Stress and Accent in Sanskrit

As Linguistic or Phonetic Features

1. loudness,
2. length,
3. pitch movement,
4. high pitch,
5. lexical stress that doesn't appear on the surface
6. theoretical marks in theoretical machinery indicating theoretical consequences, some of which may fit some facts known to some, Use your judgement.
7. vowel peripherality (See Veatch 1991, esp Jim and Judy), or tenseness (see Labov, Yaeger and Steiner's work on Philadelphia's accent), or movement (diphthongization e.g. Spanish 'poder' vs 'puedo')
8. etc. etc.

Of these (3) and (4), as in falling or high pitch without other co-occurring features of length and loudness, may be termed "pitch accent", typified by Japanese where mora-timed units in sequence follow (for example, 5 mora) patterns LHLLL, LHHHL or LHHHH, the point of linguistic interest being the presence of a rise from Low to High and then whether it drops again immediately, at the end, or never.. (A mora is a short vowel; two moras is a long vowel or a short vowel plus a consonant ending a syllable).

Many languages have stress predictable by rule, canonically Latin where if the penultimate syllable is long then it is stressed, otherwise the antepenultimate is stressed.

Per Western Sanskritists such as the Goldmans

Stress:

Stress accent, while perhaps not as important in Sanskrit as in English, is usual in proper pronunciation. The general rule is that the penultimate syllable (next to last) receives mild stress if it is as heavy (guru). If the penultimate is light (laghu), the stress falls on the syllable that precedes it (antepenultimate) regardless of its weight. Secondary stress is not usual; all the remaining syllables receive equal stress.

The latter is certainly true, since Panini spends most of the 6th Dhyaaya among the Ashta (8) dhyaayi (books) on the placement of accent. However stress and (pitch) accent need not be entirely related or identical, but orthogonal dimensions of intonational description.

In English for example, stress is partly lexical, partly phrase-structure-influenced, while pitch accents or tunes or melodies may be semantically separate from the content of the sentence (as suggested by the question mark, if it indicates final rise, it doesn't change the words or phrasing in the corresponding sentence which ends with a period). And intonational phrasing is another different but interacting question as well. See Liberman 1974, The Intonational System of English (under Halle, Chomsky and Kiparsky), which is just the beginning of a large literature on English intonational phonology.

If we may distinguish these, then, it appears that if Sanskrit lacks facts or reliable theory in the domain of stress, yet it certainly has lots going on in the domain of accent defined as pitch or tune contrasts.

How do I know this?

Per Others

One: "Extant manuscripts in Sanskrit number over 30 million, one hun- dred times those in Greek and Latin combined, constituting the largest cultural heritage that any civilization has produced prior to the invention of the printing press." (p. V)

But I'm seeking out accent.

In "Formal Structure of Sanskrit Text: Requirements Analysis for a Mechanical Sanskrit Processor" 2007 by Gerard Huet, p190, he says "Knowing the meter of metrical poetry will be a tremendous help to the segmenter. Knowing the accent will permit to recognize bahuvrihi compounds. The enormous Indian tradition of prosody, chandas, has to be exploited for proper computational modeling." To me this suggests there must be a lot more material to study than Google immediately revealed.

I admit at this point, this is a report of an exploration, not an exhaustive and authoritative summary. Still, I have some jewels to reveal. Where, though?

In the same 2007 book on the Computer Science of Sanskrit Goyal et al, "Computer Simulation of Ashtadhyaayı: Some Insights", say: "The sixth, seventh and eighth adhyaayas deal with the suutras that bring about a series of transformations related to continuous text (samhita ̄), word accent, and stem shape." We must aim our ship toward PaNini.

This exploration will share what I know about Sanskrit accent from Saussure, Liberman, Halle, Kiparsky, and PaNini.

Per Halle

Per Saussure

Forgive this interesting aside, close to but not quite about Sanskrit accent.

Pitch accent in Proto Indo European daughter languages has, strangely but actually, been the prototypical focus of study of the great phonologists of history.

Ferdinand De Saussure, who defined the structuralist method on which he built Linguistics as (may we call it) the Science it is today, used IE accent as his basic justification.

In particular, he worked on a discovery by Friedrich Kurschat, a Lithuanian nationalist and linguist and folklorist and journalist who wrote a 476 page grammar of Lithuanian in which he came up with the stress marks still used in Lithuanian studies of accent.

Kurschat had found that Lithuanian il/ir/in ending words with Sanskrit i:l/il, i:r/ir, i:n/in ending cognates happened to have a particular accent in Lithuanian where the Sanskrit vowel was long, and a different accent where the Sanskrit vowel was short.

