2.2 Diverse views on word-loaning processes
2.2.3 The perception-production account
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
38
2.2.3 The perception-production account
Proponents of this account address that perception and production play equal roles in loanword adaptation. Phonetic details and L1 phonology altogether shape the output form of perception, i.e. the underlying representation, which in turn serves as the input to the production grammar.
Starting from Silverman (1992), where he investigates English loanwords in Cantonese, this model suggests that loanword adaptation involves perception and production, which are governed by two different grammars. A crucial point is that the loanword input does not enter L1 with the phonological representation of L2, but instead the L1 speaker hears a sequence of non-linguistic signals. His multi-scansion model consists of two separate, ordered levels, namely Perceptual Level (Scansion One) and Operative Level (Scansion Two). In the Perceptual Level, constrained by the native segment and prosodic inventory, the adapter perceives the input as a string of non-linguistic signals, which is parsed into segment-sized chunks. For example, a Cantonese speaker cannot tell apart English voiced and unaspiratd voiceless obstruents for the lack of such a contrast in Cantonese and hence perceive both as voiceless, as the native phonological system does not provide access to the contrast.
For another example, [s] is the only coronal fricative in Cantonese, and the adapter thus have trouble distinguishing English [s, %, z]. The resolution to this is to replace [%, z] with the acoustically closest [s], as shown below.
(15) English [s, %, z] as mapped to Cantonese [s] (Silverman 1992)
Input Perceptual Level
soda → /.s .ta./
show → /.sou./
size → /.say.si./
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
39
In this phase, meanwhile, phonetic salience plays a crucial role, and less salient segments are less likely to be perceived. For instance, a stop adjacent to a fricative is considered perceptually weaker and tends to be ignored, as in “shaft → [.s/f.]”.
Similar effects are also observable in our loanword data, as we will see in later chapters.
In the Operative Level, on the other hand, various phonological operations are triggered by the L1’s phonotactic constraints and the ‘raw’ material (the output of Scansion One) is adjusted to fit the native syllable structures. For example, in “[.f lm.]
film → [.fey.l/m.]”, since Cantonese has fricatives in the phoneme inventory and the [f] occurs in the onset, it is completely preserved in the surface form. When [f] fills the coda position, however, it may undergo feature change in this level, as in “[.l ft.]
lift → [.lip.]”, as Cantonese bans a [+continuant] consonant from being the coda, except for the glides [j, w].
Following Silverman (1992), Yip (1993) agrees that the perception of the L1 speaker is governed by the native phonotactics, but unlike Silverman, she adopts a constraint-based framework, an early version of OT, to account for the phonological processes in the Operative Level. The asymmetry of deletion and preservation of certain consonants in loanword adaptation is attributed to their degrees of acoustic salience. For example, [s] is saliently perceived from surrounding stops or vowels for its high-intensity noise and thus it tends to be preserved in the loaning process. This argument can well account for the three-way distinction that in her data, [s] is never lost because it is salient, liquids are sometimes lost because they are not very salient, and pre-consonantal stops are always lost because they are not detected at all.
Along similar lines, Kenstowicz (2003b) reviews Gbéto’s (1999) study of French, Portuguese, and English loanwords in Fon and singles out a few noteworthy problems which can be better accounted for by means of perception cues. Compared with Yip
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
40
(1993), in which only the Operative Level is analyzed within OT framework, he sketches two different rankings with the same constraint set for both “Perception Mapping” and “Production Mapping”. For example, in dealing with sC cluster, he proposes “Dep-V, *stop/obstruent_# >> Max-C” for perception and “Max-C,
*stop/obstruent_# >> Dep-V” for production. This approach well accounts for the Fon adaptation of word-final obstruent-stop clusters from French, as sketched in (16).
(16) A bipartite OT-based model (Kenstowicz 2003b, revised) a. Perception Mapping
[.post.] Dep-V *stop/obstruent_# Max-C
a. ☞/.pos./ *
b. /.post./ *W L
c. /.pos.tu./ *W L
d. /.po.sut./ *W L
e. /.po.su.tu./ **W L
b. Production Mapping
/.pos./ Max-C *stop/obstruent_# Dep-V
a. ☞[.po.su.] *
b. [.pos.] *W L
c. [.po.] *W L
In Perception Mapping, the acoustic input [.post.] is interpreted as the underlying form /pos/ due to the dominance of Dep-V over Max-C. The dominance is however reversed in Production Mapping and the output [.po.su.] results.
