Consonant Substitution Patterns in Japanese Speakers’ English

Consonant Substitution Patterns in Japanese

Speakers’ English

Takehiko Makino(Chuo University, Tokyo, Japan)

[email protected]

Accents 2008: II International conference on native and non-native accents of EnglishUniversity Conference Centre, Łód , Polandź12-14 December 2008

1. Introduction

• Consonants are one of the most difficult parts of English pronunciation for Japanese speakers.

• This obviously reflects the difference between the consonant systems of the two languages: Japanese has only fourteen consonant phonemes while American English has twenty-four.

• It is too simplistic to assume that Japanese speakers just reduce English consonants to their own counterparts.

• We all know of “common errors,” but they have been observed only informally and the whole picture is something up in the air.

• This is understandable because a systematic study would have required large-scale databases or corpora of Japanese learners’ English pronunciation.

• However, the time is changing and we are beginning to find a number of phonetically-transcribed speech corpora projects and also some good tools for building them such as – Phon (http://phon.ling.mun.ca/phontrac)– ELAN (http://www.lat-mpi.eu/tools/elan/)

• Linguistic research outside of phonetics is already making use of learners’ corpora, and we see the findings are being applied to foreign learners’ dictionaries and grammars.

The screenshot of PhonThe screenshot of Phon

The screenshot of ELANThe screenshot of ELAN

• This study is a sequel to the preliminary observations of the range of deviation from a native-speaker norm using a sample set of transcriptions as a micro-corpus (Makino 2007) and a companion to the study concerned with vowel substitutions (Makino 2008). It is still preliminary in nature because it uses the same set of data as the previous papers.

2. The purpose of this study

• The first of the previous studies only made a list of types of deviations Japanese speakers made and looked for the types which have not been mentioned as “common mistakes” by observers of their English pronunciation. I did not look at relative frequency of different types of deviations because I thought that the quantity of observed data was too small to get a statistically significant result.

• However, I opted to look at relative frequency for vowels in Makino (2008). After doing so, a study of relative frequency for consonant substitutions was naturally expected and that is what I will do here.

It will take a lot more time to begin studying the remainder of the phonological system, i.e., prosodic aspects, because I have not transcribed them due to theoretical problems.

3. Methods in the present study

• The material I observed here is the same as that in my previous studies: the transcription of part of English Learners’ Speech Database (Minematsu et al. 2002). ELSD is a collection of about 68,000 recordings read aloud by 202 Japanese university students.

• Originally I planned to transcribe one sentence for each speaker and picked up 210 different sentences from the phonologically-balanced subset. I did so because I wanted to collect as many individual variations as posssible. Due to time constraints, however, I managed to transcribe only 74 sentences of the above 210.

• The principle of transcription was phonetic. I referred to the waveform, spectrogram and fundamental frequency traces made with Wavesurfer version 1.8.5 on Windows PC. Efforts were made to make as narrow segmental transcriptions as possible.

• As I noted above, I have not transcribed prosodic aspects. I am aware that this could be a problem because phonological processes are conditioned by those aspects in many cases.

• I compared the transcribed consonants with their English norms (as are shown in Jones, Roach, Hartman and Setter 2006) by looking at the transcriptions one by one.

• I tried to classify English consonants according to both the kinds of phonological processes concerned and their difficulties to Japanese speakers.

4. The consonant systems assumed in this

study• There are fourteen consonant phonemes in Japanese: /p, t, k, b, d, ɡ; s, z, h; m, n; j, w; r/.

• In addition, there are moraic consonants /H, J, Q, N/ which can constitute a mora alone (but not a syllable, like vowel phonemes). – Of these, /H/, the lengthening of its preceding vowel, and /J/, a

n [i] which occurs after other vowels, are vocalic, /Q/ (essentially the lengthening of the immediately following consonant) is consonantal and the moraic nasal /N/ can be both vocalic and consonantal according to the phonetic contexts.

• Japanese basically has no consonant clusters (except for cases where /j/ follows a consonant), and syllables can be closed by /Q, N/ only.

• All the Japanese consonants are always more or less palatalized when they occur before /i, j/. Some are so heavily palatalized that they change the place of articulation nearer to the hard palate (and in cases of /t/ and /d/, affricated).– In loanwords this heavy palatalization do not usually o

ccur for /t, d, s, z/ (because the speakers are aware of phonetic differences), but this does not apply to /n, h, N/.

• /s/ is usually pronounced [s, ɕ], but a significant number of younger people tend to pronounce it as [θ] these days.

• /z/ is phonologically the voiced counterpart of /s/, but it is usually fortified as an affricate [dz] (and [dʑ] before /i, j/) in phrase-initial positions and after [n] (an allophone of /N/ before [d]). This fortition can also occur intervocalically.

