310 likes | 486 Views
IS SENTENCE VIABLE? The 3 rd International Conference on Cognitive Science Moscow, June 21, 2008. Andrej A. Kibrik ( kibrik@comtv.ru ) Vera I. Podlesskaya ( podlesskaya@ocrus.ru ). Does spoken language consist of sentences?. Sheer facts: Spoken language is the primary form of language
E N D
IS SENTENCE VIABLE?The 3rd International Conference on Cognitive ScienceMoscow, June 21, 2008 Andrej A. Kibrik(kibrik@comtv.ru) Vera I. Podlesskaya(podlesskaya@ocrus.ru )
Does spoken language consist of sentences? • Sheer facts: • Spoken language is the primary form of language • Spoken language does not contain periods, question marks and other explicit signals of sentence boundaries • Research question: • Is sentence, as a theoretical construct, as identifiable and as basic for the primary form of language as it is (or as it is thought to be) for written language?
Sentence in spoken language • Position 1: sentence is a universal and basic unit of language • Assumption typically held by not only by linguists but also by other cognitive scientists • “With no more than 50 to 100 K words humans can create and understand an infinite number of sentences” (Bernstein et al. 1994: 349-350) • Psycholinguistics: “Sentence processing” • But sentence is very far from being obvious in spoken language • Position 2: avoidance of the issue, typical of discourse-oriented linguists • If so, how could sentences become so much entrenched in written language?
Night Dream Stories • Corpus of spoken Russian stories • Speakers: children and adolescents • Subject matter: retelling of night dreams • Discourse type: monologic narrative (personal stories) • Speech act type: declaratives
Two basic features of spoken discourse • Segmentation • Transitional continuity
Segmentation • Elementary discourse units (EDUs) • Identified on the basis of a conjunction of prosodic criteria: • Tempo pattern • Loudness pattern • Integral tonal contour • Presence of an accentual center • Pausing pattern • Speakers tend to organize EDUs as clausal units
Example of segmentation Z54 Discourse transcription
Transitional continuity • Term by J. DuBois et al. 1992 • Alternative term by Sandro V. Kodzasov: phase • Discourse semantic category: ‘end’ vs. ‘non-end’ (=expectation of a forthcoming end) • Hierarchical nature of phase • End of tentative sentence – falling tonal accent • Non-end – rising tonal accent
A canonical example of the transitional continuity distinction z57:15-16 • ..(0.4) /\Мы-ы’ ..(0.4) \как бы за них /взя-ались, ..(0.4) /\My-y’ ..(0.4) \kakbyzanix /vzja-alis’, We sort of at them got.hold • ...(0.5) и-и ввь= || ..(0.2) полетели \вве-ерх. • ...(0.5) i-i vv’= || ..(0.2) poleteli \vve-erx. • and flew upward • Rising (“comma”) • Non-end • Falling (“period”) • End • If things were that easy, sentence would be uncontroversial
....(1.5) /\Озеро ...(0.5)какое-то, ..(0.3) (Или /\речка, или /\озеро, но по-моему \озеро, потому что’..(0.2) как-то-оw...(0.6)\маленькоетакое, \небольшое.) ....(1.0) ’и-иh ...(0.7)через/него..(0.3)как-то\бревнокакое-то, типа\моста. ....(1.5) /\Ozero ...(0.5) kakoe-to, Lake some ..(0.3) (Ili/\rečka, Either river ili/\ozero, or lake no po-moemu\ozero, but I guess lake potomučto’..(0.2) kak-to-oWbecause somehow...(0.6) \malen’koe takoe, small such \nebol’šoe.) minor ....(1.0) ’i-iH ...(0.7) čerez /negoand across it ..(0.3) kak-to\brevno kakoe-to,somehow log some tipa\mosta.like bridge Uncanonical situation: Non-end with a falling tonal accent
The problem of two kinds of falling • The existence of non-final falling may call relevance of sentence into question • However, the distinction between two kinds of falling is very systematic • The two kinds of falling: • are prosodically distinct • have distinct discourse functions
