Abstract and Keywords
This chapter looks at processes operating in loanword adaptation. Starting out with a typology of adaptations, looking at what we know empirically about how loanwords are adapted to the phonological system of the borrowing language, this chapter then examines the central controversy surrounding loanword adaptation, whether adaptations are based on phonological equivalence or on phonetic/perceptual similarity. Data from a range of languages show that no explanation alone is sufficient. I therefore present a novel proposal, which first of all redefines the idea of phonological adaptations as equivalences based on contrast and opposition, and which secondly adds a sociolinguistic perspective to the psycholinguistic perspective that is standardly assumed. Rather than seeing individual speakers as the main locus of adaptation, I argue that loanwords are adapted by the speech community, with the consequence that there will be variation and effects of conventionalization, plus the possibility of different outcomes depending on the contact situation, with some situations inviting perceptual adaptations more than others.
Loanword adaptation has seen renewed interest in the phonological literature, accompanied by a fierce theoretical debate about how words are adapted, whether adaptations are phonological in nature, or based on phonetic and perceptual similarity; this has a number of connections to diachronic phonology and the addition of words to a language is in itself a type of change, although the adaptations of individual words themselves are not canonical cases of diachronic phonological change. This chapter reviews both the empirical facts (section 37.2) and the theoretical debate (section 37.3), proposing a way in which the opposing positions may be unified, suggesting that the two major factors in adaptation are similarity and phonological contrast and arguing that the role of contrast and opposition has received too little attention so far. The relative importance of these two factors also seems to be mediated by sociolinguistic factors, though. I suggest a sociolinguistic turn in section 37.4: we can only understand how words are adapted when we understand the sociolinguistic setting in which they are adapted. Section 37.5 concludes.
37.2 Phonological Adaptations: An Overview
When words are borrowed, they are frequently (but see section 37.4) adapted to the phonological system of the borrowing language, both segmentally, identifying native phoneme categories to express the sounds of the original form, and suprasegmentally, making the borrowings conform to native phonotactics and syllable structure constraints, and adapting them to the native stress or tone system. This section looks at what we know about how words are adapted at different levels of phonological description (p. 645) before moving on to theories of loanword adaptation, discussed and compared in section 37.3.
37.2.1 Segmental Adaptations
When we think about how the sounds of an L2 can be expressed in an L1 (see also Eckman & Iverson, this volume, for a general overview of L2 acquisition and phonological change), two possibilities come to mind. First, borrowers may take the incoming raw phonetic signal and find an underlying representation in their L1 whose output is as phonetically similar to the original form as possible. Alternatively, they may analyse the L2 in terms of phonemes and underlying contrasts and map these more abstract categories onto L1 categories. This distinction is at the heart of the difference between perceptual or phonetic and phonological approaches, discussed in detail in section 37.3. It is useful to bear the two approaches in mind when looking at the main patterns of segmental adaptations discussed here. For ease of exposition, let us begin with the hypothesis that adaptations are phonological in nature, mapping an L2 phoneme inventory onto an L1 phoneme inventory, and see if this assumption is sufficient.
We may naively think of adaptation as a process that occurs only when the borrowing language does not have a phoneme which is found in the donor language. However, all borrowings involve adaptations, since the borrower is faced with the non-trivial task of establishing what L1 phoneme should count as equivalent to a specific L2 phoneme. This task is not trivial because sounds, which we may readily class as phonologically equivalent, for example by using the same IPA symbol in transcription, or by assigning them the same distinctive feature values, usually still differ in their exact phonetic realization. For example, both German and Norwegian borrowed English beat (as in beat music, beatnik), and the word may be transcribed /biːt/ in all three languages, suggesting that no segmental adaptations took place. There are, however, subtle differences in how the three component segments are produced, with variable VOT for initial /b/, somewhat different realizations of long /iː/, which is diphthongized and variably fronted in English (e.g. as [ij, ɪj]), a stable monophthong [iː] in German and realized with a centralized offglide in Norwegian [iə]. Final /t/ is often unreleased in English, may be preglottalized, an ejective, or a glottal stop, but can be aspirated in German and Norwegian. Hence, the task of establishing phonological equivalence is not straightforward, even less so where the differences between two potentially equivalent phonemes are phonetically less subtle.
A notorious case of establishing equivalence where a phonetically similar sound is not available is that of English interdental fricatives /θ, ð/ (see e.g. Lombardi 2003 and Blevins 2006c for an overview), which do not exist in many languages. They are adapted as /t, d/ in some languages, as /s, z/ in others. Thus, a phoneme category found in the donor language but not in the borrowing language is mapped onto an existing category in the borrowing language, frequently involving mergers—here mergers with alveolar stops of fricatives. Mergers are unavoidable if the L1 has fewer contrasts than the L2. A well-known rather extreme example is that of Hawaiian, which has only seven (p. 646) consonants, /p, k, ʔ, l, w, n, h/. Many contrasts found in English are therefore obliterated. For instance, all non-labial obstruents (/t, d, k, g, s, z, … /) map onto /k/ (with qualifications, see, for example, Adler 2006 for detail).
Mergers of L2 distinctions are one way in which differences between L1 and L2 manifest themselves in the adaptation process. Alternatively, there can be a shift in phoneme category, when the borrowing language does not have a category and one perceived as similar is chosen instead (see section 37.3 for a discussion of the notion of similarity). For example, in Burmese and Hindi voiceless aspirated stops are used to adapt an English fricative. In Burmese, /ph/ is used for English /f/ (Chang 2012); in Hindi, /th, dh/ are used for L2 /θ, ð/ (Hock 1991: 393f.). These realizations do not involve a merger, however; English stops are realized by separate phonemes. In Burmese, English /p/ is realized as /p/ (even though /p/ is aspirated [ph] in many positions in English), and /b/ is realized as /b/. In Hindi, English /t, d/ are adapted as retroflexes /ʈ, ɖ/, thus avoiding merger with /θ, ð/. (1) gives an overview of the Burmese adaptations. The three-way phonemic contrast of English is preserved, but there is a case of a phonological non-equivalence when the fricative /f/ maps onto aspirated /ph/. The mappings of /b, p/ > /b, p/ could then be seen as mappings between equivalent categories, say between [labial] stops. (This simplifies matters, given the question whether the underlying contrast in English is one of voicing or aspiration, see e.g. Honeybone 2005 on ‘laryngeal realism’; see also section 37.3.2.)
(1) illustrates phonological adaptations, matching up phoneme equivalences in the two languages, disregarding phonetic detail in the L2 (here: aspiration). Is it the case, then, that allophonic, non-contrastive information is uniformly disregarded in the adaptation process? One could imagine a situation in which the borrowing language has a phonemic contrast which is only allophonic in the donor language. Will the L2 phoneme then be consistently matched with an L1 phoneme (phonological adaptation), or will the allophonic variation be reflected in the choice of L1 phoneme? The evidence for this is mixed. It seems that in some languages, allophonic information is considered in the adaptation process, while in other languages, segmental adaptations are largely phonemic (see also section 37.3.4).
