Quantification of first language transfer effect on second language acquisition

Shi, Shuju

Permalink

https://hdl.handle.net/2142/117664 Copy

Description

Title

Quantification of first language transfer effect on second language acquisition

Author(s)

Shi, Shuju

Issue Date

2022-12-02

Director of Research (if dissertation) or Advisor (if thesis)

• Shih, Chilin
• Hasegawa-Johnson, Mark
• Shosted, Ryan

Doctoral Committee Chair(s)

Shih, Chilin

Committee Member(s)

• Yan, Xun
• Tang, Yan

Department of Study

Linguistics

Discipline

Linguistics

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Second Language Acquisition
• L1-mandarin L2-english
• Vowel Inventory

Language

eng

Abstract

Theories for second language acquisition of phonology/phonetics and pronunciation/accent often resort to the similarity/dissimilarity between sound inventories of the first language (L1) and the second language (L2). In reality, the measurement of similarity between the sound inventories of two languages can be complicated by the distribution of sounds within each inventory as well as the interaction of phonology and phonetics within each inventory and between the two inventories. This dissertation attempts to quantify similar- ity/dissimilarity of sounds between two language inventories, examine how well the quantified measures could explain L1 influence on L2 acquisition and realize automatic prediction of error patterns of L2 production and level of difficulty of sounds in the L2 based on the sim- ilarity/dissimilarity between L1 and L2 sound inventories. Three studies are conducted for each of the mentioned aims. Study 1 reviews the phonological representations and examines phonetic variations of vowels in Mandarin and English first, followed by acoustic analysis of the vowel inventories in each language, a comparison of vowel inventories between the two languages and finally it presents the analytical results of vowel production by L1-Mandarin L2-English learners. The results show that: (1) L1 transfer effect is most salient on L2 production by low proficiency level speakers and L1 assimilation are best explained at the phonemic level; (2) Production by learners at the low proficiency level shows little sign of allophonic variation but production by learners of the high proficiency level show clear signs of approaching L1-like allophonic variations. Study 2 proposes an approach to quantify the L1 transfer effect by characterizing the phonetics-phonology interface between learners’ L1 and L2. Methodologies used include: (1) L1-Mandarin L2-English pronunciation is analyzed in a way to address both tonal influence from Mandarin and syllable condition influence from English in addition to the influence of vowel categories; and (2) Principal Component Analysis (PCA) is adopted to simulate learners’ phonological space(s), i.e., the real-valued vector embedding(s) of their L1 and L2 vowel inventories to test three different hypotheses: whether speakers use the feature and feature weights in their L1, those in the L2 by native speakers, or the combination of the two to differentiate sounds in their L2. Results show that the predicted pronunciation patterns by the proposed approach can in general be related to the actual L2 pronunciation patterns especially for learners at the low proficiency level. In addition, the current findings seem to favor the hypotheses that learners’ could be using features or feature weights of their L1 or a combination of that from both their L1 and the target language to differentiate sounds in the target language but not by the feature and feature weights of the target language alone. Study 3 examines intrinsic level of difficulty and L1-dependent level of difficulty for L2 learning and realizes automatic classification of level of difficulty. We use the Pillai’s trace scores between the L1 English and L2 English data as the ground truth of the level of difficulty for each sound. The posteriors of each sound derived by either the L1 Mandarin acoustic models or the L1 English acoustic models are used as features. The classification is done using Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel function in a 10-fold cross validation manner. The relatively high accuracy using either the L1-Mandarin posteriors or the L1-English posteriors suggests that both approaches are effective in modeling the level of difficulty. The fact that the accuracy using the L1-English posteriors is higher than the results using the L1-Mandarin posteriors indicates that for L1-Mandarin L2-English acquisition the intrinsic level of difficulty in English plays a more important role.

Graduation Semester

2022-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/117664

Copyright and License Information

Copyright 2022 Shuju Shi

Owning Collections