Hacker News new | past | comments | ask | show | jobs | submit login
Bhūribhirbhāribhirbhīrābhūbhārairabhirebhire (wikipedia.org)
210 points by KhoomeiK on May 17, 2020 | hide | past | favorite | 77 comments

I wrote[1] this Wikipedia article; nice to see it quoted in the wild! Happy to answer questions if any.

One thing I'd like to point out is that far from an extinct form, there are Sanskrit poets even today who can compose verses subject to such constraints, and at least two of them[2] even do it “live” on stage—compose with any of the classical constraints, on any given topic!—without pen and paper, as part of a performance[3] interleaved with many other interruptions.

Edit: About the verse in the title, just now found a Facebook post[4] that points out it's part of at least 10 such two-consonant verses in that chapter (which also has verses subject to various other constraints).


[1]: I was not the original creator (https://en.wikipedia.org/w/index.php?title=Shishupala_Vadha&...), but added most of the content (https://en.wikipedia.org/w/index.php?title=Shishupala_Vadha&...): 90% according to the “Who Wrote That” extension.

[2]: Shatavadhani Ganesh https://en.wikipedia.org/w/index.php?title=Shatavadhani_Gane... and Shankar Rajaraman

[3]: https://en.wikipedia.org/w/index.php?title=Avadhanam&oldid=9... – excerpts from a recording of such a performance are at https://www.youtube.com/watch?v=m3GnorRNjXE though it's best experienced live!

[4]: https://www.facebook.com/sanskritsense/posts/438163406786809

There was a question asking if there exists a recording of the verse mentioned in the title; looks like the post got deleted (https://news.ycombinator.com/item?id=23211366) while I was answering it. Anyway: the Facebook post mentioned above links to their Soundcloud reciting this verse in ominous tones with some modern music. I'm not aware of a more "traditional" recording of the Shishupala Vadha, but there is an earlier work by a different author (Bharavi) whose 15th chapter (canto) has similar constraints (https://en.wikipedia.org/w/index.php?title=Kir%C4%81t%C4%81r... as mentioned in the article), and there's a recording of that here: https://archive.org/details/kirAtArjunIya-mUlam-vedabhoomi.o... (extracted from a recording with commentary here: https://archive.org/details/KiratarjuneeyamByMahakaviBharavi...) — as you listen to the chapter you can easily tell when these two-consonant or one-consonant verses appear :-)

When in India I was told there are even some rare families still raising their children with Sanskrit as their primary language. I think this spoken Sanskrit is however a bit different from the written version which might mirror Latin that might also have had a spoken and written version (and the former is dead).

Small remarks:

* I found the description on the 'drum' pattern difficult to understand, that might be a word understood in a particular way in your culture/region? * The introduction mentions the verse that can mean both apology or declaration of war, would be great to have this mentioned/translated too as one of the examples.

Thanks for an interesting article!

You're welcome!

* For Latin, Wikipedia gives a list of “ages” (Old Latin, Classical Latin, Late Latin, Medieval Latin, Renaissance Latin, New Latin, Contemporary Latin), with "Vulgar Latin" (Latin as spoken by the common non-elite people) and Ecclesiastical Latin (Latin as used in the Church) on the side. The main point is that, when Latin was still a spoken language within living memory, the speech of the people had already started to diverge, and that is Vulgar Latin. But when say Newton or Gauss or Erasmus or Spinoza wrote in Latin, or when George I spoke in Latin to his prime minister Robert Walpole (according to Wikipedia), they were using not Vulgar Latin but "correct" Latin; they had just learned it as part of their education. In a parallel to that, "spoken Sanskrit" of the common people evolved into various Prakrit languages and into many modern Indian languages, just as Latin evolved into the Romance languages. But the families where Sanskrit is spoken (usually as just one language among many) all involve "correct" Sanskrit; it's just that the parents have learned it and teach it to their children.

* Yes the "drum" pattern does not look like a drum to me either; it's just what the constraint is called :-) It's based on the strings around a certain kind of drum (possibly mṛdanga = mridangam); I've uploaded a few illustrations here https://imgur.com/a/8c6Sxet — they're from books/blogs; will try uploading one of them to Wikipedia under a fair-use license.

