In revising a paper on complex onset phonotactics involving laterals, a question has come up about the status of /tl/ in Mexican Spanish loanword adaptations from Nahuatl. First, a description from Hualde (1999:171-172):
A word such as atlas ‘atlas’ is pronounced [ˈa.tlas] in almost all of Latin America and in areas of western Spain, while in central and eastern Spain it is pronounced [ˈat.las] ~ [ˈað.las]. … In Mexican Spanish the /tl/ cluster appears even in word-initial position, in toponyms and borrowings from Nahuatl such as Tlaxcala (place name), tlapalería ‘hardware store’, etc.
Lope Blanch (1972:97-98) ascribes this characteristic of Mexican Spanish to the influence of Nahuatl, which has a voiceless dentoalveolar lateral affricate /tɬ/. Presumably, when Spanish speakers were confronted with this phoneme in Nahuatl loanwords and Aztec toponyms, they interpreted it as a bisegmental sequence of coronal /t/ followed by the lateral liquid /l/, both of which exist independently in Spanish. The other possibility is that what is typically transcribed as [tl] is still, in fact, a monosegmental affricate, which might explain why the heterosyllabic parse of medial [t.l] is out (at least for Nahuatl-Spanish bilinguals?).
So, I’m just curious as to what kind of arguments (empirical, theory-internal, or otherwise) would be necessary to motivate the mono- versus bisegmental status of Mexican Spanish /tl/…
Hualde, José Ignacio. 1999. La silabificación en español. Fonología generativa contemporánea de la lengua española, ed. by R. Núñez Cedeño and A. Morales-Front, 177-188. Washington: Georgetown University Press.
Lope Blanch, Juan M. 1972. Estudios sobre el español de México. Mexico City: Universidad Nacional Autónoma de México.
The only commonly-accepted type of evidence I know of for deciding this sort of question is of a distributional sort: for example, if affricates are found word-finally but “genuine” clusters are not, then that’s evidence that the affricates are monosegmental.
But are the English post-alveolar affricates monosegmental or bisegmental? They appear not to cluster with liquids or glides word-initially, unlike other obstruents, so that seems to indicate that they’re bisegmental. On the other hand, a bisegmental analysis leads you to some curious conclusions. For example, the only stop-fricative clusters allowable word-initially are post-alveolar. (A monosegmental analysis doesn’t explain why the only affricates are post-alveolar either, but my impression is that most folks think this to be a lesser problem.)
I’ve always found these distributional arguments to be pretty weak on their own. Basically, I think that we don’t have a well-enough-established or complete enough theory of segments on the one hand and segmental distribution on the other (though we often pretend to).
Another line of argumentation might focus on vowel quality. In many varieties of Spanish, vowels exhibit two allophones, a “long” and a “short” version, expressed in open and closed syllables respectively. Clusters in Spanish tend to be unstable; speakers prefer an analysis that moves the beginning of an onset cluster to the coda of the preceding syllable. On this view, one would expect something like [a.tɬæs] if the affricate is a single phoneme, and [æt.læs] otherwise.
Thanks for your comments, Eric and ACW. Thinking more about distribution, I realize that the Spanish voiceless alveo-palatal /tʃ/, uncontroversially an affricate, cannot occur in coda position, and word-final clusters are pretty much non-existent in the native vocabulary (cf. loanwords such as fax and vals). So with respect to /tʃ/, word-final position doesn’t provide evidence for either analysis.
Interestingly, in some cases Nahuatl word-final /tɬ/ has lost either its lateral component (e.g., tomatl > tomate) or its stop component (e.g., oyametl > oyamel ‘fir tree’). If borrowed final /tl/ is treated as a cluster, then reduction might be due to constraints against complex syllable margins. But again, since affricates aren’t otherwise allowed in coda, it’s hard to know what the relevant constraint is in this case (e.g., *ComplexCoda or *CompSeg).
ACW: I understand that the “short” vowel allophone in [æt.læs] would signal the presence of separable [t] closing the preceding syllable. But I’m still not clear on how the “long” vowel allophone in [a.tɬæs] would clarify the segmental status of what follows. Wouldn’t [a] just tell you that the first syllable is open? That is, regardless of whether you have a single segment or a complex onset cluster, you would still have the “long” [a] in both cases, no?