We saw that in Burmese allophonic aspiration does not influence the adaptation pattern. English /p/ is adapted as /p/ in all positions, keeping it distinct from /f/, which is adapted as /ph/. We find a similar pattern in Korean (e.g. Lee 2003, Kenstowicz 2005) where English /b/ is uniformly adapted as lax /p/, and English /p/ is adapted as aspirated /ph/, even in contexts where it is never aspirated, as in /sp/ clusters. Thus, spy > [sɨphai], with aspiration. As Kenstowicz (2005) points out, it is not the case that Koreans do not (p. 647) pay attention to phonetic detail in laryngeal contrasts in borrowings: French /p/, which is unaspirated, is generally adapted as tense /p’/, as in Paris > [p’ari]. However, allophonic variation does play a role in Thai adaptations of English words (Kenstowicz & Suchato 2006). There is aspiration in pin > [phin] but not in spare > [səpee], unlike in Korean. We will suggest reasons for this variant behaviour later.
In their discussion of French loans in Moroccan Arabic, Kenstowicz & Louriz (2009) show that allophonic information can be taken into account in the adaptation process in a different way as well, not by mapping an allophonic distinction in the L2 onto a phoneme distinction in the L1, but by utilizing an L1 allophonic distinction to capture an L2 phoneme contrast. Like many Arabic dialects, Moroccan has only three vowels /i, a, u/. However, these vowels have lowered/retracted allophones [e, o, ɑ] when in the same syllable as an emphatic consonant. Thus, French words containing mid vowels or /ɑ/ are adapted with emphatic consonants in order to faithfully reproduce the mid vowels. A contrast which is not found in the donor language (emphasis, as pharyngealization or uvularization), is therefore used to preserve a vocalic contrast in the donor language.
Finally, borrowing may aid in the phonemicization of an allophonic contrast. Picard & Nicol (1982) cite an example from Québec French, where a long vowel diphthongizes allophonically in a stressed closed syllable. This process is not found in English loanwords containing a diphthong, however. The diphthong is maintained across phonological contexts, and the allophonic alternation is therefore suspended in loanwords. Consequently, minimal pairs may emerge. (2) provides some examples, from the Québec French alternation (a) and the blocking of this alternation in loans (b). Note the homophonous pair fête = fight with a derived contrast fêter = fighter.
In sum, we can understand segmental adaptations to a large degree as the mapping of an L2 phoneme system on the categories of the L1. However, allophonic information may interfere in this process. We will discuss the theoretical relevance of this in section 37.3.
37.2.2 Suprasegmental Adaptations
There are two main types of suprasegmental adaptations, syllable structure adaptations (and more generally phonotactic adaptations) and adaptations regarding stress and tone. This section summarizes the main findings.
(p. 648) 220.127.116.11 Syllable Structure and Phonotactics
Syllable structure adaptations occur when the borrowing language has tighter syllable structure constraints than the donor language, when the donor language allows segment sequences that the borrowing language does not allow. Languages may disallow syllable codas or have restrictions on licit coda segments (Japanese only allows a placeless nasal or the first part of a geminate in coda position, for example). Languages may disallow consonant clusters, especially tautosyllabically, as complex onsets or codas. Languages may have strict sonority sequencing restrictions, disallowing, for example, sonority reversals in onsets and codas, [sC] clusters or imposing a minimal sonority distance on clusters. When languages with some such constraint on possible syllable structures borrow words from languages with laxer constraints, the shape of these words has to be adapted to satisfy the borrowing language’s stricter constraints. This can be done in two different ways: by deleting offending segments or by inserting material into disallowed segment sequences.
To be more concrete, assume a language with strict CV syllable structure (as is found in many Niger-Congo or Oceanic languages) borrowing from English, with its tolerance of syllable codas and consonant clusters. When borrowing a word like bus or truck, the language therefore has two options: to delete the coda consonant and one of the segments in the onset cluster, or to insert vowels to achieve strict alternations between consonants and vowels. Both strategies are attested, as the borrowings from White Hmong (Golston & Yang 2001) and Shona (Uffmann 2007) show in (3).
Shona inserts a vowel into onset clusters and to avoid codas; White Hmong deletes consonants. Both end up without onsets or coda clusters (though Hmong allows a range of complex segments).1 While one may expect to find both deletion and insertion as commonly attested strategies, Shona type repairs with epenthetic vowels are much more common than the White Hmong pattern. Paradis & LaCharité (1997) refer to this as the ‘Preservation Principle’ which they argue is operative as a general principle in loanword phonology. Indeed, vowel epenthesis rather than consonant deletion is attested in many languages, for example Tswana (Batibo 1995), Fijian (Kenstowicz 2007), Samoan (Uffmann 2007), Japanese (Park 1987). Cases of deletion then need to be explained by other overriding principles. In fact, there is a very strong word size restrictor constraint (p. 649) operating in Hmong: words are preferably monosyllabic. The preference for epenthesis over deletion in loanwords is noteworthy, since it has been argued that in regular language change, deletion is a more common process (Vennemann 1988, Singh & Muysken 1995).2
18.104.22.168 Accent and Tone
Loans are also adapted with regard to accent and tone. L1 and L2 may have different stress systems, or one (or both) may be a tone language, raising the question of how stress in a donor language may map onto the tone system of the borrowing language, or vice versa. The interested reader is also referred to the excellent overview of phenomena in this area of suprasegmental adaptation in Kang (2010), and see also Lahiri (this volume) for a discussion of the effect of borrowing in stress systems, and Ratliff (this volume) for some discussion of borrowing in tone systems.
With respect to stress, a frequently found pattern is that the native stress system is used, overwriting any input stress. See Kang (2010) for a host of examples from different languages. This is the case when stress is entirely predictable in the borrowing language (for example, main stress consistently on the initial or final syllable). Peperkamp & Dupoux (2002) refer to this as stress ‘deafness’, arguing from psycholinguistic evidence that speakers whose L1 has fully predictable stress lose the ability to accurately perceive stress that deviates from their L1 norm. So speakers of such languages (for example, Hungarian, Finnish, French) ignore L2 stress and simply apply L1 stress assignment principles.
It is not the case that L1 stress is always ignored, however. The claim of ‘deafness’ is a bit strong, as there are cases in which a word is modified in order to align the L1 and L2 stress patterns. For example, Hungarian speakers can delete an initial unstressed vowel in the loan to bring the loanword in line with the Hungarian pattern of initial main stress (Kang 2010). Kenstowicz (2007) reports that Fijian loanwords follow regular Fijian stress assignment, which builds moraic trochees from the right. However, vowel quantity may change (vowels can lengthen) in order to align the stress in the source word with the stress in the loan. Thus, tobácco is borrowed as [ta(váko)], with short vowels, matching loan stress with the original stress, while the initial syllable in cólony undergoes lengthening in order to be footed and stressed, even though it only receives secondary stress: [(kòː)(lóni)].
