Podcast episode 53: Paul Kiparsky on Pāṇini

In this interview, Paul Kiparsky introduces us to the ancient Indian grammarian Pāṇini and the philosophical significance of his grammatical description of Sanskrit.

Download | Spotify | Apple Podcasts | YouTube

Feature image: Pāṇini stamp, India (2004). Wikimedia Commons

References for Episode 53

Bloomfield, Leonard. Tagalog Texts with Grammatical Analysis, 1917. https://www.gutenberg.org/ebooks/50797

Bloomfield, Leonard. The Menominee Language, Yale University Press, 1962.

Kiparsky, Paul. Panini as a Variationist, MIT Press, 1979.

Kiparsky, Paul. “On the architecture of Pāṇini’s grammar.” International Sanskrit Computational Linguistics Symposium. Berlin, Heidelberg: Springer Berlin Heidelberg, 2007.

Kiparsky, Paul.  Panini. In Elan Dresher and Harry v.d. Hulst (eds.), Handbook of the History of Phonology. Oxford University Press, 2022.

Ostler, Nicholas. Case-linking : a theory of case and verb diathesis applied to classical Sanskrit. MIT Dissertation. 1979.

Transcript by Luca Dinu

JMc: Hi, I’m James McElvenny, and you’re listening to the History and Philosophy of the Language Sciences podcast, online at hiphilangsci.net. [00:16] There you can find links and references to all the literature we discuss. [00:21] Today we’re talking to Paul Kiparsky, who’s a professor at Stanford University in California, and Paul’s going to talk to us about the philosophical significance of Panini’s grammar. [00:35] Maybe you could tell us who Panini was and what his grammar was. [00:39]

PK: We actually don’t know very much about Panini at all. [00:44] We think he worked somewhere between 500 and 350 B.C. in the Northwest, what is now Pakistan, and we know his mother’s name, and that’s just about all we know about him. [01:00] [chuckles] [01:02] So we know a lot about the grammar, of course. [01:06] It’s a system of 4,000 rules, very condensed so that you can print it on 30 pages, and it’s a generative grammar of Sanskrit. [01:19] It’s about the high standard of his day that is comparable to something like Classical Arabic in the modern world, the formal level, but it was an actual spoken language, because the grammar includes lots of information about how to, for example, address a shudra, a member of the lowest caste, with what intonation and so forth, so showing that it was a language that was spoken. [01:56] It also deals selectively, not completely, with the Vedic, earlier form of the language and treats it as a kind of aberrant system. [02:08] You know, it’s not based on Vedic, it’s based on the norm of classical Sanskrit. [02:16]

JMc: So we don’t know very much about who Panini was. [02:19] Do we know that he was an actual person, or is he more a sort of legendary figure, like a character who’s invented to be the author of the grammar? [02:29]

PK: Well, there must have been someone. We know that there were many other grammarians, because Panini quotes a number of them. [02:42] I think seven or eight of them are at least mentioned, and credited with this or that observation. [02:48] They all seem to have had the same basic approach. [02:52] There’s no difference that we can detect in theory or method, but sometimes they disagree about the data, naturally, the dialectal differences and so on. [03:05] And we, I think, can suppose that it’s kind of like Homer. [03:11] He was the last one in a series of scholars, in this case, who contributed to this enormous work. [03:19] It’s nothing that one person could do in their lifetime, and so I think he was the last in the line and the greatest. [03:29]

JMc: OK. So apropos of Homer, you said that the whole grammar can be printed on about 30 pages, so was it originally meant to be written, or is it something that was memorized and only later committed to writing? [03:44]

PK: No one exactly knows, and actually there’s disputes about the extent to which writing, if at all, was used. [03:53] I personally, for whatever it’s worth, think that both were used. [03:59] That’s certainly the practice in modern Panini scholarship among the traditional pandits. [04:06] They will do all of their computation — that is, all the derivations — essentially in their head, so they can give them a word and they will recite the derivation, 20-30 rules, in order, without using any books or anything. [04:26] And they will own a few books. [04:28] For example, they will have an edition of the Mahabhashya, which is the big commentary, and maybe the Kashika and some other works of interpretation, which nobody memorizes as far as I know, but there’s also pandits who memorize the whole thing, but the ones who simply can recite it from beginning to end in a go are not the people who actually work with the text as a derivational mechanism. [05:03] The people who do work with it that way just know all the rules; they can’t necessarily recite them all in order. [05:10] And so I suspect that that’s how it’s been all the way from the start. [05:15] It’s hard to imagine fixing the order of 4,000 rules just orally. [05:23] It had to take some kind of notation to fix it, it seems to me. [05:28]