• The voiced plosive phonemes /b, d, ɡ/ and the voiced “affricate” /z/ tend to be weakened to fricatives [β, ð, ɣ] and [z, ʑ]. – In conservative pronunciations, /ɡ/ is weakened to a

nasal [ŋ]. This occurs after /N/ ( [ŋ] before /ɡ/) as well as intervocalically. As a result, the realization patterns of intervocalic or post-nasal /ɡ/ varies greatly with [ɡ ~ ɣ ~ ŋ] among individuals, or even intra-personally.

• The realization of /r/ varies greatly among individuals: [ɾ ~ɽ ~ ɖ ~ɭ ~ l] or even sometimes [ɹ] among people influenced by the learning of English.

• /j/ is an approximant [j] intervocalically. When it occurs after a consonant, its only function is to palatalize the consonant.

• /w/ has usually neutral (unrounded) lips. It could also be transcribed as [β ].ˠ

• /Q/ is the lengthening of the immediately following consonant. It occurs only before obstruents; in case of affricates, the stop part is lengthened.– Traditionally it occurred only

before voiceless obstruents, but in loanwords it also occurs before voiced obstruents, which tend to be devoiced when occurring word-finally.

• /N/ has the most various allophones. The only common denominator is that they are nasal(ized):– Before plosives, affricates and nasals, it is realized as a nasal ho

morganic to the following sound.– Before vowels (which means between vowels), it is realized as a

nasalized vowel. The quality of the vowel is influenced by the flanking vowels.

– Before /h/, it is realized as a nasalized vowel influenced by the preceding vowel and the one after /h/.

– Before /s/ [s, ɕ] it is realized as [ , ] respectively.ɨ � ĩ– Before /j, w/, it is realized as [ , ]ĩ ɯ�

• ELSD was recorded after training the speakers in so-called General American English pronunciation. Thus the speakers’ pronunciations were targeted to that norm. GA has the following twenty-four consonant phonemes: /p, t, k, b, d, ɡ; f, v, θ, ð, s, z, ʃ, ʒ, h; tʃ, dʒ; m, n, ŋ; j, w; l, r/.

• Allophonic variations are not as great as in Japanese; devoicing of (the first part of ) vowels and approximants /j, w, l, r/ after aspirated voiceless plosives, devoicing of voiced obstruents before a pause or a voiceless consonants, and regressive assimilation of place of /t, d/ and /n/ before another stops are some of the notable allophonic variations.

• In Makino (2008) I regarded postvocalic r’s as an off-glide of r-diphthongs (and noted that they were just dropped or realized as a vowel such as [ɐ]). So only pre- and intervocalic r’s are regarded as a consonant in this study.

5. Findings

• Total of 323 substitutions were found. The average substitution rate is 4.36 per sentence.

• I have not counted the total number of consonants in the data, so I cannot do any kind of statistical analysis.

• The kinds of substitutions I found in any large numbers will follow.

• This is the substitution involving the /r/-/l/ contrast, of which Japanese speakers are notoriously inadept.

• The total occurrence is 64.

Target Actual Category

ɹ ɾ 20 L vs R

l ɾ 11 L vs R

l ɹ 5 L vs R

-ɫ ɾɨ 5 L vs R

ɹ l 5 L vs R

Cɹ Cɰ 3 L vs R

ɫ ʊ 2 L vs R

tl tɹ 2 L vs R

ɫ ɭ 1 L vs R

ɫ ɚm 1 L vs R

ɫ ɨ 1 L vs R

ɫ ə 1 L vs R

ɫ ɔ 1 L vs R

ɫ ɨɾ 1 L vs R

ɫ ɨɫ 1 L vs R

ɹ ɜ 1 L vs R

tl təɹ 1 L vs R

uɫ u 1 L vs R

θɹ- θ 1 L vs R

• This is the substitution seemingly involving “lenition.”

• The types which do not occur in Japanese can be found here.

• The number of occurrence is 66.

Target Actual Category

ʤ ʑ 10 lenition

ɡ ɣ 8 lenition

b- β 6 lenition

-b- β 6 lenition

-l- 0 4 lenition

wu 0u 4 lenition

-b β 3 lenition

-ɹ- 0 3 lenition

d ð 2 lenition

ʤ z 2 lenition

-m m 2 lenition

pʰ ɸ 2 lenition

-ʧ ɕ 2 lenition

-w- 0 2 lenition

-b ɸ 1 lenition

dɹ ð 1 lenition

-dz z 1 lenition

-g x 1 lenition

ji 0i 1 lenition

-k -x 1 lenition

-kt xt 1 lenition

-m ə� 1 lenition

-ŋk -ŋ 1 lenition

-p ɸ 1 lenition

• This is the case where consonants are in some ways fortified.

• I found 32 of such cases.

Target Actual Category

ð dz 12 fortition

ð d 10 fortition

hw ɸ 4 fortition

v b 2 fortition

ð dð 1 fortition

-ʤd ʨt 1 fortition

-ŋ -ŋk 1 fortition

-z -d 1 fortition

• This is the case where quality of sibilants and non-sibilants are concerned.