A more recent work that holds the perception-production view on word-loaning processes is Boersma and Hamann (2009), which shows that loanword adaptation can be understood entirely in terms of the native perception and production mechanisms, and like the previous studies along this line, each is handled by the interactions of OT constraints. Unlike those authors, however, their rationale is based on a finer-grained
‧
bidirectional model that distinguishes three levels of representation, namely the underlying form, the phonological surface form, and the auditory-phonetic form. The model is shown in (17).
(17) A single model for L1 processing and loanword adaptation (Boersma 1998, 2000, 2007ab; Boersma and Hamann 2009) adaptation, each of the four processes, i.e. perception, word recognition, phonological production, and phonetic implementation, is described in terms of an interaction between two of the three types of constraints: FAITH(FULNESS), STRUCT(URAL),and CUE. What is crucial within this mechanism is that structural constraints play a role in both production and comprehension.
A fact that is often commented in Korean is that the process of vowel insertion is more commonly observed in loanword adaptation than in the native phonology, in which typical phonological alternations include neutralization, assimilation, and deletion. From the OT perspective, then, the faithfulness constraint Dep-V should rank high in L1 phonology. In contrast, a predominant process found in Korean
‧
that are the same for speakers and listeners. In what follows, a couple of tableaux are extracted from the paper to illustrate their primary postulation. The focus, however, will be laid on the processes that gain more widespread recognition in literature:perception (bottom left) and phonological production (upper right).
(18) Perception grammar for both native words and loanwords in Korean (Boersma and Hamman 2009, revised)
(a) Korean perception of the native word [.m 1k2_.]
[.m 1k2_.] *[burst]
(b) Korean perception of the English word deck [._d 56k2._kh.] *[burst]
As can be seen, the main ingredients that constitute Korean perception grammar are cue constraints. While the same constraint ranking leads to vowel insertion in the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
43
loanword in (18b), it does not result in the same consequence for the native word in (18a). The crucial reason for this is that the L1 final plosive in the auditory input to the perception grammar is unreleased (in the narrow auditory transcriptions, “k2”
stands for the formant transition from a vowel to into the velar stop, and the underscore “_” stands for the silence that occurs in plosives), whereas the final plosive in the loanword input is released (“k” stands for the velar plosive release burst and “ ” stands for the moderately strong aspiration noise). It is this auditory distinction in input that gives rise to the different consequences in output. Specifically, in (18b), the high-ranking cue constraint *[burst]/C(.)/, saying that “an auditory release burst should not be perceived as a phonological consonant in coda,” favors the winner over Candidates (b) and (c), both of which interpret the syllable-final released plosive as the coda. In (18a), however, the same cue constraint *[burst]/C(.)/ is vacuously satisfied by all candidates since the plosive coda in the input is unreleased.
Candidate (a) is then preferred by all constraints and wins, while the most competitive Candidate (e) fails to win since it is disfavored by the bottomed *[ ]/!/ for an inserted vowel.
(19) Production grammar for both native words and loanwords in Korean (Boersma and Hamman 2009, revised)
(a) Korean production of the native word |.p t .|
|.p t .| */+asp./ Dep-V Max-C Ident(asp) */C./
a. ☞/.p t./ * *
b. /.p t ./ *W L *
c. /.p .t !/ *W L L
d. /.p ./ *W L L
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
44
(b) Korean phonological production of the English loanword |.t .k !.|
|.t .k !.| */+asp./ Dep-V Max-C Ident(asp) */C./
a. ☞/.t .k !./
b. /.t .k!./ *W
c. /.t k ./ *W *W
d. /.t k./ *W *W
Likewise, both native words and loanwords are processed by a single production grammar. In (19a), for the input |.p t .|, the best candidate is the phonetically second closest /.p t./ under the evaluation of faithfulness constraints, as the most faithful Candidate (b) is not preferred by the undominated structural */+aspirated./ In (19b), the most faithful Candidate (a) is favored by all constraints over the harmonically bounded competitors.
Still on the perception-production basis, Peperkamp and Dupoux (2002) draw a very similar model of loanword adaptation to Boersma and Hamann’s, but earlier than theirs. On the two sides of perception and production, they distinguish phonetic/phonological decoding and phonetic/phonological encoding, respectively, performing inverse functions, as shown below.
‧
(20) Peperkamp and Dupoux’s (2002) loanword adaptation model
Perception Production
As indicated in this diagram, there are two loci for loanword adaptation: phonetic assimilation takes place in phonetic decoding and phonological regularization in phonological encoding. Boersma and Hamann’s (2009) examples well illustrate their own model as well as this version. This is shown below.