Finally, there are cases in which heavy borrowing modified the original stress system, where loanwords now provide crucial evidence for regular stress assignment in these languages. Crucial cases are Germanic languages that abandoned initial stress as a consequence of borrowing especially from Romance languages and developed innovative stress patterns, as in the analyses of Rice (2006) (Norwegian), Fikkert et al. (2006, (p. 650) see also Lahiri, this volume) (English), and Alber (1997), Speyer (2009) (German). This is not restricted to Germanic, of course. Fitzgerald (1999) uses loanword evidence in her discussion of stress in Tohono O’odham. Pitch accent in Japanese loanwords also seems to constitute an innovation with the emergence of a Latin-style accentuation system, although Kubozono (2006) shows how the emergence of such a system can be understood from the interaction of borrowing with the native prosodic system.
When tone languages borrow words from stress languages, they have essentially two options, to map stress onto tones, using the effect stress has on pitch, or to disregard stress and use default tones or let some other factor decide what tone(s) to use in the loan (see also Ratliff, this volume). Languages with a small set of level tones typically equate high (H) tones with stressed syllables and low (L) tones with unstressed syllables (e.g. Hausa, Leben 1996; Yoruba, Kenstowicz 2006). The same pattern, though with a few complications, can hold in languages with contour tones, where level high or falling tones can be associated with stressed syllables in the etymon, as in Cantonese (Silverman 1992) or Mandarin Chinese (Wu 2006). There are, however, cases in which input stress is disregarded for tone assignment. Hsieh & Kenstowicz (2008) find that the choice between H or L tone assignment in English loanwords in Lhasa Tibetan depends on the laryngeal specification of the initial segment of the English word: L if it is a voiced obstruent or sonorant, H if it is voiceless. Golston & Yang (2001) show that a complex mix of segmental and syllabic factors plays a role in tone assignment in English loans in White Hmong, while French loans are realized with default L tones. Kenstowicz and Suchato (2006) find that in Thai, syllable structure decides what tone an English loanword will have: syllables closed by an obstruent will have a high tone, others are mid.3
There is comparatively little research on accent languages adopting words from tone languages. Ito & Kenstowicz (2009) is one of the very few studies in this area, looking at how Mandarin tones are adapted in a pitch accent system, Yanbian Korean. The challenge here is how the four contour tones (and combinations of contour tones) can be mapped onto the two pitch accent types, HL and LH. They find that the mapping is largely predictable, though complex, involving the preservation of the most salient tonal properties of the incoming form.
Finally, we can look at how tone languages adapt words from tone languages. Here, the same two options are found as in the adaptation of words from stress languages, either attempting a close match, or disregarding the tonal pattern of the incoming form. The examples discussed in Kang (2010) suggest that languages with similar tone systems can adopt each others’ tones (for example, level tones in African languages) but that the mapping is harder the more typologically distinct two systems are. Thus, Lhasa Tibetan speakers disregard tones in Mandarin borrowings but use a strategy similar to the one they use for English loans, assigning L tone to sonorant-initial forms and H to obstruent-initial forms.
(p. 651) In sum, then, one finds a wealth of phenomena when looking at suprasegmental and prosodic adaptations. As with segmental adaptations, the question is what motivates the shape of the borrowings, what exactly happens in the adaptation process, and what this can tell us about language processing or phonological theory. It is to these questions that we turn now.
37.2.3 How are the Adaptations Motivated?
Having established a typology of phonological adaptations, an important question is how the borrower knows what to do with an incoming form to adapt it to the native phonological system. This question was particularly pressing in early generative rule-based theory: where do the adaptation rules originate? How would a speaker set up a rule of, say, /θ, ð/ → /t, d/ when their native grammar does not have /θ, ð/ and there is no existing model for such a rule? The idea of morpheme structure conditions, developed in Stanley (1967) helps motivate why a rule should be set up, but there is no indication of why the rule should be in this specific shape.
It is of no surprise, then, that loanword adaptation received renewed interest with the development of constraint-based models of phonology in which rules are demoted to the status of automatic repairs, as in Paradis (1988), or done away with completely, as in Optimality Theory (OT; Prince & Smolensky 1993/2004, and see also Holt, this volume). For Paradis, rules are replaced with repair strategies that are activated as the result of a constraint violation. These strategies are not arbitrary, but the phonological grammar is strongly constrained with respect to what a possible repair is. The repairs found in loanword adaptation can then emerge from abstract principles of repairs. In OT, there are only constraints on surface forms, no conditions on underlying forms (the principle of Richness of the Base). Thus, any structure not found in a language has to be ruled out by the constraint ranking of that language, which applies to native forms and borrowings alike. To come back to our hypothetical borrower faced with English /θ, ð/, sounds she does not have in her L1, this means that her grammar already has a high-ranked constraint *[θ, ð], and the adaptation should follow from the interaction of this constraint with faithfulness constraints such as Ident(F). The same argument holds for suprasegmental adaptations. For example, if syllable structure constraints like NoCoda and *ComplexOnset must be satisfied in native forms, their high ranking will also apply to non-native material entering the OT grammar. To scholars working on loanword adaptation, OT was thus a very attractive proposition, offering a principled way of understanding why and how adaptations take place. At the same time, loanword adaptation furnished theorists with a strong argument in favour of OT, or other surface constraint based frameworks: what if no constraints on outputs can motivate loanword adaptation and also give us an understanding of how words are adapted?
One problem remains, though. The constraint ranking that can be deduced from the adaptations may not be directly motivated from the native phonology and thus remain as stipulative as the above-mentioned repair rules. Coming back to the example (p. 652) of adapting English /θ, ð/—why would one language adapt them as /t, d/, not as /s, z/ (while other languages do the opposite)? We could argue that this is due to the relative ranking of different Ident constraints in each language, but do we always have evidence for these rankings? Worse are cases in which native processes flatly contradict whatever ranking we could stipulate for explaining adaptations. A celebrated example is that of final stops in Korean (Kenstowicz 2005, Boersma & Hamann 2009, among others). In word-final position, Korean only allows plain stops, but not tense or aspirated stops. In the native phonology, an underlying final aspirated stop will neutralize to plain, as in /pɑth/ ‘field’, realized [pɑt] (compare locative [pɑthɛ], where /th/ is retained faithfully). In loanword adaptation, however, there is no deaspiration, but a word-final aspirated stop, as, for example, in week, will usually be repaired by vowel epenthesis, moving the offending consonant into an onset. Hence, week > [wikhɨ]. Native and loanword phonology treat the same underlying structure differently. One remedy is to invoke output-output faithfulness constraints that hold specifically for relations between the loanword and its source realization, as in Kenstowicz (2005), or to argue for a distinction between perception and production grammars (Boersma & Hamann 2009), assuming that adaptations take part in perception rather than in production, a point we will discuss in greater detail in the next section.
In sum, there is no straightforward way of accounting for all adaptations just by recourse to pre-existing phonological grammars; loanword adaptation contains innovations, and the question is what resources speakers are drawing upon to motivate these innovative processes. This brings us to the more fundamental questions regarding the adaptation process mentioned earlier: what do speakers actually do when they adapt a word?
