1. Introduction
Recent research shows that second language (L2) orthography – the way L2 sounds and words are represented in spelling – can affect the way L2 speakers produce and perceive L2 sounds. However, research has so far focused on the effects of spelling in L2 learners who are native users of alphabetic writing systems. The present paper therefore investigated whether the spellings of English sounds can affect speech production in L2 speakers who are native users of a non-alphabetic writing system.
1.1 Effect of second language orthography on second language phonology
There is increasing evidence that L2 orthography can affect L2 speakers’ production and perception of L2 sounds (for reviews, see Bassetti, Reference Bassetti, Piske and Young-Scholten2008; Bassetti, Escudero & Hayes-Harb, Reference Bassetti, Escudero, Hayes-Harb, Hayes-Harb, Bassetti and Escudero2015). Although some studies reported no orthographic effects (Escudero, Reference Escudero2015; Escudero & Wanrooij, Reference Escudero and Wanrooij2010; Simon, Chambless & Alves, Reference Simon, Chambless and Alves2010) and others found facilitative effects (Escudero, Hayes-Harb & Mitterer, Reference Escudero, Hayes-Harb and Mitterer2008; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2013), many studies show that L2 orthographic forms can lead to nonnative-like pronunciations, such as the addition, deletion, and – more often – the substitution of target sounds (Bassetti, Reference Bassetti, Piske and Young-Scholten2008; Davidson, Reference Davidson2010; Escudero, Simon & Mulak, Reference Escudero, Simon and Mulak2014; Hayes-Harb, Nicol & Barker, Reference Hayes-Harb, Nicol and Barker2010; Pytlyk, Reference Pytlyk2011; Young-Scholten & Langer, Reference Young-Scholten and Langer2015). Second language speakers can add sounds that are represented in the orthographic forms of L2 words; for instance, they may pronounce comb /kəʊm/ as [kəʊmb] (Bassetti & Atkinson, Reference Bassetti and Atkinson2015). They also omit sounds that are not represented in the L2 orthographic forms. For instance, learners of Chinese as a foreign language produced the triphthong /uei/ in /suei/ “year of age” correctly when it was spelled with three letters as <wei>, but as a diphthong */ui/ when spelled with two vowel letters as <sui> (Bassetti, Reference Bassetti, Guder, Jiang and Wan2007). The most widely studied orthographic effect on L2 phonology is sound substitution, whereby L2 speakers substitute the target sound with another sound, either from the target language or absent from the target language, because of its spelling. Examples of substitutions with other L2 sounds include the substitution of word-final [t] with [d] in GermanL2 words spelled with <d>, for example, Bund /bʊnt/ “federation” (Young-Scholten, Reference Young-Scholten, Burmeister, Piske and Rohde2002) or the substitution of a flap allophone of /t, d/ with [t̪], [ð] or [d̪] in American EnglishL2 words spelled with <t, tt, d, dd> (e.g., city, chatty, body, or buddy; Vokic, Reference Vokic2011). As an example of substitutions with sounds that do not exist in the target language, EnglishL1 beginner learners of SpanishL2 produce [v] in SpanishL2 words spelled with <v> (e.g., vulgar /bulˈɣaɾ/ “vulgar”), even though this sound is not present in Spanish (Zampini, Reference Zampini1994).
Orthographic effects on L2 phonology can be intra- or inter-orthographic (Bassetti, Reference Bassetti2017). An intra-orthographic effect is when an L2 grapheme is phonologically recoded using L2 grapheme-phoneme correspondences (GPCs) – rules that govern how a unit of writing is recoded into a unit of sound. For instance, L2 speakers of German pronounced word-final <d> as [d] (instead of [t]), because in German <d> generally represents the phoneme /d/, but not in word-final position (e.g., in Bund /bʊnt/ “federation”; Young-Scholten, Reference Young-Scholten, Burmeister, Piske and Rohde2002). An inter-orthographic effect is when an L2 grapheme is recoded using L1 GPCs. For instance, native Italian speakers studying English sometimes realize <w> as [v] in words such as wood, because <v> represents /v/ in Italian loanwords such as <water> (/ˈvater/) and <watt> (/vat/; Bassetti & Atkinson, Reference Bassetti and Atkinson2015). Researchers have argued that many sound substitutions are caused by incongruences between L1 and L2 GPCs like this one, when a grapheme represents different phonemes in the two languages of an L2 speaker (Hayes-Harb et al., Reference Hayes-Harb, Nicol and Barker2010). Various studies investigated the effects of L1-L2 grapheme incongruence, and consistently found that incongruence negatively affects L2 speech production (Escudero, Reference Escudero2015; Escudero et al., Reference Escudero, Hayes-Harb and Mitterer2008; Nimz, Reference Nimz2016; Pytlyk, Reference Pytlyk2011). Escudero et al. (Reference Escudero, Simon and Mulak2014) manipulated congruency to compare the effects of congruent and incongruent graphemes on L2 word recognition in beginner and naïve listeners, and confirmed that congruent graphemes have a facilitative effect and incongruent ones have a negative effect.
Finally, with regards to the locus of orthographic effects on phonology, researchers working on native phonologies have proposed two main explanations. Some researchers have argued that these effects are due to the co-activation of the orthographic and phonological codes of a word during speech production (Muneaux & Ziegler, Reference Muneaux and Ziegler2004; Rastle, McCormick, Bayliss & Davis, Reference Rastle, McCormick, Bayliss and Davis2011). Other researchers have argued that orthographic forms affect phonological representations: that is to say, the way a word is written affects its phonological code in the mental lexicon (Taft, Reference Taft2006). However, it is not clear which of these explanations can best explain orthographic effects on L2 phonology. Bassetti (Reference Bassetti2017) argued that L2 orthographic forms are very likely to affect phonological representations, whereas co-activation of L2 orthographic forms is a less likely explanation because it would also require the co-activation and incorrect online application of L1 GPCs during L2 speech production. At any rate, research on L2 speech can contribute to understanding the nature of orthographic effects on phonology.
1.2 Effects of number of letters on L2 sound length
A series of studies by Bassetti and colleagues found that the number of letters in the spelling of an English sound affects the duration of the sound in the production of ItalianL1 speakers of EnglishL2 (Bassetti, Reference Bassetti2017; Bassetti & Atkinson, Reference Bassetti and Atkinson2015; Bassetti, Sokolović-Perović, Mairano & Cerni, Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). Unlike English, where consonant length is not contrastive, Italian has a phonological contrast between singleton and geminate (short and long) consonants (Laver, Reference Laver1994). Singleton and geminate consonants distinguish minimal word pairs (two words that differ in one phoneme, Clark & Yallop, Reference Clark and Yallop1995), such as cane /ˈkane/ “dog” and canne /ˈkanːe/ “reeds” (/ː/ represents a long sound). Italian geminate consonants are represented with double consonant letters, for example, /nː/ in canne. ItalianL1 speakers of EnglishL2 produced the same English consonant as longer when it was spelled with two letters (e.g., [t] in kitty) than one (e.g., in city; Bassetti, Reference Bassetti2017). Homophonic English word pairs containing the same consonant spelled with a single letter or double letters were produced as minimal pairs distinguished by a singleton or geminate consonant; for instance, finish was produced with a singleton [n] ([ˈfɪnɪʃ]) and Finnish with a geminate [nː] ([ˈfɪnːɪʃ]) (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). This means that ItalianL1 speakers have established a phonological contrast between geminate and singleton consonants that does not exist in English and is based on the written forms of English words interpreted through their L1 knowledge.