Saussure, 51 years younger, having read Kurschat's grammar and gone to meet him, actually travelled for weeks with Kurschat in Lithuania in the 1870's to figure these things out, only to later assert that without actually being oneself a native speaker there can be no hope in discovering the structure of anything in that language through asking informants.

Still the pair of lists of cognate pairs got him inspired and obsessed, and at 21 he published:

de Saussure, Ferdinand (1879). Memoire sur le systeme primitif des voyelles dans les langues indo-europeennes. Leipzig, DE: Vieweg

To gloss over many details, and this may be speculation, but it fits the outline of events: he decided there must be some kind of a something behind this difference in Sanskrit which makes a difference in Lithuanian. And he said, in the end, it doesn't matter what that thing is, which makes that difference. All you need is a difference that makes a difference, then you have a something that is real, it is real because it makes a difference. This is the basic structuralist theory of categories, and is the foundation of linguistic epistemology. A linguist can in fact be defined as a person who is unable to think outside the structuralist method.

A method in mathematical statistic with the same properties as this is called Fisher's Exact Test. But Fisher's test gives probabilities to even small differences that might not make much of a difference, whereas in linguistics we like contrasts for which the evidence in principle is trillions of instances, such as /p/ vs /b/ in English, and where the four celled table has actual zeroes in one diagonal (and trillions on the other diagonal), so that statistical reasoning is quite superfluous.

So whatever kind of a thing it might have been in Proto Indo European, that mystery thing which came down to us as vowel length in Sanskrit but as a kind of pitch accent in Lithuanian, Saussure thought it might be an ö, later others called it Laryngeal (which means nothing since all sounds in human languages are laryngeal including laryngeally voiced and laryngeally unvoiced, which means all of them), and Wikipedia today enumerates three h's, h₁, h₂, and h₃, for different subgroups of them, and various theories have up to 8 of them, some might be velars like [k] or other consonantal or vocalic or timing features.

The kicker came when Kurylowicz realized that a certain mystery Hittite letter found in enscriptions in Turkey in the 1900's corresponded exactly to that weird claim by Saussure years before of an ö or something in proto Indo European. After that, although it took a generation, acceptance of Saussure's theory was inevitable.

But even now PIE accent or stress or ö's or whatever it is, is still partly behind the curtain of mystery, and that's Saussure's point: it doesn't matter what thing it is, what matters is that it makes a difference, and so it (its presence vs absence) is a difference that makes a difference, what we call a Contrast, which is the basic conceptual unit of Language. Go forth, then, and discover a forest of Contrasts, group them into some system, and that's called Theoretical Linguistics.

Ready, set, Go!

Funny about Lithuanian accent, there it goes again. And here we are again.

My point, here, is to let Halle give us a sense of what modern theoretical linguistics does in this domain, and also how he characterizes Sanskrit accent to his certain level of approximation. For this purpose, I here extract everything he said about Sanskrit stress or accent, ignoring the proto language and the other Slavic and Baltic daughter languages. Halle claims that Sanskrit has lexically marked accent, using a rule system shared with other Indo-European languages which stayed mostly in place even as other IE languages lost lexically marked stress.

He asserted that IE languages are of two types: they may have moveable or non-moveable stress. The latter, like Sanskrit, mark stress on morphemes in the lexicon, you have to memorize them all. The former, like Latin, evolved from the latter by forgetting all of them, and using initial or final, penultimate, antepenultimate stress in simple rules.

To handle the moveable stress case Halle proposes a computational system like this:

1. Parallel to the segmental plane of phonology which carries the features of the sounds in a text, he uses a stress/accent "plane": an abstract grid of discrete places for discrete objects, written as asterisks, that mark each levels of stress/accent in association with the aligned sounds/segments
2. Mark all the stress-bearing elements with an asterisk on "line 0", and
3. Use inserted parentheses and sentence (phrase?) edges to bound intonational constituents
4. Use a rule system to add further asterisks to lines 1 2 etc. to figure out relative stress and surface stress.

1. There can be lexical accents: mark with left paren, (, before the leftmost.
2. Edge-mark RRR (right paren to the right of the rightmost element) on Line 0.
3. Head-mark L (leftmost asterisk in the constituent or foot) w.r.t. Line 0 on Line 1.
4. Edge-mark LLL (left paren to the left of the leftmost element) on Line 1.
5. Head-mark L (leftmost asterisk in the constituent or foot) w.r.t. Line 1 on Line 2.
6. Head gets HIGH tone, remainder get LOW tone.