37.3 How are Loanwords Adapted?
We have seen that there are two main positions regarding how loanwords are adapted: either it is done by choosing an underlying form whose realization is phonetically or perceptually as close to the source phonetic form as possible, or by setting up more abstract phonological correspondences between sounds. Different models have been proposed in either camp, and this section will introduce the different models, outline problems, and make a new proposal that incorporates the findings from both camps, arguing that there are two main principles at work, similarity and contrast.
37.3.1 Perceptual Adaptation
One position in the debate that has gained in popularity in recent years (see e.g. Kenstowicz 2005, 2007, Adler 2006, Shinohara 2006, Peperkamp et al. 2008, Boersma & Hamann 2009 among others) holds that loanword adaptation is a process of primarily (p. 653) phonetic approximation: loanwords are nativized in a phonetically minimal way so that the surface form of the loan deviates as little as possible from the original in perceptual terms. Put more simply, borrowers want the loan to sound as much as possible like the original L2 form. Within this line of thought, there are two opposite camps, however, with respect to where these adaptations take place, in the phonological grammar or already in raw phonetic perception.
The latter point is argued for explicitly in Peperkamp (2005), Peperkamp & Dupoux (2003), Peperkamp et al. (2008), and Silverman (1992), who argue in favour of a perceptual filter that already modifies the incoming acoustic form to match it with native phonological categories (although Silverman proposes an additional phonological level of adaptation). According to Peperkamp, the incoming form is already fully adapted before it reaches the level of phonological computation. She refers to this as perceptual ‘deafness’: borrowers do not hear that their (adapted) form of the loanword sounds different from the original. Support for this position comes from perception studies indicating that speakers are indeed unable to distinguish between two phonetic stimuli that are non-contrastive in their L1. A particularly striking study is Dupoux et al. (1999) who report that Japanese subjects perceive illusory vowels in certain phonotactic environments. Recall that Japanese does not allow consonant clusters, and that clusters in loanwords are broken up by a vowel, mostly default [u]. Dupoux et al. (1999) found that speakers report hearing medial [u] in stimuli like ebzo and fail to distinguish ebzo and ebuzo. In its focus on (mis-)perception this approach is thus very similar to listener-based approaches to language change (see Ohala 1981, Blevins 2004a, Blevins, this volume, and Yu, this volume), which also rely on speakers not reliably perceiving a distinction and therefore modifying underlying forms.
Alternatively, perceptual adaptation may happen in the phonology. Borrowers are able to perceive the incoming sounds more or less accurately and then match them with the phonetically closest native categories in their phonological grammar, a view espoused in Yip (1993), Shinohara (2006), Kenstowicz (2007), Boersma & Hamann (2009). Perceptual similarity is thus an active choice rather than selective deafness. Shinohara (2006) provides a good example of an adaptation that cannot be purely phonetic because it takes into account prosodic information, the fate of liquids in onset clusters in Cantonese loanwords. /Cl/ clusters are disallowed in Cantonese and have to be repaired. /l/ is deleted if the resulting word is at least disyllabic. If deletion were to result in a monosyllabic form, however, the cluster is broken up by epenthesizing a vowel between the two consonants, creating a disyllabic form. Hence freezer, place > fisa, peysi (deletion) but cream, fluke > keylim, fuluk (preservation of the liquid). To Shinohara, this is conclusive evidence that English /l/ is accurately perceived but that additional grammatical factors (here, a word size constraint) decide whether it is also going to be pronounced.
Additional evidence for the phonological view of perceptual adaptation comes from follow-up studies to the illusory vowel perception study of Dupoux et al. (1999), which are more sceptical of Peperkamp’s and Dupoux’s interpretation of the results. Monahan et al. (2009) look at epenthetic [o] in Japanese, which is inserted after /t, d/, and fail (p. 654) to replicate the effect while confirming illusory perception of [u]: Japanese speakers fail to pick up the difference between ebzo and ebuzo, but they have no problem with edma vs edoma (and eduma). Monahan et al. (2009) conclude that phonological knowledge must play a role in the perception task. Speakers parse the signal and know where syllable structure constraints require a vowel. Since /u/ is often devoiced and shortened in Japanese, speakers know that the absence of good cues for [u] in the acoustic signal does not equate the absence of the vowel in the underlying form. This is not the case for Japan-ese /o/, which is never devoiced or shortened. Speakers therefore expect robust perceptual cues for [o] and accurately perceive the difference between edma and edoma.
Formally, there are two types of proposal for modelling perceptual adaptation in the phonological grammar, both using OT. Kenstowicz (2007) uses the P-map proposal from Steriade (2001a), in which speakers have so-called perceptual maps at their disposal, encoding the perceptual similarity between sounds. Speakers then draw upon this knowledge when the grammar requires an unfaithful parse of the input and choose a candidate form that deviates perceptually the least from the input form.
Boersma & Hamann (2009) take a different approach, analysing loanword adaptation within their bidirectional model of phonology, which unifies production and perception grammars in one constraint ranking. There are, however, different constraints for perception and production, and loanword adaptation occurs in the perception component of the grammar. The approach can thus explain mismatches between loanword adaptations and native processes. They analyse the above-mentioned Korean mismatch by modelling laryngeal neutralization in a traditional OT (production) grammar. Epenthesis in loanwords, however, is perceptually motivated by a high-ranked constraint that prohibits mapping the perception of aspiration to a [–aspirated] segment, and since only [–aspirated] segments can stand in word-final position, epenthesis is necessary to satisfy the constraint on the preservation of perceptual aspiration. The achievement of Boersma & Hamann (2009) is that they do not need to stipulate any loanword-specific rankings or mechanisms. Instead, they can motivate their perception rankings directly from native Korean perception where, indeed, the presence of aspiration in the phonetic signal entails underlying phonological aspiration. The question is, however, if all adaptations are perceptually motivated, especially in the light of the above finding that allophonic information is often not drawn upon in the adaptation process. This brings us to the alternative proposal, that adaptations are based on phonological equivalences.
37.3.2 Phonological Adaptation
The central idea of the phonological approach is that it is not perceptual similarity that counts but phonological equivalence. The approach goes back to Haugen (1950a) and Hyman (1970a), among others, and it is defended and developed in Paradis & LaCharité (1997), LaCharité & Paradis (2002, 2005), Paradis & Tremblay (2009). The core idea is that L2 phonemes are identified and matched with L1 phonemes that are analyzed as equivalent. In many cases, the perceptual approximation account and the phonological (p. 655) equivalence account converge since phonologically equivalent structures should also be phonetically similar, under the relatively uncontroversial assumption that phonological representations are grounded in phonetics (e.g via distinctive features which have consistent phonetic correlates). In some cases, however, they make different predictions, and LaCharité and Paradis support their position by pointing out adaptations that do not go for the closest phonetic match but seem to map phonologically equivalent sounds (see e.g. LaCharité & Paradis 2002, 2005, Paradis & Tremblay 2009). I will now review some of these cases before coming to a critical discussion of their proposal and an attempt to unify the two positions.
