Up to this point I have painted two radically different pictures of Vedic Sanskrit syllabification. On the one hand, there is the language where consonant clusters split in two, with one consonant being syllabified with the preceding syllable, regardless of the quality of the consonants in question. On the other hand there is the language where the bulk of the phonology and much of the morphology centres around the distinction between consonant clusters that do and those that do not syllabify together into onsets. There would seem to be a fundamental discrepancy here.
It has been suggested that the two systems be reconciled by assigning them to two different historical periods. Chatterji (1926:252) suggested that Vedic Sanskrit was a cluster splitter, but Classical Sanskrit was an onset maximizer; in fact, he seemed to believe that all(9) intervocalic clusters in Classical Sanskrit formed onsets, which for him explained why the native scripts write virtually all consonant strings plus the following vowel (C*V) as a single bundle. Vaux (1992) preferred the more moderate approach of saying that Classical Sanskrit changed from a CS strategy to an OM strategy where onsets are structured pretty much as presented here (a major difference being that he agrees with Steriade 1982and Cho 1990 that sibilants do not syllabify with a following stop). These approaches, however, do not take into consideration the unity of sandhi and morphophonemics that was outlined here, which requires OM syllabification even within the Vedic period. The hypothesis of an historical change would necessitate either developing a completely different explanation of why the phonology seems to behave as if there is OM, or a radical reconstruction of the text of the R.gveda, arguing that much of the sandhi visible in the text is an innovation of later editors. It also requires one to reject synchronic explanations for those features in later Sanskrit that still require CS syllabification, such as variations in the weight of reduplicate prefixes. All such forms would have to be seen as fossilized survivals, and correct formulations for new words would have to be seen as artificially constructed. All this is possible, of course, but is on its face a rather complicated argument.
Selkirk (1982:360-379) favours resyllabification as a general solution to such syllabification paradoxes. Under such a scheme, there would be an OM limited to suffixes, with clusters forming onset clusters when legal, and others splitting into coda and onset, with the codas subject to neutralization. In addition, there would be a later resyllabification not only across word boundaries but between all syllables. At this point, a word like va-jra would become vaj-ra, making the verse metre work out right. This could be refined by providing that OM takes place only in an intermediate lexical stratum; reduplicating prefixes are not subject to it because they are affixed earlier in the derivation, and compounds are not subject to it because they are later.
What is troubling about this analysis is that it posits that the postlexical phonology undoes much of what the earlier phonology worked to accomplish. Whereas onsets are preferred in suffixation, the postlexical phonology prefers to make heavy syllables, even if it has to take onsets apart (Borowsky et al. 1984:36). Whereas the lexical phonology is hostile to having segments like the palatal j in a coda, the postlexical phonology will take j out of a perfectly legal onset cluster in order to place it in a coda. It is curious that different parts of the phonology should have such violently conflicting notions of optimality. A more empirical problem is that some parts of the grammar appear to be both CS-oriented and OM-oriented at the same time, defying any neat stratification. Consider again the alternation of i versus zero in the perfect forms. The i appears between consonants after a heavy syllable, but it is omitted before vowels. Since the measure of heaviness is indifferent to whether the final cluster of the stem could form a legal onset (as does the jn~ in jajn~is.e), this i must be deleted (or added) in the CS stratum of the grammar. But the i is also sensitive to the phonology of the suffix, in that it is always omitted before a vowel. This latter behaviour is idiosyncratic, since i normally becomes y before a vowel, so it cannot simply be referred to the postlexical phonology. Now suffixes would be added at the OM stratum of the grammar, as is clear from the case of yuyujma, where j is not neutralized because it forms part of an onset cluster with the suffix. So the treatment of i appears to be determined at two different strata with conflicting syllabification properties. An even clearer case of the inadequacy of resyllabification is presented by jajn~is.e itself. The i is present because the preceding syllable is heavy, but the second j exists as such because it is in an onset cluster jn~. Postlexical resyllabification will not account for the discrepancy that the j behaves both as a coda element and as an onset element at the moment when the perfect suffixes are added.
The cleanest solution involves recognizing ambisyllabicity. Instead of concluding that CS and OM occur at different times historically or at different parts of the grammar, it is better to say that both are present simultaneously. For jajn~is.e, the second j is both in the first syllable, making it heavy, and in the second syllable's onset, rescuing it from neutralization. The rule would be that in any domain allowing free syllabification (i.e., a root and its right-appended elements, viz., suffixes), an onset is maximized according to the template discussed in the section on evidence for OM. But the first consonant of a cluster is also ambisyllabically linkedForward to Motivation