JMc: So the grammar is still used today by pandits? [05:32]

PK: Well, it’s used in attenuated form. [05:36] It’s become the basis for a lot of practical Sanskrit grammar, school grammar, and people still study it from that perspective, but as a formal grammar, it’s nothing that you could use practically. [05:54] It’s far too complex to use as a vehicle for learning Sanskrit, for example, so essentially, you already have to know the language to a certain extent, at least to be able to understand what’s going on there. [06:14]

JMc: So what is its purpose? Is it meant to be sort of normative, or is it just meant to be a thing of beauty? [06:20]

PK: Both. It’s certainly normative. [06:26] That is, I think, it was originally, and certainly continues to be, normative in that it defines the correct form of classical Sanskrit, but it was also intended to be a record of the actual speech, which was, at that point, perhaps felt as being threatened by the vernaculars. [06:52] The Prakrits had already emerged when the Aṣṭādhyāyī was composed, so the Buddha, who is roughly, very roughly, a contemporary, before Panini, was writing in a much more vernacular form of Sanskrit. [07:13] I mean, that was part of the political content of Buddhism, right? [07:18] It was more popular, but at that time, the classical Sanskrit described by Panini was clearly still in use among very educated people. [07:30]

JMc: So it was essentially, I mean, it was a conservative project, both in the sense of conserving or preserving an ancient high variety, but also in that contemporary political context of preserving ancient learning in the face of a popular religion, namely Buddhism. [07:49] Is that… Have I understood that correctly? [07:51]

PK: Yeah, I don’t know about ancient learning, about perhaps preserving the, maybe sociolinguistically preserving the Brahmin identity in some way, but much more important it is that it’s a scientific venture, that is, it goes very deeply into the structure of the language in a way that linguists can still… have always learned from, at least since the early 19th century, and still continue to learn from. [08:23]

JMc: Yeah. So the 19th century, which is the period in which modern academic linguistics in the Western world really got started. [08:32] So would you say that Panini’s grammar influenced the development of Western linguistics? [08:39]

PK: I’d say in very fundamental ways. [08:43] So if you go back to the early 19th century, the linguists like Bopp, who founded comparative Indo-European grammar, learned from him about morphology. [08:58] So, until then, people thought in terms of whole words, so in Latin grammar, you have, for example, servus ‘servant’, is cited in the nominative singular, and then you do something called inflection to it. [09:20] You change the endings, so servi, servo, servorum, etc., and these things are understood as processes that you apply to the base form, which is an actual case form, the nominative singular. [09:39] Panini introduced the idea that there was a stem, and there were some endings, and that you form the different case forms by adding to the same stem different case endings, and then maybe doing some phonology on the result of that composition, and output is then the actual word. [10:01] And the reason this was important is that it enabled Bopp to get started with comparative grammar, because it’s those things, the stems and the endings, that are reconstructible. [10:16] Actual words, you can reconstruct post hoc, but those are not the entities on which you do your comparative work. [10:25] You work with roots, you work with affixes, stems, etc., etc., and until that insight came along, there was just some kind of global idea, like you find in William Jones and so on, that “Well, this word looks like that word, and they must be related.” [10:45] And also, the other thing that was a big contribution was the idea of phonological structure, so classes of sounds, for example. [11:00] In fact, the classification of sounds was actually much better than for most of the linguistics that was to follow. [11:15] So, for example, vowels and consonants were classified together by the same feature system, so you had palatal vowels, and palatal consonants, and velar vowels, and velar consonants. [11:32] That’s the system that was abandoned by the British phoneticians. [11:37] They decided to classify vowels in one way and consonants in another way, which introduced all kinds of problems, and it was not until Jakobson that linguists reverted to a unified feature system. [11:54] And interestingly, Uralic linguistics never abandoned this idea. [12:01] I mean, they still have, in works on Uralic languages, they talk about palatal vowels, and velar vowels, and all of that, and it’s actually a very good way of dealing with vowel harmony and all of those things that are prominent in those languages. [12:16] But apart from that basic feature, even the idea of having a classification of sounds was something that basically came into linguistics at that time through Indian science. [12:34]

JMc: So, the topic that you wanted to talk about was the philosophical significance of Panini’s grammar, so what would you say the philosophical significance of Panini’s grammar is? [12:46]