A strong case in point is the preservation of laryngeal contrasts such as in /p/ vs /b/ in borrowings between languages that have a plain voicing contrast [p] vs [b] (such as French, Italian, Dutch, Russian) and languages in which this opposition is expressed primarily as an aspiration contrast [p] vs [ph] (such as English, German, Korean, Mandarin). Crucially, the ‘weaker’ member of the aspiration pair corresponds phonetically and perceptually most closely to the stronger member of the voicing pair. One could thus expect French /p/ to be adapted as /b/ in English, or English /b/ to be adapted as /p/ in French, but according to LaCharité & Paradis (2002, 2005) this is never the case. Instead, contrast is preserved, and /p, b/ in one language are borrowed as /p, b/ (or /ph, p/) in the other language, irrespective of whether the contrast manifests itself phonetically as voicing or aspiration (see also Iverson & Salmons 2008). Phonetic detail does not seem to play a role: although aspiration on English /p/ is contextual (as is devoicing of English /b/), all English /p/ map onto Mandarin aspirated /pʰ/, irrespective of whether they are aspirated in English (Paradis & Tremblay 2009).
This preservation of underlying phonological contrasts despite a potential perceptual pull towards neutralization is not restricted to laryngeal contrasts, as Paradis and LaCharité show. Another case they cite is the cross-linguistic identification of rhotics despite their phonetic variability. So, French /r/, which is a uvular continuant [ʁ ~ ʀ], is adopted in Arabic as [r], although Arabic has uvular fricatives, which are perceptually very close to the French rhotic. A similar perceptual mismatch can be seen in English loans in Japanese, in which the English rhotic, phonetically [ɻ ~ ɹ ] is always borrowed as Japanese /r/, a tap or flap, although perception studies show that English /r/ is perceptually more confusable with Japanese /w/. Conversely, intervocalic English /t, d/ which are realized as a flapped [ɾ] in American English, never are adopted as /r/ in languages such as Mexican Spanish where the rhotic is flapped [ɾ]. Vowel adaptations show the same pattern of phonological rather than phonetic equivalence, according to Paradis & LaCharité. A case in point here is seen with the English lax vowels [ɪ, ʊ], which are (p. 656) phonetically closer to /e, o/ than to /i, u/ in languages that do not have a tense-lax contrast. Nevertheless, they are consistently adapted as /i, u/, reflecting their phonological status as [+high] vowels (LaCharité & Paradis 2002).
There has been much recent work on the perception of L2 contrasts, especially associated with Flege’s work (e.g. Flege 2003, Flege & MacKay 2004). What is still missing is a systematic survey of the extent to which loanword adaptation patterns match up with findings on perceptual similarity. A casual glance suggests that there may be quite a few mismatches, going far beyond the examples cited by LaCharité & Paradis. For example, Flege & MacKay (2004) show that Italian learners are prone to confusing English /æ/ and /ɛ/ but do not confuse /æ/ and /ʌ/. Yet, in loanwords, both /æ/ and /ʌ/ usually map onto Italian /a/, while /ɛ/ maps onto /ɛ/. Thus, club, chat, track, hacker are all realized with /a/ (data from Morandini 2007).4 From the perceptual viewpoint, such mappings are unexpected.
One final point that is problematic for the perceptual approach concerns variable adaptation. We have seen that English /θ, ð/ are variably adapted as /t, d/ and /s, z/ in different languages. The perceptual approach would have to show that in these languages, the respective choice would, for some reason, be the closer perceptual match. More problematically, the actual closest perceptual match is [f], also witnessed in historical change (Blevins 2006c). In his discussion of vowel epenthesis strategies, Uffmann (2007) points out that different languages use different strategies to establish the quality of the epenthetic segment. Under the perceptual approach, one would have to claim that the different strategies have a different perceptual salience in different languages, for whatever reason. Instead, there seem to be different ways of reducing salience, and that the decision taken in a language, which path to go, is ultimately phonological: very often there is no default perceptual repair, but there is choice.
A necessary condition for adaptations to reflect the phonological rather than the phonetic status of a segment is that borrowers are aware of the phonological status of the borrowed segment. Supporters of the phonological approach therefore argue that loanwords are first introduced by competent bilinguals (Paradis & LaCharité 1997) with a profound knowledge of both phonological systems. The position taken by Dupoux and Peperkamp, on the other hand, requires a relatively low level of competence in the L2 in order for perceptual deafness to play a role. Given that fluent bilinguals have native or native-like perceptual abilities in both languages (see, for example, Flege 2003), their role must be marginal. They may introduce words to a speech community, but it is the monolingual community that adapts them. The phonological-perceptual approach (Kenstowicz 2005, 2007) is comparatively agnostic with respect to this question, although a certain level of perceptual accuracy seems to be required.
All these arguments do not invalidate the perceptual approach, however, since there are phenomena that are hard to capture under an approach that only takes into account phonological equivalence. We have already seen examples of adaptations that are distinct from processes in the native phonology, as in the adaptation of final stops (p. 657) in Korean. The phonological approach has no principled explanation for this, and can only stipulate that different processes apply to loans. Additional support for the relevance of perceptual factors comes from positional asymmetries that can occur in adaptation.
Kenstowicz (2007) reports two such asymmetries from the adaptation of English loanwords in Fijian. The first concerns the adaptation of voiced stops. Fijian does not have a series of voiced stops, but it has prenasalized stops. In word-initial position, voiced stops are adapted as prenasalized stops, thus desk > [ndesi]. In non-initial position, however, voiced stops are adapted as voiceless stops, thus rabbi > [ra:pai]. Instead, nasal+stop sequences are adapted as prenasalized stops, such as bank > [mbaŋge]. Kenstowicz’s explanation is that the nasal portion of prenasalized stops is very weak in initial position, where it lacks support from a preceding vowel and therefore is perceptually similar to a voiced stop, due to its very weak nasal cues, while in non-initial position, the nasal cues are much stronger. The second asymmetry concerns consonant clusters. Onset clusters in Fijian are invariably broken up by an epenthetic vowel, preserving all input consonants. Coda clusters are often adapted by deleting one of the consonants, however. Thus, east > [isi] while steak > [siteki]; *[seki, teki]. Kenstowicz explains this asymmetry with the weaker perceptual cues that coda consonants have, making them more vulnerable to deletion.
A further point in support of perceptual adaptation is discussed by Hsieh et al. (2009), who find an intriguing contrast shift in Mandarin Chinese. While /n/ and /ŋ/ are two separate phonemes in Mandarin, [æ] and [ɑ] are allophones: front [æ] is used before /n/, back [ɑ] is used before velar /ŋ/. Thus, [æn, ɑŋ] are possible, but *[æŋ, ɑn] are not. Rather than preserving nasal contrast in loanwords (as in ran vs rang), however, speakers choose to preserve the vocalic distinction between, for example, sang vs song, by selecting the appropriate nasal. Preservation of the accurate vowel thus trumps preservation of the accurate nasal, although the former distinction is allophonic and the latter is phonemic. Hsieh et al. (2009) argue that this reallocation of contrast is perceptually motivated. Nasal place contrast is perceptually weak and therefore lost vis-à-vis the vowel contrast.
