Phonological segmentation neighborhoods

Abstract

Phonological networks constructed from a metric of sound similarity between lexical items, commonly referred to as phonological neighborhood density (PND), have revealed network characteristics of the mental lexicon through the combined use of graph theory and psycholinguistic tasks. PND has long been a metric to account for co-activation of lexical items in the mental lexicon. The purpose of this dissertation was to investigate the role that co-activation plays in speech processing in a tonal language under the lens of a complex system built upon phonological similarity. The study begins by constructing a Mandarin syllable inventory from participant-based phonological associations collected in a neighbor generation task. This inventory was then used to build fourteen phonological networks (8 with tone and 8 without tone) based on existing proposals of Mandarin syllable segmentation. A model selection procedure was used with mixed effect models to identify the optimal fit between the participants' reaction times and the lexical statistics representing each network schema in the remaining 5 tasks. The goal of each analysis was to identify the segmentation schema and neighborhood statistic/s that best accounted for each task. The findings reveal that the spreading of activation through similarity neighborhoods during auditory lexical processing is adaptive to the demands of the task at hand, both in the segmentation schema and the directional effect. In explicit mental search (neighbor generation) lexical items were activated through a network that was nontonal and unsegmented (CGVX), where greater density aided speech production. Through the use of the model selection procedure, a false positive was identified in an auditory shadowing task. In a second task, with stimuli of greater regularity between segment and syllable length, greater density sped reaction times according to the tonal complex rime segmented schema (C_G_VX_T). Finally, an auditory lexical decision task was implemented that featured two classes of Mandarin nonwords: tone gap nonwords, which consist of existing syllables in the syllable inventory combined with one of the four lexical tones that together do not point to an existing lexical item; and syllable gap nonwords, which consist of existing lexical tones assigned to syllables that do not exist in the inventory but whose segments do. Activation spread through a tonal unsegmented schema for monosyllabic words (CGVX_T), while similar to words in the neighbor generation and auditory word repetition tasks, greater density facilitated reaction times. The tone gap nonwords identified a tonal onset/rime schema (C_GVX_T), while the syllable gap nonwords revealed another false positive. Contrary to real words, greater density slowed reaction time for the tone gap nonwords. An account is proposed that places the PND effect at a post-phonological stage, wherein segmentation is indicative of the integration of orthographic and phonological representations (CGVX and CGVX_T), or the lack of an orthographic influence (C_G_VX_T).