Perceptual Effects on English Consonant Interpretation
5.4 Vowel insertion as perceptual hallucination
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
177
mapping of aspiration and voicing is significant in statistics (χ2 = 255.9, df = 4, p
< .0001). It is then assumed that a voiceless unaspirated stop in coda behaves more similarly to its aspirated counterpart in acoustics (VOT), as it may occasionally be released with aspiration when the word is uttered in an emphatic way. When an English aspirated stop follows [s] in the onset and undergoes the process of deaspiration, however, its VOT is closer to that of a voiced stop (around zero ms) In our terms, for the voiceless unaspirated stop that ends a syllable, the absence of auditory voicing weighs heavier than the absence of auditory aspiration noise. For the ones that follow [s], conversely, the absence of auditory aspiration noise weighs more than the absence of the auditory voicing. This generalization may explain the asymmetric mapping of voicing and aspiration between the two languages.
5.4 Vowel insertion as perceptual hallucination
The resolution of vowel insertion to L2 clusters and codas that are illicit to the L1 phonological grammar has been widely attested in literature on loanword phonology, as exemplified cross-linguistically in (119) below (Paradis and LaCharité 1997, Kenstowicz and Sohn 2001, Pollard 2008, Repetti 2009):
(119) Vowel insertion in loanwords across languages
L2 Source L1 Adaptation Gloss Locus
a. French [.klas.] Fula [.ka.la s.] ‘flag’ first onset b. English [.k l z.] Inuktitut [.ka.la.si.] ‘Claus’ first onset c. English [.b s.] Korean [.po.s .] ‘boss’ simplex coda d. English [.b uz.] American-Italian [.b u.za.] ‘bruise’ simplex coda
In the native language that disallows consonant clusters, an English source word with a complex onset is prone to vowel insertion for the first onset of the cluster (see Kang 2003 for a general review that mentions languages that favor consonant deletion
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
178
as the resolution), as illustrated in (119ab). In addition, for an L1 that prohibits any consonant or a certain category of consonants in the coda position, a common repair is to insert a vowel, leading the coda to act as an onset and forming an additional syllable, as shown in (119cd).
If it is true that perception of a novel sound combination is a process that involves interpretation and computation of the foreign sequence (Calabrese 2009), this dissertation adopts the view that an epenthetic vowel can be introduced in the perceptual process, contrary to the widely approved argument that vowel insertion is accomplished in production grammar (Silverman 1992, Yip 1993, 2006, Kenstowicz 2001, among others). As cited in Boersma and Hamann (2009), Polivanov (1931) proposes that Japanese listeners perceive Russian words[._tak _k.] ‘so’ and [. drama.]
‘drama’ (narrow phonetic transcriptions, where “_” refers to the silence that occurs in the articulation of plosives, “k ” to the formant transition from a vowel into the velar stop, and “ 3” to the vocalic murmur of a voiced plosive) as /.ta.ku./ and /.do.Aa.ma./ in the underlying configuration, respectively. He attributes such percepts to Japanese structural constraints that forbid coda consonants and consonant clusters.
The assertion gains its support in the experiment conducted by Dupoux et al.
(1999, introduced in more detail in Chapter 2), which shows that Japanese listeners fail to distinguish between the sound sequences [.eb.zo.] and [.e.bu.zo.], confirming that Japanese speakers have perceptual vowel insertion. Furthermore, Broselow (2004) reports her approval of the proposal by Schütz (1978) that Fiji speakers hear English whiskey as /.wi.si.ki./.
Cross-linguistically, a vowel is not necessarily inserted after the problematic consonant, but it may occasionally be inserted ahead of a consonant to fix an onset cluster, though cases as such are relatively fewer in the literature. For example, for the English string of “voiceless sibilant-stop”, a vowel tends to be inserted before the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
179
cluster in Hindi, as in “[.skul.] school → [. s.kul.]”, where the sibilant changes from the first onset of the syllable in English to the coda of an additional syllable in Hindi.
Similarly, in Turkish loanwords from English, the word skeleton ([.sk .l .t n.]) is borrowed as “[.is.ke.let.]”. Fleischhacker (2001) attributes it to perceptual similarity, as experimental studies show that an [S1T2] (where “S” stands for the sibilant [s] and
“T” for a stop) cluster is judged to be more similar to [.V1S1.T2] than [.S1V1.T2]. The result may run counter to the more frequently attested [.S1V1.T2] resolution to the [S1T2] cluster, however. Either way, a conclusion can be drawn that the L1 adapter tends to resort to perceptual closeness in dealing with foreign sound strings.
Turning back to our research, along the same lines, we assume that TM adapters’
insertion of vowels for ill-formed structures in both onset and coda is majorly accomplished in the phase of perception. Examples of vowel insertion are given in (120).
