Pitch Prominence Matters: Perception of Thai Tones by Seoul Korean and Kyungsang Korean Speakers

This paper is a follow‐up analysis of a previous study on the perception of Thai tones by speakers of various L1s differing in pitch prominence. Varying degrees of pitch prominence (Mandarin, tone = high; Japanese, pitch accent = intermediate; English, word stress = reduced; Korean, none = low) globally resulted in the expected hierarchy of performance: Mandarin (M = 87% correct), Japanese (M = 77%), English and Korean (M = 67% for both). However, the equal performance between the English and Korean participants was not predicted. In this study, we examined whether differences in dialectal exposure among the Koreans influenced results. Three speakers of the Korean Kyungsang dialect featuring pitch accent performed with higher accuracy than Seoul dialect speakers, reaching comparable accuracy levels with the L1 Japanese speakers. Additionally, the Seoul dialect speakers performed less accurately than the L1 English speakers. Both results confirm the originally predicted hierarchy of performance. Vance Schaefer and Isabelle Darcy 598 This study specifically examines how Thai tones are perceived by naïve listeners (= non‐learners of Thai) who speak a range of languages varying in the usage of lexically‐contrastive pitch: a tone language (Mandarin Chinese), a pitch accent language (Japanese, Kyungsang Korean dialect), a word‐stress language (English), and a language which does not employ pitch in any way to distinguish the meaning of words (standard Korean). Pitch is generally defined as the variations in the height of the voice or fundamental frequency (abbreviated as f0) used to distinguish words. We ask whether pitch prominence shapes tone perception. Prominence here is based on the function of linguistic pitch to signal lexical contrast (cf. Feature Hypothesis, McAllister, Flege, & Piske, 2002). Our working hypothesis is that the greater the prominence of lexical pitch in the first language (L1), the better the perception of non‐native tones, resulting in a hierarchy of performance among the various L1s, relative to the prominence of lexical pitch in the L1. We selected the languages we use in this study according to the role of pitch for each. The three languages of Mandarin, Japanese, and English use pitch to distinguish words (even though the domain at which they do so varies) whereas standard Korean does not use pitch to contrast words but uses pitch in a domain larger than a word (i.e., intonational phrase) like all languages that we are aware of. Thus, naïve listeners who encounter a new sound system which utilizes pitch in a new manner do not approach the task without any “tools”: their perception will most likely rely on the pitch usage defined by their L1 phonological grammar. For example, the presence in the L1 of lexical pitch contrasts has been shown to facilitate perception of non‐native tone (Burnham et al., 1996). To further characterize the role of pitch in the languages used in this study, we could arrange them according to language types on a functional scale, such as one proposed by Van Lancker in 1980, summarized in Table 1. We see that languages use pitch contrasts in different domains (from small to large), and also in the function that the pitch fulfills, whether lexical or non‐lexical (i.e., from tonal contrasts to marking focus structure or affect, for example). When the domain and function of pitch overlap between the L1 and L2, we (generally) expect perception of the non‐native linguistic pitch system to be more reliable. Combining the idea of functional use of pitch with the prominence idea, we can establish a scale of prominence for lexically‐contrastive pitch going from maximal to low (see Table 1). We also have an additional distinction between English and Japanese based on whether or not pitch can be exclusively used to distinguish words; English word stress has Pitch Prominence matters: Thai tones 599 multiple correlates in addition to pitch while Japanese relies on pitch only. As a result, we predict that if the prominence of pitch to distinguish words in your first language is high, then your accuracy in tone perception will also be high, and conversely, if it is low, then accuracy will be low as well. Based on the prominence for lexically‐contrastive pitch in a given language, specific predictions for accuracy performance on naïve tone perception by the speakers of this language can be made. However, languages do not necessarily neatly coincide within these four language types, especially when dialects are considered. For example, Japanese is often cited as a typical pitch‐accent language, but features dialects which vary in pitch‐accent systems as compared to so‐ called standard Japanese or are considered to be pitch‐accentless, e.g., Fukushima and Kumamoto dialects. Another example is standard Korean which is considered to be a language which does not use pitch to distinguish lexical items (Kim‐Renaud, 2009). However, the Kyungsang dialect spoken in the southeast region of South Korea is a pitch‐accent dialect. Furthermore, the contrastive usage of pitch may be appearing among younger speakers of the Seoul dialect (Silva, 2006). According to our scale of prominence, we would predict that speakers of a pitch‐ accentless dialect of Japanese and a pitch‐accent dialect of Korean would perform less or more accurately, respectively, than their counterparts on the perception of another lexically‐contrastive pitch system, i.e., Thai tones. Hence, care must be taken when describing the use of lexical pitch in a language, highlighting the necessity to narrowly define one’s L1 as one’s first dialect. A previous publication of part of this study’s results (Schaefer & Darcy, in press) demonstrated that the accuracy rates in the naïve perception of Thai tones as evidenced in an AXB discrimination task indeed appear to be determined by the degree to which lexically‐ contrastive pitch is used in one’s L1. These results, summarized in the following section, are in line with the Feature Hypothesis (McAllister, et al., 2002) which predicts, whether for segmental and suprasegmental dimensions alike, that the more a certain phonetic or phonological dimension is prominent in the L1, the easier it might be to learn to discern and use that dimension for L2 phonological processing (see also Dupoux et al., 2008). Accuracy scores from the study resulted in a hierarchy of performance for Thai tones discrimination: L1 Mandarin > L1 Japanese > L1 English = L1 Korean (see Table 1 for