While orthographic effects on consonant duration were clear, effects on vowel duration were more nuanced (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). Vowel length is not contrastive in either Italian or English, but English vowels can be long or short. English long vowels are typically represented by vowel digraphs; for instance, /iː/ is represented by double letters <ee> in sheep. However, long vowels can also be represented by a single letter, such as /iː/ in he and ski (Carney, Reference Carney1994). Italians, who generally do not make the qualitative distinction between English tense and lax vowels, can distinguish these vowels using duration, and can rely on the number of letters in a vowel's spelling to determine its duration. Bassetti and Atkinson (Reference Bassetti and Atkinson2015) found that ItalianL1 speakers of EnglishL2 produced the same English vowel with longer duration when it was spelled with double vowel letters (e.g., moon, /muːn/) than when it was spelled with a single vowel letter (e.g., June, /dʒuːn/). Bassetti and colleagues (Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018) confirmed these results using homophonic word pairs distinguished by the number of letters in the spelling of the same long vowel, such as seen and scene (both /siːn/). Italians produced these words as non-homophonic (producing a longer vowel in seen than in scene), whereas native English speakers produced the two vowels with the same duration.
1.3 Open questions in research on orthographic effects on L2 phonology
In spite of the rapidly growing body of evidence, some aspects of orthographic effects on L2 phonology have not been investigated yet. First of all, only a limited range of languages and scripts have been researched, with the focus so far on European languages written in the roman alphabet. Further cross-linguistic research is needed in order to establish when orthographic effects are likely to happen. In particular, hardly anything is known about orthographic effects on L2 phonology in native users of a non-alphabetic writing system. At a more general level, English word spellings, which represent phonemes, may not influence L2 speakers whose L1 writing system represents syllables or morphemes rather than phonemes. At a more specific level, when the L2 speakers’ native writing system is not alphabetical, there are no incongruences between the GPCs of the native orthography and of the L2 orthography. A few studies, mostly by Hayes-Harb and colleagues, investigated cross-scriptal orthographic effects; their focus was on the learning of sounds of a novel language through a novel orthography (Hayes-Harb & Cheng, Reference Hayes-Harb and Cheng2016; Pytlyk, Reference Pytlyk2011; Showalter, Reference Showalter2018; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2015). A more recent study looked at cross-scriptal effects in EnglishL1 learners of RussianL2 (Simonchyk & Darcy, Reference Simonchyk and Darcy2018), whose two writing systems are both alphabetic. There has been, to the best of our knowledge, no research on orthographic effects on L2 speech production in biscriptals whose L1 writing system is not alphabetic.
Second, there have been very few studies of experienced L2 speakers producing real L2 words. Researchers have rarely investigated orthographic effects on the speech production of bilinguals, focussing instead mostly on learners of novel languages (Davidson, Reference Davidson2010; Hayes-Harb et al., Reference Hayes-Harb, Nicol and Barker2010; Pytlyk, Reference Pytlyk2011) and beginner L2 learners (Young-Scholten, Reference Young-Scholten, Burmeister, Piske and Rohde2002; Zampini, Reference Zampini1994). The few studies of experienced L2 speakers investigated the effects of orthography on the learning of pseudowords (Escudero et al., Reference Escudero, Hayes-Harb and Mitterer2008; Piske, Flege, MacKay & Meador, Reference Piske, Flege, MacKay and Meador2002). Only a small number of recent studies have investigated orthographic effects on the production of real words in sequential bilinguals living in a target language environment (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018; Vokic, Reference Vokic2011).
Given the absence of cross-scriptal research and the very limited research on sequential bilinguals, the present study aimed at investigating orthographic effects in JapaneseL1–EnglishL2 sequential bilinguals.
1.4 Consonant and vowel length in Japanese and English
In Japanese, length is contrastive in both consonants and vowels. The geminate-singleton opposition is used to distinguish minimal pairs such as /kata/ “frame” and /katːa/ “bought” (Kawahara, Reference Kawahara and Kubozono2015), and the opposition between short and long vowels is used to distinguish minimal pairs such as /to/ “door” and /toː/ “ten” (Kubozono, Reference Kubozono and Kubozono2015a).
Japanese and English consonants
The consonant inventories of Standard Japanese and Standard British English both include the voiceless plosives /p, t, k/ (Okada, Reference Okada1999; Roach, Reference Roach2004), which were the object of the present study. Consonant length is not phonemic in English. Voiceless plosives can be geminated in Japanese (Kubozono, Reference Kubozono and Kubozono2015a); voiced ones are only geminated in loanwords (Kawagoe, Reference Kawagoe and Kubozono2015; Otaka, Reference Otaka2009).
The primary acoustic correlate of gemination in Japanese is consonant duration: geminates are two to three times longer than singletons (Kawahara, Reference Kawahara and Kubozono2015). Secondary correlates include durations of the preceding and the following vowel (Idemaru & Guion, Reference Idemaru and Guion2008; Kawahara, Reference Kawahara and Kubozono2015), as well as non-durational correlates such as intensity, fundamental frequency fall across the surrounding vowels, and voice quality (Idemaru & Guion, Reference Idemaru and Guion2008).
Japanese and English vowels
Standard Japanese has five short monophthong vowels /i e a o u/ (pronounced [i e a o ɯ]) and five corresponding long monophthongs /iː eː aː oː uː/ (Kubozono, Reference Kubozono and Kubozono2015a; Okada, Reference Okada1999). Long vowels are two to three times longer than short vowels (Hirata, Reference Hirata2004; Hirata & Tsukada, Reference Hirata and Tsukada2009; Ueyama, Reference Ueyama2000), but they do not differ in quality (Kubozono, Reference Kubozono and Kubozono2015a; Tsukada, Reference Tsukada2009a; Ueyama, Reference Ueyama2003; but see Hirata & Tsukada, Reference Hirata and Tsukada2009).