* About the chapter with two meanings: I didn't have a reference then, but I found one now and have added an example to the article! https://en.wikipedia.org/w/index.php?title=Shishupala_Vadha&...

Thank you for your contributions to wikipedia.

Are you aware of free and open source recording of shlokas that a beginner can hear and learn to recite?

The Padhuka Sahasram of Vedanta Desikan aka KaviTarkika-Simhan has a section called Chitra Paddathi which paints pictures in words in a matrix of artistry in praise of the feet of the Lord. You can hear it being recited at (1) read the shlokas a t (2) and see the artistry at (3) Of note, the solution to the Knight's Tour can be seen at (4)

(1) Recital by Malola Kannan https://www.youtube.com/watch?v=Bj4sWsHC6Ck

(2) Follow the recital starting at shloka 911 here http://www.ibiblio.org/sripedia/ebooks/vdesikan/rps/92.html

(3) See the artistry of the shlokas here http://www.ibiblio.org/sripedia/ebooks/vdesikan/rps/chitra.h...

(4) The Knight's Tour solution here http://www.ibiblio.org/sripedia/ebooks/vdesikan/rps/knight.h...

Sorry, I'm not sure if I understood your question correctly. If you just mean recordings of any shlokas, then you can just type the name of your favourite one into, say, YouTube and find thousands of hours of listening material. Is that all you meant? For example, I listened to the entire Ramayana recited wonderfully clearly (at https://pravachanam.com/pravachanambrowselist2/296/37/230 but you can also find that recording elsewhere now). If you meant specifically a recording of such chitra-kāvya verses, see the other answer you got. Or if your question was specifically about licensing (not sure if that's what you meant by open source), search instead on archive.org; you'll still find lots of results.

I want to learn the shlokas from a priest verse by verse. Since all scriptures are out of copy right, it will be great if the priests who know how to utter the words correctly record it, people who don't access to priest are free to learn it. For example, M. S. Subbalaksmi's Vishnu Sahasranamam is very fast and Brahma is uttered as Bramha or Bramma sometimes. I am not trying to critique her work, I think it is great. It's just that it is very fast for a new student to grasp; a verse by verse reading from a priest would be very useful. Thank you for the resource you suggested, I will take a look.

Ah I see. About whether “Brahma” is pronounced as written or as “Bramha”, there is regional variation and difference of opinion even among the best scholars; recommended is to find a tradition and follow it. Overall M. S. Subbulakshmi's rendering may not be perfect but it's better than most, and if you can match it, it would be great. You can slow the playback if it's too fast; just tried it and at something like 0.7x speed it's still sounds fine while being slow enough. Anyway, you can look for recordings by people trained in Vedic recitation (not exactly same as priest…), like the Challakere brothers. Also, you may benefit from an actual class with feedback from teacher(s), etc; and Vyoma Labs (sanskritfromhome.in) have many "Learn to chant…" courses which are (incredibly!) free. Good luck!

Thank you for commenting! I was looking for a performance somehwere, but couldn't find one.

That’s it, next scrum meeting I’m saying mahakavyas instead of epic.

'Maha kavya' simply means 'Great poem' or 'Large poem' or ... 'Epic poem', where 'kavya' means 'poem'.

So, for your intended use case, I'd suggest saying 'maha' instead of 'mahakavya' since that is closer in semantics to your intention.

maha is just an adjective, epic is both an adjective and a noun. That usage is correct.