(21) Korean adaptation of the English words deck and tag (Boersma and Hamman 2009)
tag: [._tha-67.(2 3 (.] → /.t æ.k!./ → |.t æ.k!.| →/.t æ.k!./ → [._th7 -17(2 3(!.]
deck: [._d 56k2._kh.] → /.t .k !./ → |.t .k !.| → /.t .k !./ →[._d8e91k2._ kh!.]
The bipartite processing model that is sketched for this research is shown in (8), reduplicated in (22) for convenience, is in essence parallel to the above two authors’
loanword adaptation models in (17) and (20).
phonological
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
46
(22) A bipartite processing model for English loanwords in TM adaptation (also (8))
Perception Production
/underlying representation/ /underlying representation/
perception grammar (SPG) production grammar
[auditory form] [articulatory form]
L2 lexicon L1 lexicon
The difference lies in two aspects. Firstly, more attention in this dissertation is paid to modeling Boersma’s processing of perception and phonological production, which are equivalent to Peperkamp and Dupoux’s (2002) phonetic decoding and phonological encoding, i.e. the two loci for loanword adaptation governed by the perception grammar and production grammar, respectively, in our model. We do not go deep into, in Boersma’s terms, the “lexical storage” (e.g. “/.t æ.k!./ → |.t æ.k!.|”) and “phonetic implementation” (e.g. “/.t æ.k!./ → [._th7 -17.(2 3(!.]”) processes, as this dissertation lays the most focus on sketching a formal analysis for the phase of perception (in Boersma’s model)/phonetic decoding (in Peperkamp and Dupoux’s model). Secondly, while the processing of phonological encoding in (17) does not incur major segmental changes (e.g. “|.t æ.k!.| →/.t æ.k!./”), the L1 production grammar in our model is very likely to make a difference since, as shown in (22), the L1 adapter’s semantic preferences exert a great influence on his or her choice of L1 characters.
While most studies arguing for the perception-production account are modeled in STRUCT
CUE
STRUCT
FAITH
articulatory constraints semantic preferences
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
47
OT, Kim (2008, 2009) adopts a non-OT approach to Korean adaptation of English affricates and fricatives, as shown below.
(23) A feature-driven model of loanword adaptation (Kim 2008, 2009)
L2 acoustic output (= L1 input)
L1 output (surface representations)
In her “feature-driven” model, the loanword adaptation is divided into three major levels, where she assumes that acoustic parameters and cues are extracted in the first phase of L1 perception (23a) and meanwhile mapped into L1 distinctive features and syllable structure. In the mental lexicon, the filtered L2 information is represented as a sequence of distinctive feature bundles and lexicalized in the long-term memory.
Finally the underlying representation is subject to L1 phonology and then surfaces as the L1 output. Kim gives an example for the phase of L1 perception in (23a), as
a. L1 Perception
i. extraction of acoustic parameters and cues
ii. L1 Grammar
(mapping into features and syllable structure)
b. L1 Lexical representations (mental lexicon)
c. L1 Phonology
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
48
briefly elaborated below.
English palato-alveolar [%] is interpreted as a sequence of [s] and [j] before a non-front vowel, as in “show → [.sjo.]”, otherwise as a sequence of [s] and [w] before a front vowel, as in “Shell → [.sw l.]”. In the initial phase of L1 perception, she assumes that two acoustic aspects are extracted. The first is the high acoustic intensity of the onset [%], which is then parsed for cues to the features [+continuant, +strident]
within L1 grammar in the second phase. The second is the frequency energy around 2000-3500 Hz and locus, which are parsed for cues to the features [+anterior, coronal]
and one of the glides /j/ and /w/ by L1 grammar in the second phase too. Both of the extractions are based on the L1 distinctive feature and syllable structure. A noteworthy explanation for the [sj]~[sw] patterns is that the sequence of [j] plus a front vowel violates Obligatory Contour Principle (OCP), and hence [w] replaces [j] when preceding a front vowel and [j] remains elsewhere.
In summary, it is generally held in the perception-production account that perception and production are equally influential in the transformation of an L2 input into the L1 output form. This account features a viewpoint that differentiates it from the others: perception is phonological, too. Given the evidence and arguments provided in both the production-based and perception-based accounts, as well as the systematic differences between the two languages under investigation, we are led to recruit a perception-production stance on the central issues in this dissertation, as a pure production-based or a perception-based account is more or less rendered biased with respect to the language background of the adapters. The perception-production account seems to be the perspective from which we are able to see the whole truth, preventing us from being a blind man touching the elephant.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
49