Standard British English has seven lax monophthongs /ɪ e æ ʌ ɒ ʊ ə/ and five tense monophthongs /iː ɑː ɔː uː ɜː/ (Roach, Reference Roach2004). They differ in both quantity and quality: tense vowels are longer than lax vowels (Hillenbrand, 1995; Kondaurova & Francis, Reference Kondaurova and Francis2010; Leung, Jongman, Wang & Sereno, Reference Leung, Jongman, Wang and Sereno2016; Peterson & Lehiste, Reference Peterson and Lehiste1960; Williams & Escudero, Reference Williams and Escudero2014), and they are more peripheral (for example, /iː/ is higher and more front than /ɪ/; Ladefoged & Johnson, Reference Ladefoged and Johnson2011; Peterson & Barney, Reference Peterson and Barney1952).
Sound length representation in written Japanese
Japanese is written in a mixture of kanji (characters), which are phonologically opaque, and kana, generally called ‘syllabaries’: hiragana is used for native words and katakana for loanwords (Coulmas, Reference Coulmas2003). Each kana symbol represents a mora; for instance, /hon/ is written with two kana symbols ほ /ho/ and ん /n/ (Otake, Reference Otake and Kubozono2015; Ueyama, Reference Ueyama2000). Kana symbols are highly phonologically transparent. For instance, the mora /ka/ is always represented by the graphemes か in hiragana and カin katakana, and vice versa. Other scripts include furigana, which is used to gloss the pronunciation of kanji, and romanisation, which is used to input text in computers and other electronic devices.
Length is represented by the addition of a kana symbol (Laburne, Reference Laburne2012). Gemination is represented by a reduced-size TSU symbol before the consonant symbol (っin hiragana and ツin katakana), as in the minimal pair おと/oto/ “sound” and おっと/otːo/ “husband”. Vowel length is represented in hiragana by an additional vowel symbol after the first mora: for instance, the vowel symbol お /o/ distinguishes short and long vowel in the minimal pair とり /tori/ “bird” and とおり/toːri/ “street”. In katakana it is represented by a line ー, for instance, カー /kaː/ “car”.
1.5 Phonology in JapaneseL1 speakers of EnglishL2
JapaneseL1 learners of EnglishL2 need to acquire a more complex phonological system including consonantal contrasts and the qualitative distinction in vowels that do not exist in their L1. Most research on Japanese speakers’ pronunciation of English consonants has focused on the /l/-/r/ contrast (Aoyama, Flege, Guion, Akahane-Yamada & Yamada, Reference Aoyama, Flege, Guion, Akahane-Yamada and Yamada2004). For vowels, most researchers investigated the lack of qualitative distinction between /ɪ/ and /iː/, both generally produced as Japanese /i/ (Tsukada, Reference Tsukada2009b; Ueyama, Reference Ueyama2003), the consistent substitution of schwa with other vowels, and vowel epenthesis (for effects of orthography on these, see e.g., Detey & Nespoulous, Reference Detey and Nespoulous2008; Dupoux, Kakehi, Hirose, Pallier & Mehler, Reference Dupoux, Kakehi, Hirose, Pallier and Mehler1999). These effects of L1 phonology may in fact be reinforced by Japanese orthographic input, as kana does not distinguish /l/-/r/ (e.g., both loanwords <glass> and <grass> are クラ ス in katakana; Okada, Reference Okada, Cook and Ryan2015).
There is evidence that English orthography can affect pronunciation in JapaneseL1 speakers of EnglishL2. For instance, Japanese learners of EnglishL2, who often cannot distinguish /l/ and /r/, can produce the correct sound if they know the word's spelling (Brown, Reference Brown1998; Sheldon & Strange, Reference Sheldon and Strange1982). English orthography can also affect EnglishL2 pronunciation indirectly, via the phonological and orthographic form of English loanwords. English consonants written with double letters can undergo gemination in loanwords; for example, inner /ˈɪnə/ is adapted in Japanese with a geminate /nː/, becoming /inːaː/. Gemination can even distinguish minimal pairs of loanwords, such as /bagu/ “bug” and /bagːu/ “bag” (English /bʌg/ and /bæg/ respectively; Kawagoe, Reference Kawagoe and Kubozono2015; Kubozono, Reference Kubozono and Kubozono2015b). English tense and lax vowels can be adapted as the same vowel qualitatively, but with different duration; for instance, /ɪ/ in pin /pɪn/ is adapted as Japanese /i/ in the loanword /piN/, whereas /iː/ in key /kiː/ becomes /iː/ in the loanword /kiː/ (Irwin, Reference Irwin2011).
1.6 The Present Study
This study investigated for the first time whether L2 orthography affects L2 speech in biscriptal sequential bilinguals with a non-alphabetic L1 writing system, by testing the effects of number of letters in English word spellings on the duration of consonants and vowels in English words produced by JapaneseL1–EnglishL2 sequential bilinguals. If JapaneseL1–EnglishL2 bilinguals produce the same English sound as longer in English words spelled with double letters and as shorter in words spelled with a single letter, this would mean that English orthographic forms result in nonnative-like speech production in these biscriptal speakers of EnglishL2. Previous research has demonstrated that Italian speakers of EnglishL2 have established a long-short contrast in their L2 phonological systems that does not exist in English, and rely on orthographic forms to determine sound length: they produce the same English consonant or vowel as longer when it is spelled with a digraph than with a single letter, such as in Finnish vs. finish (Bassetti, Reference Bassetti2017; Bassetti & Atkinson, Reference Bassetti and Atkinson2015; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). These effects are caused by the interaction between the phonological and orthographic systems of Italian and English. The present study is aimed at disentangling the respective effects of the native phonology and orthography as follows:
First, this study investigated orthographic effects on L2 phonology in native users of a non-alphabetic writing system. Previous findings of orthographic effects on L2 consonant duration were attributed to L2 speakers decoding L2 orthographic forms – specifically, number of letters – using their native GPC rules, where double letters represent geminate consonants (Bassetti, Reference Bassetti2017; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). However, Italian and English share the same alphabet. It is not clear whether the number of letters in English spelling affects sound duration in native users of a non-alphabetic writing system. There may well be no orthographic effects when the two languages of an L2 speaker use different scripts, because, as argued by Showalter & Hayes-Harb (Reference Showalter and Hayes-Harb2015), such speakers do not need to override L1 grapheme–phoneme correspondences. Research on learners of novel scripts or novel symbols (Hayes-Harb & Cheng, Reference Hayes-Harb and Cheng2016; Pytlyk, Reference Pytlyk2011; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2015), while important to establish how orthography can affect the early stages of L2 learning, cannot answer this question. Cross-scriptal evidence from experienced L2 speakers from a non-alphabetic background is needed to extend findings of orthographic effects on phonology beyond native users of alphabetic writing systems.