There is yet another question that has to be asked about the phonological approach: if it involves the mapping of phonologically corresponding segments, how do speakers, even bilingual speakers, establish this correspondence? If we assume a universal set of distinctive features, we might argue that it is segments with the same feature make-up, but that cannot be right: we have seen above that aspiration contrasts are mapped onto voicing contrasts and vice versa. Equivalence thus has to be calculated at a more abstract level, possibly taking into account feature underspecification (to account for the laryngeal mismatches or the mapping of phonetically distinct rhotics), but—asked polemically—if phonologists cannot agree on a universal set of segment specifications, how can linguistically naive speakers set up correspondence relations between the segments of two languages, if not phonetically? We will come back to this, arguing to replace the idea of phonological equivalence with the idea of contrast and opposition (in the classic sense of Trubetzkoy 1939).
(p. 658) 3.3 Intermediate Positions
Recently, more researchers have taken an intermediate position, endorsing the perceptual approach but allowing more abstract phonological principles to play a role as well, as for example the analysis of Korean in Kenstowicz (2005) and the analysis of Sesotho in Rose & Demuth (2006). In his discussion of Korean adaptations, Kenstowicz (2005) distinguishes between perceptually motivated adaptations and adaptations that seem to require a more abstract phonological mapping. Rose & Demuth (2006), while generally taking a phonological approach, also notice a few wrinkles in their analysis of Sesotho borrowings and suggest phonetic approximation in these cases. What is often missing from the mixed approaches, though, is an attempt to explain why certain adaptations are phonological while others are perceptual.
Two studies on Japanese bring us closer to understanding how the two principles interact, Smith (2006) and Dohlus (2010). Smith (2006) looks at doublets in Japanese loanwords and finds that many doublets can be explained by the application of one of the two processes, for example, the adaptation of jitterbug as either [dʒittabaɡɡɯ] or [dʒiɾɯba]. The former follows established conventions for Japanese borrowings (see, for example, in Park 1987, Katayama 1998) and also Paradis’s Preservation Principle, not deleting underlying material. The latter is odd from that perspective, deleting a final coda and also adapting (flapped) English /t/ as [ɾ], a perceptual assimilation of the kind that Paradis and LaCharité claim does not exist. Smith argues that this is an indication of a purely perceptual borrowing, whereas the longer form is also mediated through orthography. It is unclear, though, how common forms such as [dʒiɾɯba] are, given that it is a non-canonical adaptation. Knowledge about the frequency of such perceptual adaptations could help us understand what the relative role of perception and phonology in a specific contact setting are.
Dohlus (2010) looks at the fate of the front rounded vowels /ø, œ/ in German and French loanwords in Japanese, which lacks these vowels, and finds different strategies in the two languages. German /ø, œ/ are adapted as /e/ (unrounding) while French /ø, œ/ are adapted as /u/. Her perceptual experiments show that Japanese /u/ is indeed the closest perceptual match with /ø, œ/. Dohlus therefore concludes that the French adaptation is perceptual while the German adaptation is phonological (best match in terms of distinctive features, [–back, –high]). The reason for the difference may be that German borrowings are generally older and from the domains of arts and sciences, thus more likely to be introduced via the written form and via educated, possibly bilingual speakers. French borrowings on the other hand are more widely used and found in the oral domain, facilitating perceptual adaptation. She also notices that there used to be variation in the transliteration of German loans, leading to a drive towards standardization in the 20th century, which means that German adaptations are also highly conventionalized.
(p. 659) 37.3.4 Unifying the Positions
To summarize the discussion, there is evidence for both perceptual factors in adaptation, borrowers trying to yield a phonetically minimally deviant adaptation, and for more abstract phonological principles. How can we bring the different ideas together?
To begin, the alleged dichotomy between a perceptual and a phonological approach is somewhat misleading. As we have seen, a divide goes right through the proponents of the perceptual approach, dividing those that believe in purely phonetic adaptations (such as Peperkamp & Dupoux) from those who want to integrate perceptual similarity into the phonological grammar (like Kenstowicz). The divide between this position and the ‘phonological’ approach of Paradis & LaCharité then is less about the fundamentals of adaptation—both camps would agree that adaptations occur in the phonological grammar—but more about what the grammar consists of, whether the phonological grammar can take into account and measure phonetic similarity, or whether it is a fairly abstract symbolic manipulation device. This debate goes beyond loanword adaptation to the heart of many debates in phonological theory today.
We cannot say what the nature of phonological computation is, and will remain agnostic regarding the question of perceptual knowledge in phonology, but we can take a more general look at what kinds of phonological knowledge are utilized in loanword adaptation, without having to commit ourselves to a specific theoretical angle. Two main principles seem to be relevant in loanword adaptation, and these are crucial for the question of what phonologically equivalent or corresponding sounds are: phonetic similarity and phonological contrast. A phoneme system is defined by the segmental contrasts of a language and the oppositions that individual phonemes enter, via distinctive features (see Dresher 2009 for an overview, and also Dresher, this volume, on change and contrast). Identifying two corresponding segments in another language thus consists of two distinct operations: identifying the oppositions that the segment enters in the L2 and finding a phonetically similar segment in the L1 (but not necessarily the most similar segment) that stands in similar oppositions. Thus, voicing contrasts can be mapped onto aspiration contrasts and vice versa, by identifying segment pairs that show some laryngeal contrast and by identifying the stronger and weaker member of the pair. This idea of systemic equivalence may also become clearer by revisiting the vowel adaptations discussed in LaCharité & Paradis (2002). Recall that English /ɪ/ is adapted as /i/ in languages lacking a tense-lax contrast, although it is phonetically closer to /e/. In terms of contrast and opposition, however, /ɪ/ is the highest front lax vowel in English, standing in a three-way height opposition with /ɛ, æ/ and in a backness opposition with /ʊ/. This position within a system of contrasting segments motivates the mapping to /i/, which is in a similar position (highest front vowel).
Contrast can also explain why phonetic detail sometimes seems to matter and sometimes does not. Recall that in Thai, English voiceless stops map on plain voiceless stops or aspirated voiceless stops, in line with English allophony (Kenstowicz & Suchato 2006), but that in other languages with aspirated and unaspirated voiceless (p. 660) stops, English stops map categorically onto the aspirated stop, irrespective of whether it actually is aspirated in that position (for example, in Korean, Burmese and Mandarin Chinese borrowings). The reason for this may lie in the preservation of contrast. Thai has a three-way contrast, and can thus map English /p/ on two distinct categories, /p/ and /ph/, without obliterating input contrast. In Burmese, however, which also has a three-way contrast, the aspirated stops realize input fricatives. Realizing some English /p/ as [ph] would therefore cause a partial merger with English /f/. Mandarin has only a two-way contrast, therefore also goes for the mapping that preserves contrast rather than the phonetically closest mapping. Phonetic similarity thus plays a role but is mediated by systemic considerations. When contrast is no longer at stake because a merger is unavoidable, phonetic similarity can come to the fore, whence the Japanese adaptation of French /œ, ø/ as /u/ (Dohlus 2010): when a large vowel system like the French one has to be mapped onto the 5-vowel system of Japanese, loss of contrast is unavoidable. In the absence of clear evidence regarding phonological patterning, the closest perceptual match can then be chosen. How the two factors, similarity and contrast, can exactly be brought together in a formal analysis, is an important question, but has to be outside the scope of this chapter.