• Non sibilants were made sibilants in 25 cases, while sibilants were made non-sibilants in 10 cases.

Target Actual Category

ð z12

sibilant fortition

z s 9 sibilant fortition

θ s 4 sibilant fortition

s θ 4 sibilant weakening

z ð 3 sibilant weakening

z θ 2 sibilant weakening

ts tθ 1 sibilant weakening

Target Actual Category

j- i- 2 syllabification

-d dɨ 5 vowel insertion

kɹ kɯɾ 3 vowel insertion

-z -zɨ 3 vowel insertion

dɹ dVɾ 2 vowel insertion

-s -sɨ 2 vowel insertion

•This is the type where vowels are inserted after final consonants or between consonants in consonant clusters.•The total number is 39.

-ʤ ʥi 1 vowel insertion

dɫ dɨɾ 1 vowel insertion

dɹ dɹ 1 vowel insertion

-ɡ ɣɨ 1 vowel insertion

-m -mə 1 vowel insertion

-m mɨ 1 vowel insertion

-nd -ndɨ 1 vowel insertion

nɫ nəɫ 1 vowel insertion

-ŋ -ŋɯ 1 vowel insertion

-p -pɨ 1 vowel insertion

p ##p˺ p ##pʰ 1 vowel insertion

p d˺ p dʰ 1 vowel insertion

-ps -pəs 1 vowel insertion

p t˺ p tʰ 1 vowel insertion

-st -stɔ 1 vowel insertion

stɹ stəɹ 1 vowel insertion

-t -tɨ 1 vowel insertion

-ʧ -ʨi 1 vowel insertion

t d˺ t dʰ 1 vowel insertion

tɹ tɨɾ 1 vowel insertion

ʧt ʨit 1 vowel insertion

-zd -zɨdɨ 1 vowel insertion

• This is the case where the wide variety of Japanese /N/ allophones are considered to be concerned.

Target Actual Category

Vn VV 15 N merger

-ns sV 9 N merger

-n -ɴ 4 N merger

-n- V 3 N merger

-nz zV 2 N merger

Vnd dV 2 N merger

ðm zz� 1 N merger

-n -m 1 N merger

Vnd nV 1 N merger

Vnð zV 1 N merger

• This is the cases where palatalization seems to be concerned.

• In some cases, consonants are de-palatalized. This might reflect the current tendency in Japanese.

Target Actual Category

ni ɲi 6 palatalization

si ɕi 2 palatalization

ti ʨi 2 palatalization

-d -ʥ 1 palatalization

di ʥi 1 palatalization

ili iʎi 1 palatalization

nj ɲj 1 palatalization

si ɕ 1 palatalization

ʤ z 2 palatalization hypercorrected

ʃs s 1 palatalization hypercorrected

ʒ z 1 palatalization hypercorrected

• This is the case where other types of transfer from Japanese are considered to be concerned.

Target Actual Category

f ɸ 6 Japanese transfer

v β 5 Japanese transfer

-ŋ(V) -ŋɡ(V) 3 Japanese transfer

fo ho 1 Japanese transfer

hu ɸu 1 Japanese transfer

v b 1 Japanese transfer

6. Discussions and Conclusion

• It is obvious that the fact that I have not counted the total number of consonant tokens undermines the result. We do not see what proportions are substituted in each phonemes.

• My categorization of the substitution process are more or less arbitrary.

• However, this study has revealed to some extent which English consonants are difficult for Japanese speakers and how they mispronounce them.

• It is necessary to build a larger corpus as soon as possible to get a more reliable picture.

Acknowledgment

• This work was partly supported by Grant-in-Aid for Scientific Research (B) No. 20300265 (project leader: Nobuaki Minematsu) from the Japan Society for the Promotion of Science.

ReferencesJones, D., P. Roach, J. Hartman and J. Setter. 2006. Cambridge English Pronou

ncing Dictionary, 17th ed. Cambridge: Cambridge University Press.Makino, T. 2005. Nihonjin no Tame no Eigo Onseigaku Lesson. [Lessons in Engl

ish Phonetics for Japanese Speakers.] Tokyo: Taishukan Publishing Company.

Makino, T. 2007. “A Corpus of Japanese Speaker's Pronunciation of American English: Preliminary Research.” Proceedings of Phonetics Teaching and Learning Conference 2007 (PTLC2007) CD-ROM. Also found at http://www.phon.ucl.ac.uk/ptlc/proceedings/ptlcpaper_02e.pdf

Makino, T. 2008. “Vowel Substitution Patterns in Japanese Speakers’ English.” Paper presented at The 1st Belgrade International Meeting of English Phoneticians (BIMEP 2008), University of Belgrade, Serbia. (The proceedings are forthcoming.)

Minematsu, N. et al. 2002. “English Speech Database Read by Japanese Learners for CALL System Development.” Proceedings of International Conference on Language Resources and Evaluation (LREC2002), pp.896-903.

Dzi kuj !!ę ęDzi kuj !!ę ę