorial grammar. In Proceedings of the Formal Grammar Workshop. Levy, Roger. 2001. Feature indeterminacy and the coordination of unlikes in a totally well-typed HPSG. Unpublished ms, Stanford University. Levy, Roger and Carl Pollard. 2001. Coordination and neutralization in HPSG. In F. van Eynde, L. Hellan, and D. Beermann, eds., On-line Proceedings of the HPSG 01 Conference. Pullum, Geoffrey K. and Arnold M. Zwicky. 1986. Phonological resolution of syntactic feature conflict. Language 62(4):751 773. Voeltz, Erhard. 1971. Surface constraints and agreement resolution: Some evidence from Xhosa. Studies in African Linguistics 2(1):37 60. Zaenen, Annie and Lauri Karttunen. 1984. Morphological non-distinctiveness and coordination. In Proceedings of ESCOL 84, pages 309 320. (67) kauft: ( ) {x, y} ( x pers) = 2 ( x num) = pl ( y pers) = 3 ( y num) = sg 9 10