Following Paradis and LaCharité, my proposal requires competent bilinguals as agents of change, speakers who are able to assess phonological equivalence between segments as a measure of both phonetic similarity and similar systemic properties, including systems of oppositions and natural class behaviour. Does this mean that perceptual deafness has no role to play? That claim would also be too strong, as the analysis of Japanese doublets by Smith (2006) showed. It is therefore likely that in the process of borrowing there is initially a pool of different variants, with phonological adaptations by competent bilinguals coexisting with some monolinguals’ phonetic approximations. This pool of variants then undergoes levelling and conventionalization, and it is possible that the role of conventionalization has also been underestimated in much present research. What this means, though, is that loanwords are not only adapted by individuals, but by speech communities as a whole, suggesting that a sociolinguistic perspective is also necessary. Section 37.4 will look at non-adaptations, variation, and briefly touch upon the role of code-switching, arguing that borrowings can only be understood from the perspective of the speech community, not the individual.
While adaptation is the norm, non-adaptations also occur, and a short section on non-adaptations or partial adaptations is important in order to provide a more comprehensive view of the topic. The literature knows many cases of non-adaptation, especially in variationist or explicitly sociolinguistic approaches that look at borrowing embedded in a bilingual or multilingual context, which also involves code-switching. One example of this is Chimhundu (1983), whose account of loanwords in Shona looks at both (p. 661) adaptation patterns and the linguistic situation in Zimbabwe, which is characterized by diglossia and extensive code-switching, with the result that loanwords may be adapted, non-adapted or anything in between. Thus, train may be [tireni, treni, tren], under the proviso that the unadapted forms are uttered only by proficient bilinguals—though bilingualism is common in Africa, and therefore unadapted forms are not exceptional (see, for example, also Batibo 1995 for phonotactic adaptations and non-adaptations in Tswana and Swahili).
Unadapted forms have received comparatively little attention in phonological analyses, because the research question typically is how loans are adapted; unadapted forms, then, are little more than ‘noise’. A notable exception is LaCharité & Paradis (2005), who provide rates of non-adaptations (which they call ‘importations’) and discuss some of them, noting in particular the link between rate of adaptation and degree of bilingualism: English loanwords in Québec are more likely to remain unadapted the higher the degree of bilingualism in the speech community is. They never discuss systematically what kinds of non-adaptations are found, however, whether some sounds are less likely to be retained than others, for example. A systematic survey could prove valuable for phonological theory as well, with respect to claims about (universal) markedness, gaps in inventories, and so forth. One of the few surveys I am aware of (without discussing consequences for phonological theory) is the detailed quantitative analysis of Spanish loanwords in Nahuatl in San Giacomo (2009), which gives a tally of non-adaptations per segment, showing that some segments are more likely to remain unadapted than others. More studies of this kind would be interesting for the phonologist and the typologist.
When unadapated forms become established in the borrowing language, this results in phonological change in the language, for example, by introducing new phonemes or by relaxing phonotactic constraints. Investigation into these changes has led to the idea of a stratified lexicon, to which we will turn briefly now.
37.4.1 Lexical Strata
It has long been noted that borrowing can introduce new phonemes into a language (see, e.g. Thomason & Kaufman 1988 and references therein, also Eckman & Iverson, this volume). A common pathway is via the phonemicization of a formerly allophonic distinction. The data from Québec French in Picard & Nicol (1982), shown in (2), are a good example. In the native phonology, the diphthongs [aj, aw] are allophones of long monophthongs. However, when English words containing /aj, aw/ are imported, they fail to alternate, introducing contrast and minimal pairs. Similarly, Eckman & Iverson (this volume) discuss the phonemicization of the allophonic contrast between [h] and [ɸ] in Japanese.
Alternatively, a phoneme may be imported without the previous allophonic conditioning. The likelihood of a successful import seems to depend on how well the existing system accommodates the new arrival: a sound that fills an accidental gap in the borrowing language is more easily integrated into the L1 system. A case in point is the (p. 662) reintroduction of [g] into Dutch via borrowing. Historically, earlier Dutch /g/ underwent spirantization to [χ], leaving a gap in the system: all stops come in voiced-voiceless pairs, except for /k/, which is unpaired. Importing /g/ thus fills a gap that the earlier historical change had created. Besides, the new sound only recombines articulatory gestures (or distinctive features) that are already used. Dutch has voiced stops, and it has velar stops. Compare this to the low rate of adaptation of Spanish /r/ in Nahuatl (San Giacomo 2009): Nahuatl does not have a trill; the missing sound is not a combination of familiar gestures, thus /r/ is harder to integrate into the L1 system.
When a new sound is introduced, it is only found in loanwords (although it may subsequently spread to native items). Therefore, different constraints (inventory, phonotactics) may hold for native items and loans in the lexicon of a speaker, a point made already by Fries & Pike (1949), who noted that Spanish loanwords in Mazateco may violate constraints that hold for native items. In particular, they note that a process of postnasal voicing does not hold for Spanish-origin items, thereby introducing a phonemic voicing distinction on stops. In generative phonology, the idea of a stratified or compartmentalized lexicon was developed further. Lightner (1965) notes that native Russian words undergo processes that other, imported forms do not, and that non-Russian Slavic words undergo a subset of these processes. He suggests diacritic features on lexical items, such as [±Russian] and [±Slavic]. Aronoff (1976) takes up this idea and suggests a feature [±Latinate] for the English lexicon, to account for stem-affix combinatorics, an idea that is extended also into phonology, for example by Plag (1999). The idea of affix classes goes in a similar direction; Siegel (1974) already notices the role of etymology in class membership. Ito and Mester (1995) claim that the relation between such sublexica is not random but argue for a core-periphery structure of the lexicon instead, where constraints that hold on native vocabulary (the core) are increasingly relaxed the more one goes to the periphery, where loanwords first appear. Itô and Mester model this in OT by gradually demoting faithfulness constraints from the core grammar to more peripheral grammars. In his careful look at loanwords in Slovenian, Jurgec (2008) argues that faithfulness demotion cannot be the whole story, as he finds instances where there are tighter structural constraints on loans. At the same time, he largely confirms the core-periphery structure, showing in detail how it works in Slovenian, and what intermediate stages of adaptation are found there, demonstrating once more that adaptation is not an either/or scenario.