Y̶e̶s̶,̶ ̶b̶u̶t̶ ̶w̶h̶e̶n̶ ̶p̶e̶o̶p̶l̶e̶ ̶e̶x̶c̶l̶a̶i̶m̶ ̶'̶E̶p̶i̶c̶!̶'̶ ̶—̶ ̶w̶h̶i̶c̶h̶ ̶i̶s̶ ̶t̶h̶e̶ ̶u̶s̶a̶g̶e̶ ̶I̶'̶d̶ ̶w̶a̶g̶e̶r̶ ̶m̶y̶ ̶p̶a̶r̶e̶n̶t̶ ̶w̶a̶s̶ ̶r̶e̶f̶e̶r̶r̶i̶n̶g̶ ̶t̶o̶ ̶—̶ ̶t̶h̶e̶y̶'̶r̶e̶ ̶u̶s̶i̶n̶g̶ ̶t̶h̶e̶ ̶a̶d̶j̶e̶c̶t̶i̶v̶e̶ ̶f̶o̶r̶m̶ ̶a̶n̶d̶ ̶n̶o̶t̶ ̶t̶h̶e̶ ̶n̶o̶u̶n̶ ̶f̶o̶r̶m̶;̶ ̶i̶.̶e̶.̶,̶ ̶t̶o̶ ̶d̶e̶s̶c̶r̶i̶b̶e̶ ̶t̶h̶e̶ ̶m̶a̶g̶n̶i̶t̶u̶d̶e̶ ̶o̶f̶ ̶a̶w̶e̶s̶o̶m̶e̶n̶e̶s̶s̶ ̶o̶f̶ ̶w̶h̶a̶t̶ ̶t̶h̶e̶y̶'̶v̶e̶ ̶j̶u̶s̶t̶ ̶w̶i̶t̶n̶e̶s̶s̶e̶d̶,̶ ̶a̶n̶d̶ ̶n̶o̶t̶ ̶t̶o̶ ̶c̶a̶l̶l̶ ̶i̶t̶ ̶a̶ ̶'̶g̶r̶e̶a̶t̶ ̶p̶o̶e̶m̶'̶,̶ ̶o̶r̶ ̶a̶n̶y̶ ̶p̶o̶e̶m̶ ̶o̶f̶ ̶a̶n̶y̶ ̶s̶i̶z̶e̶,̶ ̶f̶o̶r̶ ̶t̶h̶a̶t̶ ̶m̶a̶t̶t̶e̶r̶.̶

EDIT: Nvm, I was unfamiliar with the Scrum term.

An "epic" (noun) is a Scrum term for a collection of stories.

I'd say that was the intended usage.

Code can indeed be poetry :-)

This is the closest thing I've seen to Borges' fictional language described in Tlön, Uqbar, Orbis Tertius. The language has no nouns, so words are made of up sets of adjectives.

"moon" becomes "round airy-light on dark" or "pale-orange-of-the-sky".

Because of this structure, it was possible to create poems in the language that consisted of a single long word.


I'm struck with a sense that this poem with a linguistic constraint is a mirror for my life at $dayJob, with some embellishment perhaps...

"We want you to write a poem in an obsolete/mostly dead language. And oh, you may only have two consonants per stanza, no more."


"Write and maintain web-based applications in 2020 but do not stray outside of the list of approved languages (Ruby and PL/SQL), no modern advancements such as containers or CI/CD are supported, also, remote debuggers are right out."

It was a better analogy in my head, before I wrote it down.

I guess a closer analogy is the International Obfuscated C Code Contest (https://www.ioccc.org/): something that requires ingenuity and mastery over the language (Sanskrit or C respectively) and all its dark corners, but is more a dazzling display of cleverness than something appropriate for everyday use. :-)

The problem with this analogy: What you describe was an ordinary job maybe 15-20 years ago. If the analogy held, people would have spoken like this poem in daily life at some point.

They did. Much closer to it anyway. The difference is it was many more than 15 or 20 years.

I think you may be underrating the mastery required to write verses like this. Sure, Sanskrit was (and still is!) spoken as a language. But just because you can speak in Sanskrit does not mean you can come up with constrained verses like this; it requires a lot of skill. (Just as, to use the analogy above, just because you can program in C does not mean you can write a winning IOCCC entry.) The difference between "not speaking Sanskrit" and "speaking idiomatic Sanskrit" is IMO smaller the difference between "speaking idiomatic Sanskrit" and "speaking like this poem in daily life" — the former is “just” a matter of learning a language.

No need to compare to programming. I speak English as a native language but I'm not great at writing poetry. There are poets who will vastly outperform me at the task.

But I speak pretty similarly to the poetry, gramatically, lexically, phonetically, and in a few hundred or thousand years' time nobody will comprehend those written works the same way I do, just as many Shakespearian puns and jokes don't work well with current accents even though we are all speaking "Modern English".

Also, tangentially, I don't know Sanskrit or the modern languages of South Asia at all. However, given what I know about linguistics and other old languages I can guarantee you that no modern speaker of Sanksrit understands it in exactly the same way as the dead authors of those old written works.