PK: So, let me preface it by saying that — and this is partly a continuation of my answer to your previous question — there are many other insights that came out of Panini that anticipate the, or were instrumental in introducing these ideas into Western linguistics. [13:12] So, the idea of a modular organization of language. [13:17] So, you go to Bloomfield and his work, not only on the Native American languages like Menominee, but also his Tagalog grammar, which nobody reads, have this very interesting organization. [13:33] There’s a kind of level order system. [13:37] There’s primary endings and secondary endings, and they each have their characteristic phonology associated with them, and they work differently. [13:49] The primary endings go on roots, the secondary endings go on stems, and that’s all a very sophisticated system. [13:57] That’s straight from Panini, and Bloomfield was very well acquainted with Panini. [14:02] He actually wrote an interesting technical article on that. [14:07] There’s thematic roles and the idea of, if you like, a level of representation or a set of categories that come between semantics and the expression in the morphosyntax of grammatical relations, so case and agreement, and stuff like that, in a many-many relationship to both. [14:30] So, a thematic role can correspond to different semantic categories, and it can also correspond to different cases, for example. [14:41] So, that was introduced into modern linguistics in the ’60s. No, ’70s, I would say. [14:50] So, my student Nick Ostler at MIT wrote a dissertation that was very influential, where he had this, and that was inspired by Panini. [15:03] OK, so why am I telling you this? Because the remarkable thing is that Panini had no theory at all, and that’s the theoretically interesting thing. [15:15] All of these very sophisticated categories, conventions, devices, and so forth, which we admire, emerge from only two fundamental things. [15:28] One is, we’re going to have a complete description, and the second is, the total description has to be as simple as possible. [15:38] If you put those things together, they force you to introduce these devices to abbreviate the description. [15:48] So, these are not just handy conventions or anything like that. [15:55] They have a very serious purpose here. [15:58] If you removed any of them, the grammar would become longer. [16:01] It couldn’t fit into 30 pages anymore. [16:05] In fact, for parts of the grammar, you can actually prove this mathematically. [16:11] For example, the Shiva Sutras, which have the phonological classes, and define the phonological classes, they are, in a very particular sense, perfect. [16:23] So, that, I think, is really philosophically interesting, because it provides an answer to an issue that’s being discussed a lot, which is actually at the basis of, for example, the generative grammar project, which is, how is it possible to learn a language, and what do all languages have in common, and why? [16:46] And the answer is very different than what you get in generative grammar. [16:51] It also involves simplicity, but in a totally different way. [16:55] So, what I learned as a graduate student is that you have a rich, innate grammatical formalism, the theory of grammar, and then you have a simplicity principle, and you apply the formal system to the data and always choose the simplest description. [17:19] Well, what Panini has is, he says you don’t have anything except completeness and simplicity, and the structure emerges, right? [17:32] I mean, it seems to me that’s a very exciting answer. [17:35]

JMc: But is it stated explicitly that- [17:39]

PK: Oh, nothing is stated. [17:41] Nothing is stated. [17:42] We just have the rules. [17:43] There’s no theory. [17:45] And that’s a problem. [17:48] We don’t even have all of the principles by which those rules and their interaction is interpreted. [17:58] We don’t have those because, evidently, there must have been a break in the tradition at some point. [18:07] So, you find that even the earliest commentators, which are Katyayana, who writes a bunch of comments, and then Patanjali, who has a more discursive discussion, pro and con of different interpretations, they don’t even know, in many cases, exactly how the grammar works. [18:31] And figuring that out has been a project of modern Panini studies, and that’s been one of the things I’ve been interested in. [18:44] I actually wrote a book that’s called Pāṇini as a Variationist. [18:51] It turns out he uses three words to indicate that the rule is optional, and there’s a ton of optional rules, maybe 30% or so of the rules are actually optional. [19:03] So, you can add this suffix to express a certain meaning, or that other suffix also. [19:11] And he uses three words — vā, vibhāṣā, and anyatarasyām — and what I discovered is that vā actually means ‘preferably’, vibhāṣā means ‘marginally’, and anyatarasyām means ‘either way’, and I proved this by collating not only his own usage, but the usage of the texts and the usage of the Vedas with respect to the Vedic rules that he formulates to show that that fits exactly with what the text shows. [19:54] And so, that was, I think, the first clear proof that there was content in the grammar that the tradition didn’t know about, because there must have been this break. [20:10] So, that’s why it’s important that the grammar was written down, because it’s the only way in which it could be recovered, and the study of it taken up again after this disruption, and that led me in work that was done together with S.D. Joshi, who was a Panini scholar from Pune, where I spent some years, to question the principles by which rules interact, and so we formulated, I think, a much simpler and better understanding of how that works, and that’s been quite controversial, because there are many traditional people who don’t like that. [21:06] On the other hand, I think we have convinced a lot of people, so the discussion continues. [21:13] But anyway, that’s why this is a continuing quest, and we don’t have a finished system that we can just apply and talk about. [21:22] We have to understand first what that system is like. [21:26]

