All the evidence stated thus far has been from internal reconstruction of the R.gveda text. There is also evidence from Indian linguists, early inscriptions, and the structure of daughter languages. All this, however, is very late, perhaps six or seven centuries later than the composition of the core R.gveda ca. 1000 BCE. There is the very real possibility that the language is significantly different by this time: I have already mentioned that Vaux believes that Sanskrit completely changed its syllabification strategy by this time. Sanskrit was still alive as the language of the Brahmanic class, but had evolved signifcantly beyond the Vedic stage. The popular Indic speech went much farther still, such that it is considered to have developed into separate languages, and in particular the phonology was much altered. Therefore it is probably best to consider these data not as evidence for Vedic Sanskrit per se, but as the next step in the evolution of the language. They are evidence insofar as they constitute a plausible development from the state of affairs posited for Vedic.
An immediately encouraging fact is that where my account calls for ambisyllabicity, the grammarians state that the consonant may be doubled, i.e., pronounced long. One might only wonder whether I should have been talking about gemination all along rather than ambisyllabicity. But some of the grammarians said that the lengthening was optional (e.g., Pân.ini 8.4.47, under the scope of vâ in rule 45), and others (e.g. S"âkalya apud Pân.ini 8.4.51) denied the lengthening altogether, so it seems safest to assume that Sanskrit had structural bisyllabicity, which alternated between ambisyllabicity (no lengthening) and gemination (lengthening). Similarly, the earliest inscriptions often, though inconsistently, showed doubling there (Varma 1929:64-67).(14) Finally, the Middle Indic dialects, which evolved from Old Indic dialects similar to the language of the R.gveda, show clear effects of such doubling. A word like atra, for example, evolves into atta. If one takes ambisyllabicity of the t as the base form, with optional doubling, the resultant form atta is easy to understand, given the independent fact that these dialects simplified all tautosyllabic clusters. The change is difficult to motivate otherwise. (The argument that Middle Indic simplifications have an origin in Sanskrit doublings goes back at least to Jacobi 1881:609, and Cho 1990:194-231 and Vaux 1992 have worked to put that in a syllable-based context.) All this evidence indicates that the bisyllabicity of onset clusters after a vowel was a real property of early Indic.
Unfortunately, these doublings are not completely restricted to the contexts where ambisyllabicity was posited. The situation is really very difficult to evaluate. Allen (1953:79) despaired of even discussing the topic. Cho and Vaux have recently made heroic attempts to harmonize those accounts and to explain them in terms of syllable theory, but it must be admitted that many of the pronouncements make broad claims that other sources seem to step on. For example, Pân.ini doubled s before stops; the Taittirîyapraâtis"âkhya (14.9, apud Vaux 1992:286) doubled stops after s. But it is possible to reconstruct core environments that the great majority of the grammarians would agree on, and that are compatible with the evidence from the inscriptions and the daughter language developments. One is that an oral stop doubles between a short vowel and an approximant. This is quite close to my rule for ambisyllabicity. The restriction that the vowel must be short (due to Pân.ini 8.4.52) may mean that the bisyllabicity occurs only when it would actually have a non-vacuous effect on syllable weight. On the other hand, the rule could just be a reflection of the fact that the colloquial dialects resist closing a syllable after a long vowel (Hermann 1923:260).
The other doubling that occurs almost universally is that of a stop after r. Since r plus stop can never make an onset cluster, they would always be heterosyllabic, and the prior syllable therefore always heavy. I therefore know of no internal evidence that would tell whether these sequences are ambisyllabic in Vedic Sanskrit. The fact that Greek does not show this doubling may mean that this particular rule is an Indic innovation (Vaux 1992:288), and so may plausibly be post-Vedic. But if the actual rule is that heavy interludes (not just heavy onsets) want to be ambisyllabic, as suggested above, then this case naturally falls under the scope of that rule. The sequence r plus stop is the only clearly documented coda sequence in the language. Therefore it is a clear situation where a non-onset sequence can be made bisyllabic. In contrast, a sequence of two stops rarely is doubled in the inscriptions, and there is no evidence for that in the daughter languages. Any bisyllabicity there would require admitting them as coda clusters or as onset clusters, neither of which is possible for two stops.
This argumentation is a little too pat, however. One cannot afford to turn away from the cases where there is conflicting evidence. Many sources, for example, provide that nasals geminate before stops (Taittirîyapraâtis"âkhya 14.24 apud Vaux 1992:290) or after sibilants (14.9 apud Vaux p. 286), which are incomprehensible from a syllable-theoretic point of view (nt cannot be an onset, and sm cannot be a coda) and have caused much consternation over the years (Hermann 1923:256, Cho 1990:209-210). Ultimately in some dialects of Sanskrit the rule must have been restructured into a requirement that length be associated with some member of a heavy interlude. If this is the case, then it is difficult to know whether the doubling of stops after r is the result of an ambisyllabicity requirement, or whether it simply reflects the fact that r cannot be doubled, so the stop is doubled in compensation, as it were. At this point, it seems safest to consider the doubling of stops before approximants as good evidence of bisyllabicity, but to consider conclusions about other doublings to be tentative, pending a complete analysis of the Sanskrit doublings and Middle Indic developments.