I think we're just using different definitions of “closer” or “similar” to “people would have spoken like this poem in daily life”. :-) To me, “grammatically, lexically, phonetically” is the trivial part. Learning English is achievable, and many do it. But no one in the course of their daily speech accidentally starts speaking in palindromes, or single-consonant verses — the distance to it seems much greater. I do see it's just a matter of definitions though, something like measuring lexicographic distance of strings versus measuring K-L divergence of the probability distributions on strings. My point is that it would be staggering if at any point in history people spoke such a verse in daily life, and your point is that they were speaking the same language, so it's “closer” in some sense. Both are valid perspectives, I guess.

Anyway, about the tangent: both modern speakers and the dead author of this old written work are from centuries after something like Sanskrit was the spoken language of the street, and both learned Sanskrit by reading many of the same authors and aspiring to adhere to the same (very comprehensive) grammar, so the difference is not as much as one may think: I cannot enumerate any point of difference as far as the language goes (grammatically, phonetically, etc), except for things like quotations from later centuries being more familiar to later authors. This is like: Gauss was writing in Latin about three centuries after Copernicus (say), but both of them understood Latin in pretty much the same way AFAIK. (In the case of Latin the phonology was questionable maybe, but with Sanskrit such things were better documented.) No two people ever understand a language in exactly the same way of course, but when learning a fixed version of a classical language, the differences are relatively minor.

Latin was a dead language when both Gauss and Copernicus were alive. Even within the time of the Roman empire, it went through crazy amounts of changes.

For example it's pretty common to find people who can read Latin well but pretty rare to find people who understand in what time period and circumstance an /m/ would have nasalized the vowel before it. Could they "speak Latin"? It wouldn't really be the same. It's a bit like learning Esperanto.

I think it's quite likely that modern communities speaking Sanskrit are a little blinded to this sort of thing by pride in the cultural heritage and identity that Sanskrit represents.

My larger point is we don't really know what kind of utterance a casual street speaker of Latin or Sanksrit was capable of. It's very possible that something like a palindromic poem is impressive to modern eyes and ears in part because native speakers no longer exist.

Another example in English. You might be surprised at how people without much education can improvise a freestyle rap. How will that look to somebody studying current English in 1000 years? Will someone make comments like yours about the astounding difficulty of execution? I mean, yes, people doing this are clever. Not everyone can pull it off. But it is derived from existing casual speech patterns and recognizable as such.

> Latin was a dead language when both Gauss and Copernicus were alive.

That was exactly my point, and the reason I picked those examples; I guess I could have been clearer in my previous comment. The author of the poem being discussed in this thread, whose name was Māgha, lived in the 7th century. It is hard to be sure exactly when Sanskrit stopped evolving as a spoken language and became "frozen" into its classical form (or rather, the naturally evolving language stopped being called Sanskrit), but one date conventionally taken is that of Pāṇini, who lived in possibly the 4th or 5th century BCE. So at the time Māgha was born, Sanskrit was a “dead” language (in that sense) for about a thousand years, give or take a few centuries. He would have learned the language not as his primary spoken tongue of everyday life, but more as the language of education and scholarship, somewhat similar to the way Gauss or Copernicus would have learned Latin, and the way we learn Sanskrit today. That's what I meant in my previous comment by:

> both modern speakers and [Māgha] are from centuries after [Sanskrit] was the spoken language of the street […] when learning a fixed version of a classical language, the differences are relatively minor.

Now, we can be very certain that the kinds of constrained poetry we see here are the result of arduous (and masterly) construction and not a product of casual speech patterns, because:

1. On general linguistic grounds: constraints like two-dimensional palindromes and the cakrabandha (where every third syllable of the fourth line matches a specific syllable of the first three lines, and certain annuli spell out something meaningful), are not discernible to the ear (unlike say, rhyme or assonance).

We never see something like it produced in daily speech (in English or any language), and when it is done (see Oulipo and similar), it is with much effort (and the resulting language—e.g. the “Pilish” of Mike Keith, or “A Void” by Adair—can sound slightly stilted, which is the case with several of Māgha's poems too).

In Sanskrit, no one produced such stuff in the centuries when it was a language of daily speech.