JMc: Do the traditionalists have an alternative interpretation? [21:30]

PK: Well, there is a traditional interpretation, which is… Well, within it, there are many different opinions, also minor disagreements, but there is broadly a traditional consensus, and what we offer is a new one. [21:53] And there’s actually a recent work that’s just coming out, I think it’s Oxford University Press, by Rishi Rajpopat, who has a still different view, so it’s been very controversial, and I think it’s quite interesting, but ultimately not correct. [22:15]

JMc: Empirically not correct, or? [22:17]

PK: It’s just not, it doesn’t handle the… It doesn’t give you the right results, that’s a term. [22:23] I mean, if you like that, that’s a way of saying it’s not empirically correct. [22:28] These are principles about how to decide, for example, if two rules are applicable, which one do you choose? [22:36] And because only a very small part of the grammar shows you the order in which rules are to be applied, most of them just have the rules swimming around there, and you throw in a form, a derivation, and the rules all have a whack at it, and you have to decide which one is prioritized at any point in the derivation, and we have a very simple answer to that. [23:02]

JMc: At a couple of points, you’ve described Panini’s grammar as a generative grammar, and you’ve also mentioned how Panini’s grammar has inspired many developments in generative theory since the ’60s and ’70s, but do you think it’s anachronistic at all to project our modern notion of generative grammar back onto Panini? [23:27]

PK: Well, I’m just using generative grammar in a broad sense, that it’s not restricted to a corpus, but it goes beyond the data, and that it essentially provides you an analysis that relates meaning to sound. [23:45] There’s four levels. [23:48] Again, these emerge from the basic idea. [23:54] You have the simplest description. [23:57] You are forced to modularize the system, and so you have essentially input. [24:03] You can think of it as an input. [24:06] You can think of it as just all being simultaneous if you like. [24:10] It doesn’t matter. [24:11] But there’s on one end the semantics, and there’s a very regimented way in which you talk about tense and mood and all of those categories and about the meaning that the cases express. [24:28] For example, you know, separation: if there is a fixed point, then the removal from that is defined as a separation. [24:40] Then you use that to characterize, for example, a certain theta role, which is, let’s say, source. [24:51] And then you use that theta role, the source, to characterize, among other things, the ablative case, but each of those things has many functions. [25:04] That’s just the most straightforward mapping between them. [25:08] So you have semantics, you have thematic roles, and you have sort of abstract tense, if you like, abstract tense, and then you have case agreement, and you have the actual morphological tenses, and then you have morphophonology, and that is then mapped into phonology. [25:31] And following on that and not dealt with in the grammar is phonetics. [25:37] So there were other works which take the output of the grammar, or if you like, let’s not talk input and output. [25:48] Let’s just say it’s correspondence between semantics and phonology. [25:52] They take the phonology and map it into phonetics. [25:58] So this is interesting. [25:59] Again, the entire edifice of phonemes emerges just from these basic principles. [26:11] There’s no idea of complementary distribution, or minimal pairs, or distinctiveness, or whatnot. [26:21] I mean, all of this theory that goes into phonology and phonemics in Western linguistics isn’t even there. [26:30] But still, you get phonemes, which are exactly… It’s the smallest number of categories that you need in order to characterize the lexicon and the rules, [26:46] and you don’t need for that, for example, the visarga, which is the, you know, the H with a dot under it (it’s a form of S that you get at the ends of words), the anusvara, which is the nasal off-glide that you get in certain combinations, because those are all derivable by phonetic rules. [27:14] So they’re not part of the Shiva Sutras. [27:17] It’s an example of emergent structure, right? [27:23] It comes out from nothing. [27:25] I think it’s an amazing fact. [27:28]

JMc: So do you think it’s a fact about the structure of human language, or do you think it’s just a contingent result of a particular approach to analysis? [27:38]

PK: Well, you know, this is way beyond anything that I could substantiate, but one might think of the human learner as trying to construct the simplest account of the ambient language. [28:01] There’s a modern theoretical foundation for that. [28:04] It’s minimum description length. [28:07] Originated as a theory of randomness, a Russian mathematician, Kolmogorov, and it’s become pretty respectable philosophical theory, because it’s got a solid mathematics behind. [28:27] So John Goldsmith has written about it. [28:32] Yeah. [28:33] So this is kind of a very interesting basic anticipation of a modern idea that I think holds a lot of promise. [28:44]

JMc: Well, that’s excellent. [28:46] Thanks very much for answering those questions. [28:49]

PK: OK. [28:51]