Linguists of the pregenerative period however rarely adduced much internal evidence for their claims about syllable structure, and the evidence that was given centred almost entirely on syllable weight as determined by the metre. The situation changed in the 1970s, as phonologists began discovering many kinds of phenomena that were syllabically conditioned. For ambisyllabicity, a major breakthrough came with Kahn (1976), who showed how such allophonic processes as aspiration, glottalization, and tapping (reduction of alveolar stops to a tap) of stops in American English can be felicitously described in terms of syllable structure. According to Kahn, English basically syllabifies in accordance with the maximum onset principle, but a few subsequent adjustments can make a single consonant be shared between two syllables. One adjustment is that an open syllable will get a coda by sharing the first consonant of a following unstressed syllable, as the /t/ in city. Another is that a word beginning with a vowel will get an onset by sharing the last consonant of the preceding word, so that the /t/ in let Ann is ambisyllabic as well. Gussenhoven (1986) has emended and extended Kahn's analysis to account for perhaps a dozen allophonies in American and British English. A compelling feature of this sort of analysis is that ambisyllabic segments are not only subject to rules of their own, such as tapping, which in American English applies only to ambisyllabic alveolar stops, but also may participate in rules that apply to onsets and also those that apply to codas. For example, in alley, the first vowel is an /æ/, which occurs only in stressed, closed syllables, which places the /l/ in the coda; but the lateral is clear (non-velarized), which is the allophone taken by /l/ in onsets.
Convincing cases have also been made for other modern Germanic languages. Vennemann (1982:280-281) argued that a German word like Roggen `rye' must have an ambisyllabic /g/, since open /o/ otherwise appears only in closed syllables, but all stops are voiceless when not in an onset. Borowsky et al. (1984) stated the case for gemination in Danish, where stressed syllables must be heavy, and so a syllable ending with a stressed short vowel will ambisyllabically share the initial consonant of the next syllable. This consonant is now eligible to bear a stød (glottal stop) or make a preceding /a/ grave, which properties are otherwise features of coda consonants. Hulst (1985) argued that all full (non-schwa) vowels in Netherlandic must be in heavy syllables. If a full short vowel appears at the end of a syllable, a following single consonant ambisyllabically spreads to fill the preceding coda.
Several attempts have been made to account for the same phenomena without ambisyllabicity. Kiparsky (1979:432-440) proferred a foot-based approach, attempting to show that the consonantal allophony of English can be determined by first constructing metrical feet beginning at the left edge of the word and at all stressed syllables, then weakening consonants that follow a [-cons] and are not foot-initial. Selkirk (1982:360-379) more directly addressed the paradoxical status of segments that seem to have both onset-like and coda-like properties when she provided that consonants may move from an onset into the preceding coda during the course of a derivation, picking up properties conditioned by the respective syllable component or conditioning the appropriate phonological changes as they move about. Gussenhoven (1986) demonstrated that both of these solutions were inadequate on empirical grounds. But it is worthwhile to ask why people object to ambisyllabicity. A major objection is that it violates proper bracketing, or specifically, the prosodic hierarchy, which teaches that elements at one prosodic level are properly included in a single parent construct(Selkirk 1982:355). To this one may well ask why segments should be considered part of the prosodic hierarchy; or why geminates, which autosegmental phonologists agree are typically single melodic constituents shared by two syllables, are not an equally big problem; or on exactly what grounds improper bracketing should be excluded from linguistics (Lass 1985:249). A somewhat more troublesome challenge is Picard's objection (1984) that ambisyllabicity has no phonetic grounding, and so amounts to nothing more than an unverifiable formalism the phonologist can make use of. Here the appropriate safeguard would be to make sure that prospective ambisyllabic segments actually have properties both of onsets and of codas. Some phonologists have inverted the requirement, finding ambisyllabicity where segments have properties devolving neither from their status as onsets nor from their status as codas. For example, Suzuki (1985:102-104) claims to have proved that Old English had ambisyllabicity, because /h/ disappeared between sonorants but not next to obstruents or word-marginally. Certainly, /h/ could have been ambisyllabic between sonorants and as such susceptible to a special rule (like Kahn's tapping of ambisyllabic /t/ in American English), but arguing that a segment does something that is not a necessary result of being in either an onset or a coda does not seem like a very convincing proof that it must in fact be in both. In the same way, Kang's argument for ambisyllabicity in Korean (1991:90-94) hinges on the assertion that intervocalic [r] from underlying /l/ (/tali/ -> [tari] `leg') must be ambisyllabic, because not all onset /l/ or all coda /l/ change to [r] (contrast /kamlam/ -> [kamnam] `olive', /m{schwa}lli/ -> [m{schwa}lli] `far'). Or that the voiced bilabial fricative [B] must be ambisyllabic because it only appears between vowels ([t{bari}tsaBa] `hear (humble completive)'), not in other codas or onsets. Not every change has to be conditioned exclusively by syllable structure.
Admittedly, there are a few other objections to ambisyllabicity that are harder to dismiss. Picard also complains (1984:45-46) that different phonologists have significantly different criteria for ambisyllabicity in even a single language such as English. Kahn (1976), as has been seen, provides for ambisyllabicity at the left edge of unstressed syllables; Hooper (1978:189) provides for it at the right edge of stressed syllables; Anderson and Jones (1974) make stress irrelevant. There is also the sticky problem that not even Gussenhoven's recent analysis (1986) fully accounts for all the allophony data, which, to be fair, are certainly complicated by a good deal of variability, even for a single speaker. Finally, it is suspicious that most findings of ambisyllabicity have been restricted to the modern Germanic languages. Arguably, these cases all share in common the loss of geminates in a system where stressed syllables are necessarily bimoric, and all are languages that areally share many features. The case of ambisyllabicity would be bolstered by showing that it is present in a more distantly related language, with no contact to English, and under a different set of historical circumstances. Fortunately Sanskrit is an answer to this last objection.
[Forward to The representation of ambisyllabicity.] [Up to Background] [Up to table of contents]