Second, the present study investigated native speakers of a language that has contrastive length in both consonants and vowels. Previous research on number of letters effects investigated Italian native speakers, for whom consonant length is contrastive, but vowel length is not (Bassetti, Reference Bassetti2017; Bassetti & Atkinson, Reference Bassetti and Atkinson2015; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). Orthographic effects were found on the duration of both EnglishL2 consonants and vowels, however the ratio of long to short vowels (between 1.2 and 1.4) was somewhat smaller than the ratio normally found for English tense vs. lax vowels (between 1.2 and 1.7). This may be due to the interaction between orthographic input (double letters) and acoustic input (no vowel duration differences), or it could be caused by the absence of contrastive length for vowels in their native language. To further clarify the nature of these orthographic effects, it is necessary to test EnglishL2 speakers whose native language has contrastive vowel length.
This study aimed at testing whether L2 orthographic forms can result in a length contrast in the L2 production of both consonants and vowels in native users of a language with contrastive consonant and vowel length and with a non-alphabetic writing system. Japanese was chosen because it has contrastive consonant and vowel length, and a non-alphabetic writing system where long sounds are represented consistently in the orthography but not by means of letter digraphs. Furthermore, native Japanese speakers may be affected by L2 orthographic forms, because kana is highly phonologically transparent, and native users of phonologically transparent alphabetic writing systems are likely to rely on orthographic forms to establish the pronunciation of a second language (Erdener & Burnham, Reference Erdener and Burnham2005). However, it is not clear whether this reliance on L2 orthography is found in native users of a transparent writing system that is not alphabetical. Finally, participants were instructed sequential bilinguals. They had completed their English language instruction, and were living and using English in an English-speaking environment. This is because research on experienced L2 speakers is needed to show whether orthography has long-term effects beyond the early stages of learning and the instructed environment investigated by most researchers so far (for studies of sequential bilinguals in naturalistic environments, see Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018, and Vokic, Reference Vokic2011; for the distinction between L2 learner and bilingual/L2 user, see e.g., Cook, Reference Cook and Cook2002).
This study investigated whether JapaneseL1–EnglishL2 speakers would produce the same English sound as longer when spelled with double letters than when spelled with a single letter. Bilinguals were compared with a group of native British English speakers. Participants performed a delayed word repetition task, whereby they heard a native speaker's pronunciation of a phrase and repeated a target word. Delayed repetition tasks have been used to investigate L2 speech production (e.g., Flege, Munro & MacKay, Reference Flege, Munro and MacKay1995), including orthographic effects on L2 speech production (Bassetti, Reference Bassetti2017; Bassetti & Atkinson, Reference Bassetti and Atkinson2015). In the task, the orthographic input was read out loud and the output was spoken, in order to test orthographic effects on phonology in the absence of orthography. This is because orthographic effects may be stronger with than without orthographic input in beginners and in learners of novel languages or novel words (Rafat, Reference Rafat2016; Young-Scholten, Reference Young-Scholten, Burmeister, Piske and Rohde2002; Zampini, Reference Zampini1994). However, this may not apply beyond the early stages of learning in experienced L2 speakers producing real L2 words (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018).
If Japanese–English bilinguals systematically produce the same EnglishL2 sound as longer when spelled with a digraph than when spelled with a single letter, this would indicate that they have established a contrast in their L2 phonological system that does not exist in the phonological systems of native English speakers. This would show that orthographic forms affect L2 phonology in L2 speakers with a non-alphabetic L1 writing system, a population that has hitherto not been studied.
2. Methods
2.1 Participants
Participants were 16 Japanese–English bilinguals and 16 native speakers of British English, with similar gender and age compositions (females = 13 in both groups; age: M bilinguals = 34.94 years, SD = 8.11; M natives = 33.81, SD = 16.77) and no known language, visual or reading difficulties.
The native English participants were speakers of Southern British English. The Japanese–English participants were predominantly speakers of Standard Japanese (n = 7) or the Osaka variety (n = 6; other varieties: Shizuoka, Hokkaido and Iwami, all n = 1), with no knowledge of other languages with contrastive length. They had studied English as an academic subject for an average of nine years and six months, with a mean age of onset of acquisition of around 11 years. While in school, two-thirds had studied only with non-native-speaking teachers (n = 10), the rest with native English teachers or both. Participants had been living in an English-speaking country for an average of nearly five years, and about two-thirds were postgraduate students at a British university (n = 10). They rated their English proficiency on average as 3.75 on a 7-point scale, where 7 = native-like. Participants reported spending more time reading than listening to English (on average 9.44 hours/week vs. 3.14 hours/week). They considered a native-like pronunciation as quite important, with a mean rating of 5.13 on a 7-point scale, where 7 = very important. Among the students, nine had taken an IELTS test, with the average score of 6.5 (The International English Language Testing System, or IELTS, uses a nine-band scale with levels of proficiency ranging from non-user [band score 1] to expert user [band score 9]). Table 1 presents a summary of participant details for the Japanese–English bilingual group.
Table 1. Summary of participant details for Japanese-English bilinguals.
2.2 Materials
Materials were 32 English words: nine word pairs containing a target consonant and seven containing a target vowel. Consonant pairs and carrier phrases were taken from Bassetti (Reference Bassetti2017), and vowel pairs and carrier phrases from Bassetti and Atkinson (Reference Bassetti and Atkinson2015). A full list of word pairs and the corresponding carrier phrases is provided in the Appendix.
For target consonants, there were three target word pairs for each of the English voiceless plosives /p, t, k/. Within each word pair, the target consonant occurred in the same intervocalic contexts and within the same word stress patterns, but it was spelled with a single letter in one word (C-consonant) and with double letters in the other one (CC-consonant), such as /t/ in city and kitty (for more information about the words, including frequencies, see Bassetti, Reference Bassetti2017). Among the CC-words, two were loanwords in Japanese: one contained a geminate in Japanese (floppy) and one contained a singleton consonant (kitty).
The target vowels were: /iː/ (five word pairs), /ɔː/ and /uː/ (both one word pair; vowel choice reflects the relative frequencies of the GPCs for <ee> and <oo> in the English lexicon). Within each pair, the target vowel was spelled with a single (or zero) letter (V- vowel) or with double letters (VV-vowel), such as in the pair June–moon (/dʒuːn/ - /muːn/). There were three homophonic pairs, such as scene–seen (both /siːn/) and four near-minimal pairs, in which the target vowel occurred in the same rime, that is to say it was word-final or it was followed by the same consonant in both words, such as June–moon (for more information about the words, see Bassetti & Atkinson, Reference Bassetti and Atkinson2015).
Each word was presented orally within a six-word phrase, recorded by a male native speaker of Standard British English. A truncated version of each phrase was obtained by deleting the target word and all words following it. Each target word was also introduced by an image.
2.3 Task and procedure
Participants performed a delayed word repetition task, followed by a spelling task.