2. Even in the centuries before and after Māgha, only very few authors have managed it. Looking at modern speakers (which is a reasonable proxy; see above), it takes a rare skill, and the distance between not speaking Sanskrit and fluently speaking idiomatic Sanskrit is still smaller than the distance between the latter and being able to compose like this.

You have a great point though, which applies to many situations other than this one: for example, I might hear someone today rapping in Old English (the language of Beowulf) and be impressed, but for someone living back then, it may have been (like the "an ordinary job" comment that started this thread) just natural and requiring only slightly above-average competence.

Wow, it just struck me that Ruby is already about 25 years old.

Sanskrit was very much a widely used scholarly and courtly language until the middle of the last millennium.

Fascinating. They also say Ruby is mostly dead/obsolete now, (it is to say that is the parlance, not something I personally believe!)

A Ruby game engine was recently released https://dragonruby.itch.io/

Would you accept that in a Hangman game? It beats my previous favorite Eyjafjallajökull. Much harder to memorize though.


52 letters :)

>>The 34th stanza is the 33rd stanza written backwards, with a different meaning. Finally, the 27th stanza is an example of what has been called "the most complex and exquisite type of palindrome ever invented".[19] It may also be thought of as a syllabic Sator Square. Sanskrit aestheticians call it sarvatobhadra, "perfect in every direction" — it yields the same text if read forwards, backwards, down, or up:

It is instances like these when I utterly regret my school not having offered Sanskrit back then. Now I highly doubt if I would ever get the opportunity to learn this beautiful language, ever!

Fun fact → In the '80s, a NASA engineer discovered that Sanskrit can be a more viable language in NLP than any formally defined programming language[1] courtesy its constrained grammatical states (to coin a phrase)


[1] https://www.aaai.org/ojs/index.php/aimagazine/article/view/4...

This claim of NASA endorsing Sanskrit has been debunked many times. Regardless, its a favorite among a particular milieu [0] -- our beautiful heritage stabbed in the back by some evil conspiracy -- and some such. Made up stuff like that hurts the cause more than it promotes.


[1] https://www.quora.com/Is-Sanskrit-compulsory-in-NASA-for-AI-...

The “claim of NASA endorsing Sanskrit” is an exaggeration that is easy to debunk, but the paper itself is real: it presents (though in a vague way) something of historical interest, namely a dialect of Sanskrit that was (and is) used for giving precise definitions and statements, attempting to avoid ambiguity. Of course from one perspective this is “just” a historical curiosity and an irrelevant dead end as far as AI is concerned, while from other it's an important landmark in the intellectual history of the world. (See my other comments in this thread if they're visible; trying to avoid repeating myself.)

Aside, just remembered: there's a talk on this Navya-nyāya tradition that I listened to recently: https://www.youtube.com/watch?v=KRiv7uv_C90&feature=youtu.be... (covers Navya-nyāya in the 2nd half, but may be hard to follow if you're not familiar with the Sanskrit tradition). For a Western audience, these ideas may be easier to understand via the papers I mentioned at https://news.ycombinator.com/item?id=14295285, or something like the “History of Philosophy Without Any Gaps” podcast, which is good in general (but which covers only the older nyāya and not navya-nyāya so far, in its 62 episodes https://historyofphilosophy.net/india).

Of course the paper is real, but not one that's particularly influential, notable, novel or one that led to anything more substantial. Regardless, its frequently used to push a political agenda of Indian/Hindu superiority. Either overtly, or covertly.

Here's how I think about these things. You can either react to the extreme version, focus on the political consequences, and what the person-on-the-street says/thinks. This is totally valid, and closer to what happens “on the ground”. Or you can (if you can afford not to worry about it) take delight in intellectual curiosity, approach everything like a student, trying to be as scholarly and scrupulously honest as possible, gently correcting people when they make exaggerated claims, etc. I mean, no matter how much you use strong words like “debunked” etc., people will tend to just ignore it and keep bringing up the same dumb thing as evidence of whatever-they-imagine and only think you're ignoring/fighting them, while instead if you non-confrontationally give them a proper judicious account of its role, they at least have a positive valence towards what you're saying and there's a chance the accurate version will sink in. Even if not, it's also good practice for being able to extract whatever good is in something: haṃsa-kṣira-nyāya or nīra-kṣīra-viveka as they say in Sanskrit.