predictions). Vance Schaefer and Isabelle Darcy 600 Table 1. Overview of our pitch prominence typology and predictions for tone perception accuracy. Pitch pattern Prosodic Domain Lexical status of pitch use/function Prominence for lexical distinction? Predicted Sensitivity/ Accuracy in tone perception Tone (Mandarin) Syllable Lexical Maximal Highest PitchAccent (Japanese) Word Lexical HighIntermediate (pitch is exclusive) High-intermediate Word Stress (English) Word/Foot Lexical Lowintermediate (pitch is nonexclusive) Low-intermediate Intonation (Standard Korean) Intonational phrase, PPh Non-lexical Low Lowest Note: PPh = Phonological phrase However, an unexpected finding was the equal performance of the L1 English speakers and L1 Korean speakers. This paper addresses this unexpected result (see Schaefer & Darcy, in press, for a more extensive literature review). In this follow‐up to our study, we specifically ask whether the L1 English speakers performed lower than expected or whether the L1 Korean speakers performed higher than expected. That is, we question whether f0 information is less readily accessible for phonological discrimination for the L1 English speakers and more so for the L1 Korean speakers. There are two possible scenarios. The first one is that the word stress correlate of pitch in English is not robust enough for the L1 English speakers to access/transfer to their perception of Thai tones. The second one is that the Koreans were able to use the presence of L1 pitch accent patterns (either from the Kyungsang dialect or “young‐ generation” Seoul dialect) to aid their perception of Thai tones. In this paper, we explore the latter hypothesis and explicitly investigate whether dialectal exposure has influenced Korean performance. Pitch Prominence matters: Thai tones 601 THE SCHAEFER & DARCY STUDY Participants, Materials and Procedure Forty‐seven participants were recruited from five language groups: Mandarin (n=10; females=6), Japanese (n=12, females=11), English (n=13; females=10), Korean (n=10; females=7) and Thai (n=2; males=2). The Thai speakers were recruited to ensure that the stimuli and AXB task itself were valid for native speakers. The participants were primarily graduate students or former graduate students who were involved in language studies (i.e., language education, linguistics, applied linguistics) with the exception of 11 participants who were undergraduate students (n=3) or not involved in language studies (n = 8) (i.e., Mandarin = 3, Japanese = 3, English = 1, Korean = 3, Thai = 1). However, four individuals were cut from the final analysis as they had significant exposure to one of the other target languages in the study or differed in background from the target group, resulting in a reduction from 47 to 43 participants in total. Three English speakers had exposure to Japanese or a tone language (i.e., Mandarin or Vietnamese). One female Japanese student was an ESL student with lower exposure and proficiency in English as compared to the graduate student participants. As a result, only 11 Japanese‐speaking participants and 10 English‐speaking participants’ data were analyzed. Average ages were 27.1 years for Mandarin speakers (range 24‐31), 35.4 years for Japanese (range 25‐50), 31 years for English (range 25‐45) and 32.2 years for Korean speakers (27‐47). The two Thai listeners were 25 and 32 years old. The average time spent in an English‐speaking country was 3.5 years for the Mandarin speakers, 6.6 years for the Japanese, 4.5 years for the Koreans, and 2 years for the Thai speakers. The English speakers had spent an average of 1.7 years abroad in a non‐English speaking environment. The speakers of Mandarin included six speakers who also had various degrees of exposure to Taiwanese, another tonal language. Most had been exposed to another Chinese dialect even if they did not consider themselves a fluent speaker of that dialect. The speakers of Japanese, a pitch accent language, were recruited on the basis of speaking a dialect of Japanese which features pitch accent although not necessarily standard Japanese. Two speakers were from Tochigi and Ibaraki prefecture which are close to Fukushima prefecture, known for its pitch‐accentless dialect. The English speakers were native speakers of American English who had no proficiency in Thai, Mandarin, Japanese, Korean or any tone language. Vance Schaefer and Isabelle Darcy 602 Korean speakers were mainly from the Seoul area, but three were from the Kyungsang region where a pitch‐accent dialect of Korean is spoken, and one speaker was from Cholla, an area abutting Kyungsang but with a dialect not featuring pitch accent (self‐reported). However, exposure to Kyungsang dialect speakers might be expected for this speaker. The test stimuli consisted of 16 open CVV syllables, with a long vowel (VV). Each syllable was recorded with each of the five different Thai tones, resulting in 80 items (41 items being real words and 39 non‐words). Control stimuli were CVV and CVC syllables (all were real Thai words) composed of vowels or consonants similar to those used for the test items. The control condition also included more difficult vowels such as [ɯ], [ε], and [ɤ]. The syllables were then arranged in triplets for the AXB design. In an AXB design, four trials are needed for each comparison: AAB, ABB, BAA, BBA. For example, the two tones Low (L) and Mid (M) would be paired as LLM, LMM, MLL, and MML. If the syllable carrying such a comparison were [bi:], a trial would look like the following: [bi:]L – [bi:]L – [bi:]M. The experiment contained two conditions, test and control, with 48 trials each. In the test condition, the syllables within one triplet only differed by tone; the segmental make‐up of the syllables remained the same. In the control condition, all syllables in the triplet had the same tone but varied in either one consonant or one vowel. Furthermore, within the test condition, we included three subconditions in order to examine specific tonal comparisons: 1) Height, comparing flat tones, 2) Direction, comparing contour tones, and 3) Mixed, comparing flat tones with contour tones. These specific comparisons are not discussed in the present paper. Table 2 presents the overview of the conditions used in the study. Twelve triplets each for the direction and height condition, and 24 triplets in the mixed condition were created. All trials were randomized and put into three blocks of 32 items, respectively. Pitch Prominence matters: Thai tones 603 Table 2. Overview of the tonal comparison(s) and number of trials used