In the delayed word repetition task (Bassetti, Reference Bassetti2017), a phrase and a related image were used to help participants recognise the target word. First, an image appeared in the centre of the screen. The image illustrated the meaning of the phrase that followed (for instance, an image of condiments). After clicking on an on-screen button with the mouse, the participant heard over headphones a pre-recorded phrase, such as Salt and pepper, oil and vinegar. The phrase and the related image helped participants recognise the target word: in this case, pepper. Before repeating the phrase, the participant counted aloud backward from seven to one in English. This distractor task was used to eliminate traces of the phonological input from memory. After repeating the phrase, the participant clicked on another on-screen button, and heard a truncated version of the phrase, such as Salt and. They then had to produce the first missing word (the target) in the carrier sentence The word … is missing three times (e.g., The word pepper is missing). The carrier sentence was used to place the target word in the nuclear position within the intonational phrase. The three repetitions were used to obtain a reliable measure of sound duration. For a graphic representation of the task, see Bassetti (Reference Bassetti2017).
A spelling task was used to test the participants’ knowledge of the spelling of the 32 target words. This is because effects of number of letters were only expected when participants knew the correct number of letters in the spelling of the target sound. For each word, the participant saw the same image as in the word repetition task. Upon clicking on an on-screen button, they heard a recording of the word produced in isolation by the same male native speaker and wrote the spelling on paper. The words were presented in random order.
Participants were tested individually in a sound-attenuated booth. A researcher provided instructions before each task and supervised four practice trials before the word repetition task. Stimuli were presented on a MacBook Air laptop running PsyScope X (Cohen, MacWhinney, Flatt & Provost, Reference Cohen, MacWhinney, Flatt and Provost1993) and over headphones. The responses were recorded using the Praat software (Boersma & Weenink, Reference Boersma and Weenink2012) via an AKG D80 microphone connected to a computer. Participation was voluntary and unpaid.
2.4. Data analysis
For each of the target sounds, durations were measured from spectrogram and waveform displays in Praat (Boersma & Weenink, Reference Boersma and Weenink2012), using standard segmentation procedures (Foulkes, Docherty & Jones, Reference Foulkes, Docherty, Jones, Di Paolo and Yaeger-Dror2010; Keating, Linker & Huffman, Reference Keating, Linker and Huffman1983; Turk, Nakai & Sugahara, Reference Turk, Nakai, Sugahara, Sudhoff, Lenertova, Meyer, Pappert, Augurzky, Mleinek, Richter and Schlieser2006). Plosive duration was measured as the duration of the closure, which is the main phonetic cue to gemination in Japanese plosives (Kawahara, Reference Kawahara and Kubozono2015). Closure duration was measured from the offset of the preceding vowel, characterised by a drop in amplitude and a loss of the second and higher formants, to the onset of the plosive release. If there were multiple bursts, the first one was taken as the onset of the release. Pre-aspiration, which occurred frequently in the production of the native English group, was not included in closure duration (Foulkes & Docherty, Reference Foulkes and Docherty2006). For vowels, duration was measured from the onset of the clearly defined formant pattern, in particular F2 and the higher formants, accompanied by an amplitude increase and a complex waveform display. The offset of the vowels was marked at a drop in amplitude and the cessation of the second and higher formants. In words where the target vowel was followed by the nasal /n/ or the fricative /z/, the boundary was marked at the point of visible spectral discontinuity. Acoustic analysis was performed by the first and the third author. A subset of data (243 words or 8.3% of all data) was re-analysed by a third researcher, who was naïve to the purpose of the present study. The intraclass correlation coefficient for all three researchers was 0.96, 95% CI [0.95; 0.97], F(242, 484) = 80.42, p < 0.001.
In total, 3,072 words were recorded (thirty-two participants’ three repetitions of sixteen word pairs). Mean target sound durations were calculated from the three repetitions for each word for each participant. When a repetition was lost (for example, due to mispronunciation or a sudden background noise), mean durations were calculated from two repetitions. 2.54% of word pairs were lost because the participant failed to produce one or both words (22 consonant pairs and 17 vowel pairs). An additional 3.26% of word pairs (33 consonant and 17 vowel pairs) were excluded from the analysis because the participant spelled the target sound with the same number of letters (both with a single letter, or both with double letters). Since no effects of number of letters on sound production were expected if the participant did not know the spelling of the target sound, these pairs were later entered in a separate analysis reported below. Preliminary inspection of data revealed that the word pair these-cheese was anomalous, because bilinguals produced a longer vowel in these than in cheese. This happens because EnglishL2 learners are explicitly instructed to produce a long vowel in these in order to differentiate it from singular this (Bassetti & Atkinson, Reference Bassetti and Atkinson2015). Since the long vowel in these is a washback effect of instruction, this anomalous pair was eliminated from the main analysis.
For each word pair, a duration ratio was calculated by dividing the duration of the sound spelled with double letters by the duration of the sound spelled with a single letter; for example dividing the duration of [t] in kitty by the duration of [t] in city. This was done to eliminate the potential confound of different speaking rates when comparing the two groups (Hirata & Whiton, Reference Hirata and Whiton2005) and to eliminate the potential effect of different intrinsic durations of English vowels (Peterson & Lehiste, Reference Peterson and Lehiste1960). To remove outliers, the top and bottom 1% of ratios were excluded (seven consonant pairs and four vowel pairs), and data was logarithmically transformed to eliminate skewness.
Statistical analysis was performed using R version 3.2.4 (R Core Team, 2016) with RStudio (RStudio Team, 2015) and additional packages lmertest (Kuznetsova, Brockhoff & Christensen, Reference Kuznetsova, Brockhoff and Christensen2016), doBy (Højsgaard & Halekoh, Reference Højsgaard and Halekoh2016) and ggplot2 (Wickham, Reference Wickham2009). Two maximal mixed effects models were created for consonant and vowel data (Barr, Levy, Scheepers & Tily, Reference Barr, Levy, Scheepers and Tily2013), including the fixed effect of language background (native speakers, bilinguals), and random intercepts for participants and word pairs. Using backward model selection, each simplified model was compared to the previous one, and the model with the smallest AIC (Akaike's Information Criterion) was selected.
3. Results
3.1 Consonant duration ratios
The geometric mean of the CC/C ratio (the duration of the same consonant when spelled with double letters and when spelled with a single letter, log-transformed) was 1.11 in the Japanese–English bilingual group (95% CI [1.07; 1.16]), and close to one (1.01) in the native English group (95% CI [0.98; 1.04]). This reflects the fact that descriptively the Japanese produced the same consonant as longer when spelled with double letters (M CC-words = 90 ms, SD = 23) than with a single letter (M C-words = 81 ms, SD = 18), whereas the English speakers produced consonants of similar duration regardless of spelling (M CC-words = 67 ms, SD= 18; M C-words = 66 ms, SD = 17).
Table 2 shows the final model and the results of the model. The CC/C ratio was higher among bilinguals than among native speakers, p < 0.001. The random intercept of word pairs reflected the variation in ratios across word pairs.
Table 2. Results of mixed-model analysis of the effects of language background (Native English Speakers, Japanese-English Bilinguals) on logged consonant duration ratio. R code: modC2 = lmer(CC_C_RATIOLog ~ GROUP+(1|WORDPAIR),data = ProdDataCFinal, REML = FALSE).