(120) Examples of adaptations through vowel insertion
English TM Locus
a. [.sta n.we .] Steinway → [. .t an.wei.] 史坦威 1st onset
b. [.st a s.] Strauss → [. .t .l u.s!.] 史特勞斯 1st and 2nd onset c. [.f l.] Phil → [.fei. .] 菲爾 simplex coda d. [.f lp.] Phelp → [.fei. .p u.] 菲爾普 1st and 2nd codas e. [.f lps.] Phelps → [.fei. .p u.s .] 菲爾普斯 1st, 2nd, and 3rd codas
TM strictly forbids consonant clusters in any position, and hence vowel insertion may be triggered for the first consonant within an English [C1C2(V)] onset cluster, leading to the [.C1V1.C2(V)] structure in TM, as shown in (120a). The same goes for an English [C1C2C3(V)] onset cluster, which may be adjusted by inserting vowels after the first and second consonants separately, leading to the structure of [.C1V1.C2V2.C3(V)] in TM, as in (120b). By the same token, TM does not allow coda
‧
consonants other than the nasals [n, ] and retroflex [ ] in the syllable of [. .], which follows that any codas against the requirements, simplex or complex, should be fixed.
In cases where everything is preserved, vowel insertion occurs for an English illicit the adaptations of the same English source words.
(121) Vowel insertion in Cantonese and TM
English28 Cantonese TM Locus
a. [.sm t.] smart [.si.ma7t.] 士嘜 [. .m u.] 時髦 1st onset b. [.d s.k o .] disco [.tek.si7.kow.] 的士高 [.ti.s .k .] 迪斯可 simplex coda c. [.le s.] lace [.le.si.] 籬士 [.lei.s .] 蕾絲 simplex coda
As the two borrowing languages treat consonant clusters as ill-formed, for the onset cluster in the source [.sm t.] both languages epenthesize a vowel in the back of the first onset [s] word-initially, as indicated in (121a). The same goes for (121bc), where a vowel is inserted by both speakers for the simplex [s]-coda word-medially and word-finally, which is an illicit coda consonant in both languages.
27 Note that a liquid is not always syllabified by inserting a vowel ahead. For instance, English [.f lm.]
(film) is adapted as [.fe.l/m.] by Cantonese speakers (Silverman 1992), where the epenthetic [/]
appears after the liquid.
28 Due to the long British colonial era until 1997, when China resumed sovereignty, chances are that the English inputs to Cantonese speakers may carry British accent, rather than the phonetic forms that are transcribed in our dissertation, which is mainly American English. For instance, the phonetic form of the English input “smart” to Cantonese should be [.sm 7t]. However, the
“R-dropping” phenomenon in British English does not constitute a problem in our investigation on vowel insertion for obstruents.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
181
As has been discussed in the previous subsection, however, Cantonese [p, t, k]
can act as both the initial and the final consonants of the syllable. Because of this, English source words with [p, t, k] codas do not undergo vowel insertion in Cantonese adaptations, but words with the same structure may be syllabified in this way in TM.
Refer to (122) for examples.
(122) Vowel insertion for stop codas in TM but not in Cantonese
Source Cantonese TM
a. [.% p.] sharp/Sharp29 [.sa7p.] [.$ja.p u.] 夏普 b. [.k w t.] quart [.k w/t.] 骨 [.k wa.t .] 夸特 c. [.d#æk.] Jack [.tsek.] 積 [.t$ja.k .] 傑克
With no structural constraints that forbid [p, t, k] as the coda consonants, an English stop coda is thus faithfully mapped onto the same position by the Cantonese adapter, but it is syllabified through vowel insertion with the stop being the onset by the TM adapter, under the guidance of the constraints that disallow any obstruents to be the codas.
In our stance, we assume that L1 structural constraints that regulate the syllable- marginal structures function in the perceptual level, and thereby vowel insertion is accomplished early in this phase. The scenario of first perceiving the English word [.sta n.we .] (Steinway) by the TM adapter thus proceeds like this: When presented with the illicit sound string [.sta n.we .] in the phonological processor, the TM listener cannot construct a mental representation of it because of the active structural constraints in his/her perceptual grammar. The listener thereby resorts to epenthesis to fix the problem, as he/she perceives the initial onset consonant [s] clearly for its friction noise. The presence of the illusory vowel [!] in perception then occurs, since
29 English “sharp” is borrowed into Cantonese with its literal meaning “able to cut”, but in TM it is the registered brand name of a Japanese company that produces household appliances.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
182
the resultant [.s!.] is the closest approximation of [s]. This yields the mental representation of, say, /.s!.tan.wei./30, which is stored in the short-term memory in the brain, which serves as the input to the production grammar later.