The production-based account

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

27

coined in the L1 lexicon. However, most, if not all, of the surface representations have undergone adjustments, as the L1 speaker cannot speak the L2 word well. There are two logically possible “loci” where the adaptation takes place. The first is that the acoustic signal that is novel to the speaker is faithfully mapped into the abstract featural representation. This featural representation is then subject to the modification that is performed by the production grammar. The second is that adjustment is implemented as early as in perception under the government of L1 phonology and phonetic cues of the L2 input.

The second scenario seems to be pervasively approved across authors along this line. Considering English is not as widely spoken in Taiwan as it is in other Chinese communities such as Hong Kong and Singapore, this dissertation adopts this scenario too. Previous studies on loanword phonology have centered on debates between the two potential “loci” where adaptation of a given input string takes place for the native speaker. In the forthcoming subsection, representative postulations on each side will be elaborated respectively.

2.2.1 The production-based account

Authors (Calabrese 1988, 1995; Connelly 1994; Davidson and Noyer 1997; Ito and Mester 1995; Paradis 1996; Paradis and LaCharité 1997; Paradis and Prunet 2000;

Jacobs and Gussenhoven 2000; LaCharité and Paradis 2005; Paradis and Tremblay 2009) holding this view assume that word borrowing occurs in the first scenario. The role of perception in the storage process is either absent (Paradis and LaCharité 1997) or restricted to a limited number of extragrammatical adaptations to the segmental or tonal inventories of L1 (Silverman 1992, Yip 1993). They assume that the adapter is proficiently bilingual and thus the underlying representation contains the non-native segments. Adaptations of these non-native segments are performed in production so as

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

28

to avoid producing marked or ill-formed segments or strings of segments.

In the proposal of TCRS (Theory of Constraints and Repair Strategies, Paradis 1996; Paradis and LaCharité 1997), it is the expense of a repair strategy that determines the retention or deletion of a segment in loanword adaptation. The more costly it is, the less likely it is adopted. A segmental adjustment cannot go beyond the

“threshold” of modification. Under the universal Preservation Principle (Paradis and LaCharité 1997, LaCharité and Paradis 2005) in loanword adaptation, once the cost of preservation of a segment exceeds the limits of the threshold principle, deletion of it occurs. All the repairs take place in the phase of production. For example, in the case where the L2 French word [.av.ka.] (avocat) is adapted into L1 Fula [.aw.ka.], it takes only one step to change the illicit segment [v] into [w] by inserting the feature [+sonorant]. However, in another case, French [.vwa.ja.] is adapted into Fula [.a.wa.jas.], in preference to *[.wu.wa.jas.], since preservation of the latter involves a three-step repair: sequentially, insertion of a nucleus node, filling of a vowel in the inserted nucleus position, and finally changing ill-formed [v] into native [w].

TCRS may soon run into the problem of explanatory inadequacy as more cross-linguistic loanword data are being investigated. According to TCRS, vowel insertion is always more laborious than consonant deletion, which follows that deletion of a consonant should always outnumber vowel insertion cross-linguistically.

However, it is at least not true in our investigation of English loanwords in TM, where, all else being equal, consonants are retained and syllabified through an epenthetic vowel in more cases, except for a liquid coda following a [+back] vowel. Other similar oppositional stances include Brasington (1997), where it is observed that in Marshallese adaptations from English, resolution to a consonant cluster relies on syllable position and cluster type. For example, vowel insertion is found in more cases when the cluster lies in onset, while consonant deletion is more likely to happen in

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

29

coda position. Likewise, Ulrich (1997) tests the TCRS model with Lama adaptations from French and English and concludes that vowel insertion serves as the major strategy even though it involves more “steps” to repair.

Perhaps the most representative trait of a production-based analysis is that the input to adaptation is a morphophonemic representation of the L2 word that abstracts away from the detailed phonetic realization. For example, in LaCharité and Paradis (2005), it is demonstrated that in French loanwords from English, the English lax high vowels [ ] and [ ] are mapped to French [i] and [u], despite the fact that French mid vowels [e] and [o] are acoustically more similar matches (closer formant frequencies).

This is because loanwords under this rationale are adapted in terms of distinctive features, and then the configuration [+high, –ATR] of English [ ] and [ ] can be repaired into the configuration [+high, +ATR] of French high vowels [i] and [u]. In this view, perception is therefore extragrammatical and plays a passive role. Another explicit example in their study is the mapping of stop voicing from English to Spanish.

The VOT of stops in the two languages are given below.

(9) VOT correlates of stops in Spanish versus English (LaCharité and Paradis 2005) Phonetic Implementation

Phonological value Spanish English

Voiced /b, d, (/ –VOT (-40-0 msecs.) +VOT (0-30 msecs.) Voiceless /p, t, k/ +VOT (0-30 msecs.) +VOT (> 50 msecs.)

As can be seen, if perception of the phonetic form is solely responsible to adaptation, a perfect mapping would be for English voiced stops to be interpreted as Spanish voiceless stops, as both are produced with VOT ranging between zero and thirty milliseconds. The truth is, however, that English voiced stops remain as voiced in Spanish. This suggests that loanword adaptation is based on the L2, rather than L1,

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

30

referenced perception of L2 phoneme category, and phonetic approximation plays a limited role.

Production-based accounts within the OT framework propose rankings of faithfulness constraints to ensure the storage of true detailed acoustic information. For example, Kang (2003) attributes the Korean surface form [.tsi.p !.] from English jeep to the ranking “Max[release] >> Dep-V, ensuring retention of the release feature. A major problem that arises from both LaCharité and Paradis (2005) and Kang (2003) is that, generally, it is recognized that underlying forms should be as economical as possible without too much realization of phonetic details, such as height, [±ATR], and release. What constitutes a secondary problem, moreover, is that although Kang (2003) explicitly states that vowel insertion occurs in perception, her proposed underlying forms contain no inserted vowel. This is contradictory to the statements that lexical representations have been filtered by the perception process, as put forth by psychological models of speech perception (cited in Calabrese (2009)).

Still within OT, Itô and Mester (1995) posit the core-periphery structure in Japanese lexicon, which is organized into separate core-periphery strata, i.e. native vocabulary, established loans, foreign vocabulary, and unassimilated foreign vocabulary. Each stratum is equipped with its own constraint-based phonology, which follows that a foreign sound may be “tolerated” in a more peripheral partition while blocked in a partition that is closer to the core. Similarly, in the investigation of Huave (a language isolate spoken in southeastern Oaxaca Sate, Mexico) adaptations from Spanish, Davidson and Noyer (1997) propose a loanword-specific constraint Match, requiring that stress between input and output should match, to interact with other faithfulness constraints. Moreover, Jacobs and Gussenhoven (2000) even dispense with Silverman (1992) and Yip’s (1993) Perceptual Level and make the claim that the input to the L1 phonological grammar is a universal defined and fully specified

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

31

phonological representation. All the abovementioned authors contend that loanword adaptation is done by language-specific OT production grammars. If this is the case, the fundamental problem remains—how does the L1 speaker learn the loanword production grammar that is specific to a particular language, provided that such foreign input strings are impoverished in L1?

To summarize, authors along this line maintain that phonologically-informed adaptations can only be made in production, and that perception is nothing but a passive process. One premise that they share in common is that the adapter is a proficient bilingual, and correct perception of the L2 morphophonemic representation becomes possible. A possibility that should not be excluded in our investigation, however, is that adaptation of a certain number of loanwords may be done by a monolingual or a bilingual with lower L2 proficiency.