To confirm that the observed differences in consonant duration in the bilingual group were due to the number of letters in the consonant's spelling, a further analysis was performed on misspelled words. We compared the durations of the two target consonants in each word pair where both consonants had been spelled with the same number of letters. For instance we compared the duration of [k] in pairs that had been spelled *<acuse>-<acute>, or <accuse>-*<accute>. No durational differences were found in such word pairs, M C-words = 96 ms, SD = 24; M CC-words = 97 ms, SD = 26, t(11) = - 0.15, p = 0.887.
To check whether consonant duration in EnglishL2 words was affected by the pronunciation of English loanwords in Japanese, we looked at the CC/C duration ratios of the two word pairs that contained an EnglishL2 word that was a loanword in Japanese: floppy, which is produced with a geminate in Japanese, and kitty, which is produced with a singleton. Despite small descriptive differences in the predicted direction, the CC/C duration ratio was neither larger in copy-floppy than in the other /p/ pairs (M copy-floppy = 1.28, SD = 0.11; M other /p/ = 1.22, SD = 0.16, t(12) = 1.46, p = 0.169), nor smaller in city-kitty than in the other /t/ pairs (M city-kitty = 1.13, SD = 0.18; M other /t/ = 1.18, SD = 0.22, t(15) = - 0.74, p = 0.472).
3.2 Vowel duration ratios
The geometric mean of the VV/V ratio (the duration of the same vowel when spelled with double letters and when spelled with a single letter, log-transformed) was 1.12 in the Japanese–English bilingual group (95% CI [1.08; 1.16]), and 1.04 in the native English group (95% CI [1.01; 1.06]). This reflects the fact that descriptively the Japanese produced the same vowel as longer when spelled with double letters (M VV-words = 293 ms, SD = 62) than with a single letter (M V-words = 264 ms, SD = 62), whereas the English speakers produced vowels of similar duration regardless of spelling (M VV-words = 285 ms, SD = 69; M V-words = 275 ms, SD = 67).
Table 3 shows the final model and the results of the model. The VV/V ratio was higher among bilinguals than among native speakers, p = 0.002.
Table 3. Results of mixed-model analysis of the effects of language background (Native English Speakers, Japanese-English Bilinguals) on logged vowel duration ratio. R code: modV3 = lmer(VV_V_RATIOLog ~ GROUP+(1|SIDNO), data = ProdDataVFinal, REML = FALSE).
A further analysis was performed on the words containing vowel targets that were misspelled by the Japanese–English bilinguals, as was done for the words with target consonants. No durational differences were found in the bilingual group's production of word pairs where the target sound had been spelled with the same number of letters (M V-words = 215 ms, SD = 50; M VV-words = 223, SD = 27), t(9) = - 0.83, p = 0.427.
4. Discussion
The aim of the present study was to test whether orthographic effects on L2 speech production are found in biscriptal bilinguals – bilinguals whose two languages use different scripts – and whether L2 orthographic forms – single or double letter spelling – affect L2 consonant and vowel duration in sequential bilinguals whose L1 has contrastive length in both consonants and vowels. As predicted, JapaneseL1–EnglishL2 sequential bilinguals systematically produced the same consonant or vowel as longer when orthographically represented with double letters than when represented with a single letter. As these L2 speakers were native users of a non-alphabetical script, unlike Italian participants in previous studies, they did not have a correspondence between double letters and long sounds in their native orthography. Results confirmed previous findings that L2 orthography can lead L2 speakers to establish two phonological categories – long and short – corresponding to one category in the target language (Bassetti, Reference Bassetti2017; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). However, results also showed that orthographic effects are not restricted to the users of alphabetic writing systems, but are present in biscriptal bilinguals. There were no differences in the magnitude of orthographic effects in the Japanese–English bilingual group, as the duration ratio of the double-letter sound to the single-letter sound, calculated as the mean of individual ratios, was 1.14 for both consonants and vowels (with the exception of the anomalous pair these-cheese). Unlike Italian speakers of L2 English (Bassetti & Atkinson, Reference Bassetti and Atkinson2015), Japanese–English bilinguals show very similar effects in vowels and in consonants – most likely because vowel length is contrastive in Japanese, but not in Italian.
4.1 Effect of number of letters on consonant duration
The Japanese–English sequential bilinguals, but not the native English speakers, produced the same consonant as longer when spelled with double letters than when spelled with a single letter. Orthography affected these biscriptals’ L2 consonant production, in line with previous findings from native users of the roman alphabet (Bassetti, Reference Bassetti2017; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018).
This is arguably the first evidence that orthography affects L2 phonology in sequential bilinguals whose L1 writing system is not alphabetical. This shows that effects can take place across scripts, and in the absence of relevant GPCs in the L1 writing system. The few previous studies that investigated orthographic effects on phonology across scripts found little or no orthographic effects (Hayes-Harb & Cheng, Reference Hayes-Harb and Cheng2016; Pytlyk, Reference Pytlyk2011; Showalter & Hayes-Harb, Reference Showalter and Hayes-Harb2015). However, this absence of effects was probably due to lack of familiarity with the orthography or the phonology of the target language, as such studies investigated naïve learners of a novel language and orthography. The present study, by testing experienced users of two scripts, showed that orthographic effects can be found across scripts. Furthermore, Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2015) argued that the reason for the absence of orthographic effects in biscriptals is that such learners do not need to override native GPCs. In addition to that finding, the present results show that L2 phonology can be affected by orthography even when the L1 writing system does not have GPCs that can interfere with the recoding of the L2 orthographic forms. The Japanese–English biscriptals’ L1 writing system does not represent gemination with double letters, as gemination is represented by a subscript tsu symbol. These bilinguals establish a long-short contrast in their EnglishL2 phonological system, due to their L1 phonological system, and use English orthography (single or double letters) as a marker of length, even in the absence of such GPCs in their native orthography. The use of double letters to represent long consonants in romanised Japanese may play a role in this, but this cannot be considered a native GPC given that Japanese speakers hardly ever use romanisation. The effect of orthography on L2 phonology is therefore not restricted to native users of alphabetic writing systems, or to graphemes that can be decoded incorrectly due to native GPCs.
The orthographic nature of these effects was confirmed by the lack of orthographic effects in misspelled word pairs: that is to say, pairs in which the target consonant was erroneously spelled with the same number of consonant letters, as in acute-*acuse or *accute-accuse. The CC-consonant and the C-consonant were produced with similar duration when the participant thought they were spelled with the same number of letters. This confirms the close relationship between L2 speakers’ written and spoken production of L2 sounds.