After 1 and 2, the lexical accents define FEET with the lexical accent in first position.

After 3 that accent becomes the HEAD of its foot (another asterisk on a higher level)

After 4, 5, 6, a sequence of feet has the first (leftmost) head as the head of them all, and that gets HIGH, otherwise low.

Quoting Halle (1997:291):

"This brief overview of Sanskrit stress is based on the data in the grammars of Macdonell 1975 and Whitney 1941. My analysis is indebted to Sarma 1994. In its basic outline the Sanskrit accentual system is quite like that of Russian and Serbo Croatian; that is, stress is computed by the core rules in 10. Like the latter two languages, Sanskrit distinguishes between accented and unaccented morphemes. It differs from them in that Sanskrit has no postaccenting stems... It was shown by Illic-Svityc 1963 that the postaccenting paradigm is a special development of Slavic that has no counterpart in the other branches of Indo-European....

Except for those of the vocative, case endings in Sanskrit are inherently accented. Some of the accented case endings trigger the Sanskrit analog of the stress retraction rule (17) [which states, mark the lexical accent on a stem before certain suffixes with LLL, thereby retracting stress from the suffix to the stem,] and place an accent on the final stem syllable. The endings triggering retraction are those of the so-called strong cases: the nominative and accusative of both singular and dual, and the nominative of the plural. The fact that in the vocative, stress is on the initial syllable of both accented and unaccented stems will be captured by positing that the vocative desinence [remainder] is unaccented and that like other 'dominant' suffixes it triggers a rule that deaccents stems (cf Kiparsky 1982). In sum there are two kinds of stems in Sanskrit--accented and unaccented-- and three kinds of case endings--accented, preaccenting, and the unaccented vocative, which is also deaccenting. The six possible combinations of stem and case endings are illustrated in Table 4:

Table 4. Sanskrit stems and case endings.

	ACCENTED STEM		UNACCENTED STEM		ENDING TYPE
	'wind'	'sister'	'daughter'	'foot'
SG. DAT	marút-e	svásr-e	duhitr-é	pad-é	accented
SG. ACC	marút-am	svásār-am	duhitā´r-am	pā´d-am	accented, preaccenting
SG. VOC	márut	svásar	dúhitar	pād	unaccented, deaccenting

With the exception of the vocative, the accented stems svásar and marút have stress on the same syllable in all forms, as would be expected given the rules in 10. I have cited two accented stems in ETable 4 in order to illustrate the fact that stem accent in Sanskrit is not restricted to the stem initial syllable. The vocative singular form of the noun marút, whose lexical accent is stem final, has initial stress just like the inherently unaccented stem duhitar. Formally this is accounted for, as suggested above, by positing that the vocative singular suffix, which happens to be phonetically NULL, triggers a rule that deaccents the noun stem. The rules in 10 then assign initial stress.

Consider next the accusative singular forms duhitā´r-am and pā´d-am. Since the stem is unaccented and the case ending is accentend we should have expected stress on the ending, not on the presuffixal syllable.

An analogous situation arose in Russian, cf 15, where I posited the accent retraction rule 17. A similar rule must be posited also for Sanskrit. This rule differs from the Russian rule in the contexts where retraction applies; in their effects on stress location, the two rules are identical.

Notice in Table 4 that the stems duhitar and svasar lose their last vowel before 'weak' case endings: case endings that are accented but do not trigger accent retraction. This vowel loss is a Sanskrit example of the 'zero' grade ablaut, which deletes unaccented short a if the next syllable is accented. A more formal statement of the zero grade rule is given in 43.

(43) /a/ -> 0 / _____ ... V
                  |       |
		  *      (*    Line 0
		          *    Line 1

The zero-grade rule (43) is sensitive to the abstract accentual properties of the sequence rather than to its concrete stress contour. As shown by the forms duhitr-é vs. svásr-e, unaccented a in the stem final syllable is deleted before the accented dative singular ending -e, regardless of whether or not -e is stressed on the surface. The -a is not deleted in the stem-final syllables of svásár-am, duhitā´r-am because it is accented by the Sanskrit counterpart of rule 17 and protected thereby from deletion.

In sum, Sanskrit is subject to the rules in 10 plus its own version of the retraction rule 17. Sanskrit, moreover, has a rule deleting stem accent in the vocative singular and the zero grade rule (43). Sanskrit differs from Russian--and other Slavic languages-- in that from an accentual point of view it has two rather than three classes of stems: accented like márut, svásar and unaccented like duhitar, but no counterparts of the Slavic postaccenting stems.