(I think I already mentioned this paper being a “dead end” etc in my comments, even in my first comment on this topic from 3 years ago: https://news.ycombinator.com/item?id=14295285. BTW it did lead to something more: excitement over the paper led to a conference the next year in India—see the only other paper of Rick Briggs listed in DBLP—though that too fizzled out and led nowhere... I haven't even been able to find any record of its proceedings being published anywhere. Anyway, telling people about all this creates a mystery and raises questions—if a lot of people putting their minds together couldn't produce anything useful for AI from it, maybe it's not so useful for AI after all?—and invites them to reconsider their position, and generally leads to better outcomes than quick debunking and denial of false claims, IMO.)

I think we are more in agreement than in opposition. The two point of views that you mentioned should not even be complementary. Delighting in the intellectual marvel should not stop one from calling out abuse.

Yes we are in agreement! There is no opposition, and if at all there is any disagreement it is at most one of focus and emphasis, not anything substantive. Consider the chain:

<Fun fact: X, imperfectly stated> → <NASA engineer says Sanskrit can be used for knowledge representation> → <NASA endorses Sanskrit for AI> → <Sanskrit is good for AI> → <Sanskrit is the best language for computers> → <claims of superiority> → <Y: the thing you might be actually worried about>.

In any such chain X→…→Y, I think it's possible that worry about Y can hinder ability to appreciate X on its own terms as an intellectual delight, so it requires some conscious mental effort of stepping back or “depoliticization”, to do so. Then, for “calling out abuse” (as you said, they're not complementary), one again has a choice of form. A shooting-down approach, especially one where one debunks something actually further ahead in the chain, has its advantages — it is strong, and plays well to third parties and to those already on one's side. But it can be polarizing, and poor at convincing the person who was saying X. It seems preferable to me to start by agreeing with X (in this case X too was imperfectly stated so actually one would have to reach back and agree with the correct W), then add a cautionary note about the next steps in the chain that are not true. But this is ultimately a matter of preference no doubt. After all, those who react strongly seem to be more successful in most outcomes other than convincing the person they're reacting to. :-)

Anyway I guess it is a bit pointless and too meta to be talking about talking, so I'll end here; will just for concreteness repeat the correct (AFAIK) version of the “fun fact”: there existed/exists an interesting and sophisticated special-purpose usage of (a subset of) Sanskrit that was used to avoid (reduce?) the ambiguity that is common all natural languages, and Sanskrit in particular, and this was (roughly) pointed out in a paper in the 80s that caught some interest because of the author's employer (NASA) and publication venue (AI Magazine), though there are much better sources, and though it is not actually any more useful for AI than the typical approaches used (even at that time) for knowledge-representation.

>>Sanskrit that was used to avoid (reduce?) the ambiguity that is common all natural languages

I did not even bring up any other naturally spoken language. The author wants to say that we have a subset in Sanskrit using which can be much better than a formally defined language. There wasn’t even an attempt to bring any other naturally spoken language in the picture.

To be clear, I wasn't attributing anything to you: I wasn't saying that either you or Briggs mentioned other languages; I was just describing the fact that the Navya-Nyāya language was designed to reduce the ambiguities that natural Sanskrit has, like all natural languages — sorry if that wasn't clear.

As for the other part, consider what this NN language was designed for: to give precise definitions and statements, for use in debate, dialectic, etc. While it is plausible that some of these properties could make it suitable for similar purposes on a computer too, it also had other constraints (designed for humans, had to be grammatical Sanskrit, etc). How likely do you think it is that it can be much better than anything that could be designed for computers, without those constraints? Besides, if it were so optimized for computers it would be slightly worse for its actual purpose of human communication in debates. Have you considered the fact that Briggs does not claim so in his paper either?

> subset in Sanskrit using which can be much better than a formally defined language

Did you get in what ways would it be "much better", I didn't quite get that part.

To be fair to OP, they didn't not propagate the urban legend that "NASA uses Sanskrit for AI", as in the quora thread you cite. The OP simply reported that a paper was written by "a NASA engineer" that promotes Sanskrit for NLP.

Some of that seems to be true: the paper was indeed written and the contact details of the single author, Rick Brigs, are "RIACS, NASA Ames Research Centeu, Moffet Field, California 94305". That doesn't mean they were a "NASA engineer"- but they probably were a NASA employee.