Although orthographic forms resulted in the production of a longer and a shorter form of the same consonant in EnglishL2, the actual ratios were smaller than the ratios both in the participants’ native language and in ItalianL1 speakers of EnglishL2, showing that the present ratios were probably caused by an interaction of orthographic input, phonological input, and the difference between English and Japanese scripts. First, the CC/C ratio was on average 1.14, which is smaller than the geminate/singleton ratio of two or more typically found in Japanese. This suggests that orthographic effects are weakened by the absence of confirmatory evidence in native spoken input, where there are no durational differences between consonants. Previous research had also found smaller ratios in EnglishL2 than in the native language (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). Second, the CC/C ratio of these Japanese native speakers was smaller than the ratios previously found in ItalianL1 speakers of EnglishL2, suggesting that orthographic effects may be weaker in biscriptal than in monoscriptal bilinguals. Previously reported mean ratios of 1.66 (Bassetti, Reference Bassetti2017) and 1.70 (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018) in instructed L2 learners could be due to learners’ limited exposure to spoken input. However, Bassetti and colleagues (Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018) reported a mean ratio of 1.38 in intervocalic consonants produced by Italian–English sequential bilinguals, who were roughly comparable to the present study's participants in terms of length of residence in England (on average six years for the Italians and five for the Japanese), as well as in other relevant factors such as age of onset of acquisition (both around eleven), length of instructed learning (both around 8–9 years), type of teachers (mostly non-native teachers) and language teaching practices (mostly textbook-based). Although it is not possible to compare the results of different studies (Bassetti, Reference Bassetti2017, only tested learners; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018, used different materials), differences in scripts may lead to a weaker impact of orthography in Japanese than in Italian speakers of English.
Results confirmed and extended a number of previous findings on orthographic effects on L2 speech production, as follows: first, results confirmed that L2 orthography affects L2 speakers who are native users of a phonologically transparent writing system (Erdener & Burnham, Reference Erdener and Burnham2005), and extended this to native users of a phonologically transparent non-alphabetic script. This is interesting because native Japanese speakers are often presented as typical examples of readers of an opaque orthography (see reviews in Koda, Reference Koda, Cook and Bassetti2005, Reference Koda2007), because kanji are phonologically opaque. However, it appears that L2 speakers who are native users of both a transparent script and an opaque one still rely on L2 orthographic forms. Future research could investigate speakers of various native languages to compare the effect of different levels of phonological transparency of the native orthography or orthographies. Second, results confirmed that orthography affects L2 phonology beyond the early stages of language learning, as effects were found in sequential bilinguals with nine and a half years of instruction and almost five years of residence in an English-speaking country. This confirms recent evidence that orthographic effects are found in L2 speakers with long experience of L2 learning (Bassetti, Reference Bassetti2017) or with lengthy naturalistic exposure in a target-language environment (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018), and not only in beginners learners and naïve learners of novel languages (e.g., Zampini, Reference Zampini1994). Finally, there was much variation in the Japanese group in terms of proficiency, length of stay, and other variables reported in the participants’ description. This is because our sample is representative of the varied population of native Japanese speakers living in the UK. Future research could investigate whether individual variables such as length of stay, proficiency and amounts of spoken and written input can impact the orthographic effects on speech production.
Further research is needed to establish whether the phonological forms of cognates and loanwords impact orthographic effects on L2 pronunciation. In this study durational ratios did not differ significantly between English double-letter words that were loanwords in Japanese and those that were not. If the phonological form of loanwords impacted gemination in EnglishL2, there would have been stronger effects in the word floppy, whose loanword contains a geminate in Japanese, and weaker effects in the word kitty, whose loanword contains a singleton. No significant differences were found. However, previous studies found stronger gemination in ItalianL1 speakers of EnglishL2 when an L2 word was a loanword or a cognate containing a geminate in Italian, such as floppy (Bassetti, Reference Bassetti2017) and Finnish (Italian finnico, Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018). Evidence is inconsistent, but it suggests that orthographic effects resulting in gemination may be reinforced by the presence of a geminate in cognates and loanwords in L1.
The effects of number of consonants on consonant duration in Japanese speakers of EnglishL2 cannot be explained by a single factor. Since gemination does not exist in English, and is not represented by double letters in Japanese, there may be various reasons why Japanese speakers produce double-letter English consonants as longer. Below we discuss: intra-orthographic effects (incorrectly assuming a correspondence between double letters and longer sound duration); effects of knowledge of other scripts (romaji); washback of teaching practices (furigana); and spoken input (from other Japanese speakers of EnglishL2 and from English loanwords in Japanese).
An intra-orthographic effect explanation would be that Japanese speakers interpret double consonant letters in English orthography as a representation of longer sound length. As a result, they create <CC>-/Cː/ GPCs between double letters and geminates, which do not exist in either the target or the native language (e.g., <pp>-/pː/). This would be similar to what native Italian speakers do, but it would be intra-orthographic (internal to English orthography), rather than inter-orthographic (caused by the incorrect application of native GPCs to their L2). This interpretation runs against the proposal by Showalter and Hayes-Harb (Reference Showalter and Hayes-Harb2015) that cross-scriptal orthographic effects on L2 phonology are unlikely if there are no L1 GPCs to override. Another explanation may be that Japanese native speakers are also literate in romaji, or romanised Japanese, which represents Japanese geminates using double letters. However, romaji by itself would not be a sufficient explanation, because Japanese speakers are unlikely to be affected by the GPCs of romaji, a script they use extremely rarely, more than by the GPCs of English, which they have been using for years. Other factors include: the presence of geminates in the phonological and orthographic (katakana) forms of English loanwords, which may misleadingly point to the presence of long and short consonants in English; the use of the supplementary script furigana which is used in schools for glossing the pronunciation of Japanese kanji and English words, and is considered the source of various nontarget-like productions (Okada, Reference Okada, Cook and Bassetti2005). While it may well be impossible to pin down the cause or causes of these effects, it is clear that the main cause of gemination in Japanese speakers’ EnglishL2 speech production is the number of letters in English word spellings.
Finally, it should be noted that participants reported more exposure to written than spoken input, and this may partly explain why orthographic forms affect their speech production. While this study did not address this question, future research could investigate whether the amount of orthographic input impacts orthographic effects on L2 speech.
4.2 Effect of number of letters on vowel duration
The Japanese–English bilinguals, but not the native speakers, produced the same vowel as longer when spelled with double letters than when spelled with a single letter. This happened because these EnglishL2 speakers overgeneralised the English language correspondence between single letter spelling and short vowel to contexts where it does not apply, whether to the single letter <i>, which represents a long vowel in words such as ski (/skiː/), or to the letter <e> in scene, which is in fact part of the grapheme <e_e>, which represents /iː/.