I can summarize Halle for our purposes by repeating his point that in Sanskrit, accent is lexical. That is know the morpheme, know the accent. Another way to characterize is it, Halle thinks there are no rules for Sanskrit accent. Lexicality means arbitrariness. It means he has no idea what the rules are, if there are rules.

So he has given really the vaguest possible approximation to what we know in detail about Sanskrit accent, for which hundreds of rules are given in the 8 books of PaNini. If a rule is in PaNini governing placement of an accent, Halle has here classified it as lexical: rule-less, non-generalizeable, arbitrary. Lets keep that in mind as we find out more.

Per Kiparski

Kiparsky 2007 starts us out with an example:

"A descendant of Upagu is called Aupagava ́, formed by the suffix -aN, phonologically -a, with a diacritic N which causes vrddhi strengthening of the stem’s initial syllable; general rules accent the suffix, and truncate the stem-final -a before it. With the other patronymic suffixes, -aN constitutes a subclass of taddhita suffixes (‘secondary’ denominal derivational suffixes), which share a number of properties: they are added to nominal stems (prātipadikas), and they are optional, in the sense that there is a synonymous analytic expression: Aupagava ́ is synonymous with Upagor apatyam ‘Upagu’s descendant’.

Taddhita suffixes in turn are a subclass of suffixes (pratyaya), which share some more general properties, such as being placed after the base, and being accented (in the default case)."

"[Shared properties are not repeated for each suffixation rule; they are stated just once in a heading with the appropriate scope. Suffixes — essentially the items introduced in books 3–5 of the grammar — are governed by the headings (10) and (11).

(10) 3.1.1 pratyayah
     suffix-Nom
     ‘(an item introduced in the rules up to the end of 5) is (termed) pratyaya “suffix”’
(11) 3.1.2 paraS ca
     following-Nom and
     ‘and (an item introduced in the rules up to the end of 5) follows’
(12) 3.1.3 aadyudaattaS ca
     initial-accent-Nom and
     ‘and has initial accent’

Rule [12] causes suffixed forms to be accented on the first syllable of their (last added) suffix.

This is merely the default case which is realized if none of the more specific accent rules is applicable. There are many classes of suffixes with special accentual properties and many special rules for the accentuation of compounds and other derived words, which have priority over [12]. (p39)

Per PaNini

At https://sanskritdictionary.com/panini, I searched under the keyword "accent" which pulled up the following verses in Book 6:

6.1.{121,158-181,183-223}
6.2.{1-199}
6.4.{36,72,73,75,127}

1.2.{29-40}
1.3.11
2.1.2
3.1.{1-4}
3.3.{94,96}
6.2.{79,80}
6.2.{1,18,19,195,199}
6.4.{71-75}
7.1.75
8.1.{1-3,16-20}
8.2.{82-88}
8.4.{66-68} which concludes the work.

Is Halle right, then, that accent is lexical in Sanskrit?

Is Halle right, then, that being clear and disproveable is better than being unclear?

For dessert, after 6.2.80 I derive PaNini's name (search Google, I couldn't find the source) in an (accented) Bahuvrihi compound.

Bibliography

Sanskrit Computational Linguistics, 2009. Springer-Verlag. Lecture Notes in Computer Science 5402. Huet Kulkarni and Scharf, eds., First and Second International Symposia Rocquencourt, France, October 29-31, 2007 Providence, RI, USA, May 15-17, 2008 Revised Selected and Invited Papers. W. Sidney Allen,1953. Phonetics in ancient India. Oxford, OUP.

Paul Kiparsky, 1982. The lexical phonology of Vedic accent. Cambridge MA MIT (MS only).

Paul Kiparsky and Morris Halle, 1977. Towards a reconstruction of the Indo-European accent. Studies in stress and accent, ed. by Larry Hyman, 209-38. So Cal Occasional Papers in Linguistics 1. USC.

Macdonell, Arthur Anthony 1975. Vedic grammar. Delhi Bhartiya Publishing House.

Sarma, Vaijayanti. 1994. Accent and ablaut in Vedic. Cambridge MA MIT (MS)

Saussure, Ferdinand de [1894] 1922. A propos de l'accentuation lituanienne. Recueil des publications scientifiques. 490-510. Paris: Payot.

Whitney, William Dwitch. 1941. Sanskrit grammar. Cambridge MA Harvard U Press.