1. Some of these terms have been used in conflicting ways. `Ambisyllabic' has occasionally been applied, and quite reasonably so, to describe any consonant shared between two syllables (e.g., Abu-Salim 1982:23), but I here follow the more conventional practice of restricting its use to short segments. `Geminate' shall refer only to the case where a segment shared between two syllables is long. I shall use the cover term `bisyllabic' to refer indifferently to either phenomenon. Although the term `interlude' is most commonly applied to ambisyllabic consonants, I shall here employ it to refer to all intervocalic sequences of one or more consonants, regardless of their syllabification.
2. The approximants, traditionally called semivowels, comprise the liquids r, l and the glides y, v, as well as the derived l_, l~, y~ and v~. Among the sonorant consonants, they exclude only the nasals, m, n, n. and derived n", n~ and m.. The stops comprise the velars k, kh, g, and gh; the palatals c, ch, j, and jh; the retroflexes t., t.h, d., d.h; the dentals t, th, d, and dh; and the labials p, ph, b, and bh. The fricatives include the sibilants s, s" and s., as well as the unarticulated (glottal) h and the derived sounds h., f and x.
3. For ease of comparison with virtually everything else written about Sanskrit, I use (an ASCIIfied version of) standard Indological transcription. The acute accent on nuclei shows high pitch; this is placed after letters when typographically necessary. Vowels are short unless marked by a circumflex, except that e and o are always long. Vowels have values as in IPA, except that short a is a schwa. Dots after a letter show that consonants are retroflex, except in the case of m. where it shows lack of oral articulation, and h., where it shows that the [h] is voiceless (the normal h is breathy-voiced). h after a stop is a sign of aspiration, not a separate segment. A dot after a liquid shows that it is syllabic. l_ is a retroflex lateral. Tildes show that the sound is nasal, except that n~ marks a palatal nasal. s" is a palatal fricative, n" is a velar nasal. v is a labiodental approximant at the beginning of a syllable, else [w]; phonologically, it is best treated as a glide.
4. In general I will use `word-final' as a cover term embracing also the boundaries between stems in compounds, and between prefix and stem, since most external sandhi changes happen in this environment as well as between full words.
5. For Vedic, these include nasalization of stops before nasals; reduction of ay to a before a vowel; change of ns to nn or m.r before a vowel; assimilation of m to glides and l; and disarticulation of m before r. In addition, the processes of dental assimilation to coronal stops and the defricativization of s" or h after a stop only happen to occur at word boundaries and so are compatible with either theory. I know of no process that occurs only at the end of the word in the broad sense here employed (i.e., also including the end of prefixes and compounding stems), but it is the case that r desonorizes, becoming eligible for syllable-final obstruent-like assimilations, only at the end of a word proper. Crucially, the change that is conditioned by word juncture in this case is a neutralization, not an assimilation.
6. I have taken the glides y and v of the R.gveda text at face value, even though in many cases the metre depends on their being syllabic. The glide and the vowel often alternate for the same word ad libitem, and the complexity of the resultant onset cluster does not seem to be a determining factor.
7. Cho 1990:201-204 makes the appealing hypothesis that v should be treated as an obstruent, because of this distribution and the fact that it is labiodental. But in every other respect it patterns with the glide y as the nonsyllabic form of u. Besides, the labiodental pronunciation is an innovation over Proto-Indo-European; calling v an obstruent would not explain why w could precede segments like r in other Indo-European languages, leaving traces even in the English writing system.
8. As for h, Clements (1990:322) notes that laryngeals fall outside of the sonority scale, so that cross-linguistically they quite often pattern with obstruents, particularly if that is their historical origin. In Sanskrit, h derives from voiced aspirated stops.
9. The only exception is that the unarticulated (laryngeal) sounds m. and h. are represented as diacritics on the preceding bundle, probably reflecting a phonetic analysis of them as being nasal or voiceless continuations of the preceding vowel.
10. In contrast, the postlexical rule of liaison, which gives a coda consonant to the immediately following vowel anywhere within the half-verse, is a movement rule. The consonant remains in the coda long enough to undergo neutralization and assimilation, and yet after liaison, it does not count toward making the syllable heavy: tát a-stu `let that be' -> tád a-stu -> tá-da-stu (RV 1.30.12)
11. There is no direct evidence that Vedic Sanskrit had stress at all, although grammarians describe a Latin-like stress system for the later Classical language. At any rate, there is no correlation between ambisyllabicity and either that stress system or the Vedic pitch accent.
12. Aspirates are apparently an exception, since in texts they virtually never appear in a coda (in the Rigveda they do so twice, in geminates). In gemination, the first element is replaced by the corrsponding unaspirated sound. This may simply be an acknowledgement that aspiration is noticeable only when a stop is released, but the first element of a geminate is not released. Hermann 1923:110-123).