The mean VV/V duration ratio of 1.14 (excluding the anomalous pair these-cheese) was the same as the mean CC/C duration ratio. This shows that number of letters has similar effects on the production of consonants and vowels in native speakers of a language that has contrastive length in both consonants and vowels. Previous studies of Italian speakers of L2 English found smaller ratios in vowels than in consonants (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018), probably because Italian has contrastive length in consonants but not in vowels. The only exception to this pattern were the words these-cheese, whose vowel was produced with similar duration by both L1 and L2 speakers (this is because L2 participants produced a long vowel in these, as was reported by Bassetti and Atkinson, Reference Bassetti and Atkinson2015).
The mean duration ratio of 1.14 is smaller than both the long/short vowel ratio of 2–3 normally found in Japanese, and the tense/lax ratio of 1.2–1.7 normally found in English (Hillenbrand, 1995; Kondaurova & Francis, Reference Kondaurova and Francis2010; Leung et al., Reference Leung, Jongman, Wang and Sereno2016; Peterson & Lehiste, Reference Peterson and Lehiste1960; Williams & Escudero, Reference Williams and Escudero2014). As in previous studies with ItalianL1 speakers of EnglishL2, this smaller ratio is most likely due to the interference between the single vowel letters in the orthographic input, which are generally recoded as short vowels, and the long vowels in the spoken input. It is also possible that at least some of these L2 speakers may be aware that the grapheme <V_e> represents a long vowel in some of the present study's words such as scene. This possibility should be tested in future studies, and it would also be of interest to compare the duration of these orthography-induced shorter vowels with the duration of lax and tense EnglishL2 vowels to establish whether they are as long as EnglishL2 lax vowels or constitute a third category with durations that are between those of lax and tense vowels.
The mean VV/V ratio is also smaller than the 1.20 ratio found in ItalianL1–EnglishL2 sequential bilinguals with lengths of residence comparable to those of the participants in the present study (Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018, results from double letter vowels; results differed for vowels spelled with digraphs such as <ou>). The mean ratio is however close to the ratio of 1.13 found in ItalianL1 learners of EnglishL2 (Bassetti & Atkinson, Reference Bassetti and Atkinson2015). As with consonants, it is possible that the smaller orthographic effects in these Japanese–English bilinguals are due to their biscriptalism, as the effects of number of letters on the duration of L2 sounds could be weaker in non-alphabetic native readers.
4.3 The locus of orthographic effects on L2 phonology
With regards to the debate of whether orthographic effects on phonology are due to co-activation of orthographic and phonological codes during speech production (Muneaux & Ziegler, Reference Muneaux and Ziegler2004; Rastle et al., Reference Rastle, McCormick, Bayliss and Davis2011), or to effects of orthographic forms on phonological representations in L2 speakers’ minds (Taft, Reference Taft2006), the present findings strongly support the latter approach. Namely, our results are compatible with the idea that L2 phonological representations are generated during the L2 word learning process, through the interaction between L2 orthographic forms – often reinterpreted using native GPC rules – and L2 phonological forms – often reinterpreted according to the native phonology (Bassetti, Reference Bassetti, Piske and Young-Scholten2008). In the case of Japanese bilinguals, double letters may be recoded as long sounds during the process of learning English words due to the various reasons discussed above (English GPCs, romaji, furigana). English orthographic forms would then result in phonological representations of EnglishL2 words containing long consonants or vowels. At the same time, the native spoken input would provide conflicting evidence, whereby the word is heard with a short sound. This may be partially offset by spoken input from fellow Japanese speakers of English and from English loanwords. The complex interaction between orthographic and phonological input best explains why an EnglishL2 sound spelled with double letters is produced as longer than the same sound spelled with a single letter, but shorter than long sounds in the native language. To further clarify the role of orthographic forms in shaping phonological representations of L2 words, future research could investigate whether exposure to word spellings containing double letters during the early stages of spoken word learning leads EnglishL2 speakers to learn L2 spoken words containing long sounds.
Whether the co-activation of orthographic and phonological forms also plays a role is a difficult question to answer. Bassetti (Reference Bassetti2017) stated that, in the case of Italian L2 speakers, orthographic effects would occur if the L2 orthographic form is activated together with L1 GPCs. However, this explanation does not apply to JapaneseL1–EnglishL2 speakers because there are no double letters in Japanese scripts. If orthographic forms are co-activated, then JapaneseL1–EnglishL2 speakers must have established nonnative-like GPCs in English, whereby double letters correspond to long sounds. These nonnative-like GPCs could then be used to decode orthographic forms online and interfere with phonological representations as follows: if Japanese EnglishL2 speakers’ phonological representations contain short sounds, in line with phonological input they receive from native speakers, then co-activated orthographic forms may interfere if the L2 speakers have established nontarget-like English GPCs whereby double letters correspond to long sounds. If these speakers’ phonological representations contain long sounds, then co-activated orthographic forms could reinforce the effects of the phonological forms. In either case, the present study confirms that L2 speech production is affected by the phonological systems and writing systems of both the target language and the native language(s), and it is very likely that orthographic forms affect the establishment of L2 phonological forms during the word learning process.
5. Conclusions
This study extends the growing body of research into orthographic effects on second language phonology by investigating these effects cross-scriptally. Results have implications for research on L2 phonology and for L2 teaching practice.
Regarding theories of orthographic effects on phonology, as discussed above, the present findings support the view that orthography affects the phonological representations of L2 words, even in native users of a non-alphabetic writing system. Furthermore, much L2 phonology research investigates L2 speakers who have a single phonological category corresponding to two target categories, but the present results confirm that L2 speakers can have two categories corresponding to a single target category (Bassetti, Reference Bassetti2017). Third, effects of number of letters on vowel duration were found in the absence of orthographic input, confirming previous evidence from reading aloud tasks (Bassetti & Atkinson, Reference Bassetti and Atkinson2015; Bassetti et al., Reference Bassetti, Sokolović-Perović, Mairano and Cerni2018).
With regards to the teaching of L2 pronunciation, teachers should be aware that L2 spellings can affect pronunciation in non-alphabetic L2 learners (e.g., Swan & Smith, Reference Swan and Smith2001, do not report spelling among factors influencing Japanese learners), and that orthography can affect sound length (an aspect of L2 phonology that is not discussed in materials for L2 pronunciation teachers, e.g., Swan & Smith, Reference Swan and Smith2001, do not mention gemination). Since the effects of L2 spellings on L2 phonology are still found in L2 speakers with years of naturalistic exposure, L2 teachers could consider this factor in their teaching practice.
Supplementary Material
For supplementary material accompanying this paper, visit https://doi.org/10.1017/S136672891900035X
Author ORCIDs
Bene Bassetti, 0000-0003-4547-4085; Mirjana Sokolović-Perović, 0000-0003-4974-2710
Acknowledgements
This work was partially supported by a Leverhulme Trust Research Grant [grant number: RPG 2013 180] awarded to Bene Bassetti and Jackie Masterson.
We are grateful to Dr Tania Cerni for advice on the statistical analysis, to Dr Rosalba Nodari for contributing to the acoustic analysis, and to all participants.