37.4.2 The Role of Conventionalization
Now, if full adaptation is not a defining feature of loanwords, if loanwords can stay partially or fully unadapted in the lexicon, the question emerges how a loan can be distinguished from a code-switch, a question that has occupied researchers on code-switching (see e.g. Poplack 1980, and Myers-Scotton 1993, for different positions), since there are one-word switches, while loans can be larger than an individual word (as in déjà vu, je ne sais quoi). The only reliable criterion that researchers seem to be able to agree on is (p. 663) listedness (Poplack et al. 1988, Myers-Scotton 1993, Muysken 1995): loanwords are listed in the L1 mental lexicon, and are understood by a large number of speakers.
This finding has important consequences for theories of loanword adaptation, since it is not the individual who borrows an item (an individual may code-switch and idiosyncratically adapt an item) but the speech community. Consequently, psycholinguistic approaches to loanword adaptation, such as studies of online adaptations and L2 perception, which are the main empirical support for the theory of perceptual adaptation, are insufficient, as they only look at the level of the individual speaker. They need to be supported by an explicitly sociolinguistic approach to loanword adaptation, an approach that fully acknowledges the role of conventions and conventionalization in the borrowing process.
I previously alluded to the idea that some segmental correspondences in borrowings may be conventionalized, disregarding actual (synchronic) perceptual similarity or phonological equivalence. I also suggested that there can be initial variation in the adaptation of a word, depending, for example, on the degree of bilingualism, with perceptual approximations, phonological adaptations and non-adaptations competing in a pool of variants, which will then undergo levelling and focusing on one or two variants. If it is the case, though, that adaptation is sociolinguistically mediated, one should also expect the possibility of purely conventionalized mappings, which new words automatically undergo.
A good example of such a conventionalized mapping is the adaptation of /h/ into Russian. Rather than being deleted or mapped onto [x] (a common L2 strategy), it is mapped onto [g], as in gamburger ‘hamburger’ or galstuk ‘necktie’ < German Halstuch. The motivation for this surprising mapping lies in the regular correspondence between Russian /g/ and Ukrainian /h/ (and Belarusian [ɣ]). Russian speakers use this knowledge to analyse all instances of [h] as a variant realization of /g/.
Such conventionalized mappings may be more frequent than often assumed. Why is English /ʌ/ borrowed as /ɑ/ in German, but as /œ/ in Norwegian (as e.g. in muffin)? Maybe it is convention, each language reflecting different stages in the historical lowering of this vowel in English (from Middle English short /u/) with Norwegian preserving an older approximation—note that some older loanwords in German also have unexpected /œ/, for example pumps, curry. This conventionalization may be helped by imports via the written medium: orthographical influence may not be restricted to spelling pronunciations but also aid in conventionalizing segmental mappings between L1 and L2. Perhaps conventionalization also is an important factor in the conundrum of the differential adaptation of English interdental fricatives. Anecdotally, I know that German learners of English are told at school that /θ, ð/ are /s, z/ ‘spoken with a lisp’. I cannot imagine a similar didactic ‘aid’ in languages where /θ, ð/ are adapted as /t, d/.
Conventions are, of course, influenced by actual language contact. An increase in L2 proficiency may lead to the modification of conventionalized mappings. In Russian, new borrowings containing /h/ often have the perceptually better motivated mapping to [x], as in xaker, imxo, xobbi ‘hacker, imho, hobby’, or lead to doublets; the Hudson River has [g], while Sherlock Holmes’s landlady Mrs Hudson has [x]. The influence (p. 664) of L2 proficiency is also illustrated by the fate of initial /s/ in German loans. Standard German only has /z/, never /s/, in initial position, but English loans like sex, surfing, singer-songwriter introduce initial /s/, presumably under the influence of spoken English. Initial /s/ in Japanese loanwords like samurai, sushi, Suzuki, however, is conventionally realized as [z].
Further support for the idea of conventionalized mappings comes from what Janda et al. (1994) call ‘hyperforeignisms’, conventions based on overgeneralizing a perceived regularity. A striking example concerns the fate of final consonants in borrowings from French. American English speakers in particular overgeneralize the observation that final consonants are prone to deletion and consequently realize coup de grace as [kudəgrɑ:]. Janda et al. list a host of such examples, demonstrating that loanword adaptation is as much about arbitrary mapping conventions as it is about the desire to faithfully reproduce an input form. Models of loanword adaptation therefore have to take into account the role of the speech community; not every adaptation is phonetically minimal.
It is unlikely that one single strategy determines the loanword adaptation process. Instead, I argue that phonetic similarity and phonological equivalence (via contrast and opposition) both play a role, mediated by the degree of bilingualism found in the contact situation. Raw phonetic approximations should be found in cases of casual contact and contact in the oral medium only. Bilingualism and the introduction of words via the written domain favour phonological adaptations and the emergence of mapping conventions, if contact is sustained and a sufficiently large number of words is borrowed. This also means that we expect variation in the early stages of loanword nativization. I have therefore argued that the phonological and the psycholinguistic perspective, both of which are established in the literature, should be supplemented with a sociolinguistic approach. To my knowledge, there are no studies yet that try to relate the formal properties of the adapted word to the sociolinguistic setting in which it was borrowed. Future research could shed some light on this and potentially resolve the old dispute between ‘phonological’ and ‘perceptual’ approaches. Inspiration could be taken from the field of creole studies in which formal and sociolinguistic analyses have been integrated to a larger degree, to a similar end: to establish how the sociodemographic setting during creole formation influenced the outcome, the grammatical properties of the creole. Studies in loanword adaptation may benefit from a similar turn.
So how does loanword adaptation relate to other phonological change? The process of loanword adaptation is, of course, quite different from internally motivated diachronic change. We can see, however, that there are similar theoretical debates, so findings from one domain can potentially yield insight in the other domain as well. Both fields have seen a reorientation toward the listener, in listener-based approaches to diachronic (p. 665) change (Ohala 1981, Blevins 2004a, Yu, this volume) or in the perceptual approach to loanword adaptation, as an alternative to the traditional phonological approach. Central to both fields is the role of markedness, and both were given a fresh angle by Optimality Theory. Historical linguistics has also benefited from developments in quantitative sociolinguistics (see, e.g. Labov 1994, 2001, D’Arcy, this volume); tracing variation and change in real time in a speech community, and loanword adaptation could also benefit from this approach. To conclude, although the linguistic situation is very different, diachronic change and loanword adaptation seem to be subject to the same linguistic forces and principles, with similar debates dividing the scholarly community. Consequently, both fields can probably learn something from inspecting each other’s findings, and develop a fresh perspective on the field that goes beyond current divisions. (p. 666)
(2) Note that the question of what vowel is inserted is also of empirical and theoretical relevance. Many analyses identify a default vowel in a language, typically an unmarked, non-salient vowel. Uffmann (2006, 2007) shows that most languages display a complex interplay of three strategies, default segmentism, vowel harmony, and local assimilation to a consonant (such as inserting [u] after labials); see Uffmann (2007) on how the processes interact in Shona, Sranan, and Samoan, and Kubozono (2002) who identifies all three processes in Japanese.
(3) The high tone can be explained as an effect of the voiceless final obstruent.
(4) There are some instances of /æ/ > /ɛ/, but these are fewer in number. There is /ɛ/, for example, in match, sandwich, backup.