Prosodic criteria of the final vs. non-final falling distinction • Primary criteria: • Target frequency band • Post-accent behavior
Criterion 1: Target frequency band • Final falling (“period”): targets at the bottom of the speaker’s F0 range • Non-final falling (“faling comma”): targets at level several dozen Hz (several semitones) higher
F0 graph for the “lake” example 12 10 12 8 5 \ozero, \malen’koe \nebol’ \brevno kakoe \mosta. takoe, šoe. -to,
Non-final falling (210 Гц),final falling (170 Гц),rising, post-rising falling Z54: 4-5 170 Hz 210 Hz
Criterion 2: Post-accent behavior • Final falling (“period”): steady falling on the post-accent syllables • Non-final falling (“comma”): lack of falling on post-accent syllables, often rise of tone (V-curve)
V-curve z26 260 Hz 240 Hz 235 Hz
Secondary criteria • Pausing pattern • Reset vs. latching • Steepness of falling • Interval of falling
The final vs. non-final falling distinction • A speaker’s prosodic pattern must be identified • On its basis the difference between final and non-final falling distinction can be identified with a high degree of robustness
Contexts of non-final falling • Anticipatory mirror-image adaptation • Inset • Stepwise falling
Anticipatory mirror-image adaptation • ....(1.8)Когдая\услышала, Kogda ja \uslyšala,when I heard • ...(0.5)что-о/бомбагремит, čto-o /bomba gremit, that bomb growls
Inset • /Входитэто ...(0.5)/\ма-аль↑чик,/Vxoditèto ...(0.5) /\ma-al’↑čik, enters here boy • ’ ’ ..(0.1)/\нук\другому, ’ ’ ..(0.1) /\nu k \drugomu, well to another • ..(0.1) и\говорит: ..(0.1) i \govorit: and says
....(1.5) /\Озеро ...(0.5)какое-то, ..(0.3) (Или /\речка, или /\озеро, но по-моему \озеро, потому что’..(0.2) как-то-оw...(0.6)\маленькоетакое, \небольшое.) ....(1.5) /\Ozero ...(0.5) kakoe-to, Lake some ..(0.3) (Ili/\rečka, Either river ili/\ozero, or lake no po-moemu\ozero, but I guess lake potomučto’..(0.2) kak-to-oWbecause somehow...(0.6) \malen’koe takoe, small such \nebol’šoe.) minor Stepwise falling 210 Hz 190 Hz 160 Hz
The status of sentence • In the speech of most speakers final falling is clearly distinct from non-final patterns • Final intonation, expressly distinct from non-final intonation (both rising and falling), makes the notion of sentence valid for spoken discourse • Speakers “know” when they complete a sentence and when they do not • Apparently, spoken sentences are the prototype of written sentences
Functions of sentences • Ease the processing by creating intermediate informational chunks • Chafe: superfoci of consciousness
However • Identification of sentences is possible only on the basis of a complex analytic procedure • It is dependent on prior understanding of a speaker’s prosodic “portrait” • There are prototypes of final and non-final fallings, but there are intermediate instances, therefore sentencehood may be a matter of degree • A significant tune-up is necessary to apply the procedure to a different discourse type or a different language • Therefore, sentence is an elusive, intermediate, non-basic unit of language
EDUs:distribution in terms of number of words Sentences:distribution in terms of number of EDUs EDUs vs. sentences: degree of variability 53% –3±1 80% –3±2
EDUs vs. sentences: degree of variability • Unlike EDUs, sentences are highly variable • Speakers with short sentences • Speakers with long sentences equaling stories • Clause chaining
Conclusions • Sentence is an intermediate hierarchical grouping between a whole discourse and an EDU (roughly, clause) • Sentence is very far away from being a basic unit of spoken language
Acknowledgement Member of our project Nikolay Korotaev