Phoneme restoration methods reveal prosodic influences on syntactic parsing: Data from Bulgarian.

摘要

This dissertation introduces and evaluates a new methodology for studying aspects of human language processing and the factors to which it is sensitive. It makes use of the phoneme restoration illusion (Warren, 1970). A small portion of a spoken sentence is replaced by a burst of noise. Listeners typically mentally restore the missing phoneme(s), reporting that they heard a complete utterance with the noise burst overlaid on it. Phoneme restoration is used in the present study to explore the role of prosodic phrasing, specifically the impact of a major prosodic break on the interpretation of a sentence where no other type of disambiguation remains. Materials are constructed so that the phoneme(s) that listeners indicate they had heard reveal which syntactic/semantic interpretation of the sentence they had computed.;Two target constructions in Bulgarian are investigated in a series of experiments, using the same materials throughout: NP/S coordination as in (1) and RC attachment as in (2). The target items are originally disambiguated by morphological agreement: number agreement between subject and verb in (1), and gender agreement between the head noun and the relative pronoun in (2). The sentences in each pair differ with respect to one or more phonemes in the word on which the morphological disambiguation is encoded; these are the phonemes that are noise-replaced in the experimental materials. (Prosodic boundaries are indicated by ||.).;(1) a. Nakraia sreshtnahme Ani || i Ivan i Mimi biaha vav vaztorg. In the end meet-past-1p.pl Ani and Ivan and Mimi were in ecstasy 'In the end, we met Ani and Ivan and Mimi were in ecstasy.'.;b. Nakraia sreshtnahme Ani i Ivan || i Mimi bese vav vaztorg. In the end meet-past-1p.pl Ani and Ivan and Mimi was in ecstasy 'In the end, we met Ani and Ivan and Mimi was in ecstasy.'.;(2) a. Podtseniha advokata || na pevitsata koiato kupi imenieto. underestimate-past lawyer-m of singer-f who-f buy-past estate-det '(They) underestimated the lawyer of the singer who=N2 bought the estate.'.;b. Podtseniha advokata na pevitsata || kojto kupi imenieto. underestimate-past lawyer-m of singer-f who-m buy-past estate-det '(They) underestimated the lawyer of the singer who=N1 bought the estate.'.;The alternative structures for both (1) and (2) are also disambiguated by the location of a major prosodic boundary, as shown. For the NP/S coordination a boundary after the first noun (1a) signals that only that noun ( Ani) is the object of the first clause; a boundary after the second noun (1b) signals that the first two nouns (Ani and Ivan) form a coordinate object in the first clause. For the RC construction, a boundary after the first noun (2a) favors the interpretation that the RC modifies the second noun (the singer), whereas a boundary after the second noun (2b) favors RC modification of the entire complex NP (the lawyer of the singer). After noise-replacement of the disambiguating phonemes, only these prosodic boundaries differentiate the competing interpretations.;Three different response tasks are employed to tap which phonemes listeners mentally restored. In a visual word choice task, participants indicate which of two words (e.g. biaha / bee in the case of (1)) they had 'heard' in the stimulus sentence (Experiments 1a-c). In a sentence repetition task, participants repeat back the sentence after hearing it (Experiments 2a-b). In a sentence shadowing task, participants repeat back the sentence as they are listening to it (Experiments 3a-b). In all three tasks, responses confirmed that prosodic boundary location provided viable disambiguation of syntactic structure. However, the effect of prosody differed for the two constructions. Both prosodic contours disambiguated NP/S coordination equally well, but a break after the first noun was a stronger structural cue for RC attachment than a break after the second noun.;These findings for Bulgarian comport well with data from more traditional methodologies on prosodic influences on coordination and RC attachment interpretation in other languages, thereby validating the sensitivity of the new methodology. A broader range of other ambiguities in Bulgarian were explored in a subsequent pilot experiment. Also, the original target constructions were tested in written form, with a simulated 'ink blot' obscuring the morphological agreement. Two tasks were employed: silent reading with visual word choice (Experiments 4a-b) and silent reading with sentence repetition (Experiments 5a-b). Visual grouping of phrases disambiguated NP/S coordination quite effectively, but had little impact on RC attachment interpretation. (Abstract shortened by UMI.).