Besides that, there doesn't seem to be any hint that NASA, as an org, adopted the use of Sanskrit for ... something.

All that you said is true, but there is a well funded gaslighting of sorts that goes around in my country along these lines. This flows right from the top, cabinet ministers and all. So there is an understated subtext that is supposed to be understood by Indian readers.

Excellent and fair review BTW. I hope you do some reviewing for AAAI, ICML et al

Thanks. So do I- I need to publish something first :)

All the best.

I am hopeful that the time is ripe for ILP or a merger of logic and probability.

If you're interested in learning Sanskrit, I'd definitely recommend this site [1]. I was able to gain a solid Sanskrit foundation in about a couple months (though I should mention I have amateur linguistics experience which helped).

And I'm always a bit skeptical of those Sanskrit NLP claims, but it's interesting nonetheless.

[1] https://learnsanskrit.org

Thank you for the link.

Apropos of your skepticism, I tried my best to back my claim with a research published by the engineer.

> Apropos of your skepticism, I tried my best to back my claim with a research published by the engineer.

The publication is not from a "NASA engineer". Neither does the publication, back up any of your claim.

I admit I haven't read the paper, but as someone who works in NLP and is in a linguistics department I have to say that this claim seems absurd. Leaving aside what "viable in NLP" means, all natural languages are rich in ambiguities, redundancies, and hazy polysemy that don't occur in, say, FOL used for knowledge engineering. And Sanskrit, being a natural language, is no exception.

If it did turn out that Sanskrit were exceptionally well suited for something in AI, I'm willing to bet that either the domain of AI or Sanskrit (either the systems in its grammar, or our respect for it as a natural language that can only imperfectly be captured by a formalized grammar of it) would need to be so constrained in scope that this supposed exceptionality would seem rather cherrypicked.

I'm not saying Sanskrit isn't cool (it's super cool), but I am saying we should resist insinuations that it's somehow more conducive to sophisticated thought or whatever.

Think classical Latin on steroids. Thanks largely to its scriptural use and Pāṇini's absolutely magisterial grammatical description meant mostly to protect the meaning of those scriptures, it was frozen in time at a much more rigid and inflected state than any of its Indo-European cousins. Classical Greek and Latin are slovenly and full of elisions and street usage - recreating infrequently-used grammatical cases using auxiliaries rather than endings/affixes, f'rinstance - by comparison. Grammatical precision in no way prevents the construction of meaningless or muddled utterances.

Sure, it's a fact of Sanskrit that of all the well-attested classical IE languages it's done the best job of preserving PIE's case system. But what's this have to do with the question of whether Sanskrit has some especial status as a tool for thought, AI, or whatever people are claiming in this paper and elsewhere? Is it that using inflection for indicating grammatical relations is somehow superior to using adpositions? I'd be very interested to hear an argument that makes the case for this for something more than aesthetics, but I can't imagine one myself.

It's that grammatical specificity that is the core - and, truth to tell, the entirety - of the argument, as if it makes Sanskrit somehow equivalent to Loglan/Lojban-type constructed languages or a purpose-built computer language. Well, that and the prevalent idea in some nationalistic circles that Sanskrit is the original, perfect language, and that everything else is just a corrupt, fallen version of it. It's wishful thinking. Woo. It's the Lysenkoism of linguistics.

I am curious about Avestan. It is dead and sister of Sanskrt.

As an Indian I have enormous curiosity about Persian culture -- cousins somehow distanced by the force of history.

I find so many similarities in language (well of course that's a no-brainer), traditional music, recipes. I met many Iranian people while I was a student in the United States. My experience has been so absolutely wonderful.

I don't how much goodwill Iranians have towards Indians, but personally I have a lot towards them. Hopefully would be able to visit sometime. This was impossible while I was a student in US -- without burning bridges that is.

I did read the paper. It's vague, badly written and its main claim is not supported by significant evidence, either theoretical or empirical.

First of all- what is the main claim of the paper? It's difficult to tell because it's not stated clearly at any point. There are a number of sentences in the paper that seem to be trying to claim something, but the author seems to be unclear himself what that something ought to be.

For example, this sentence, in the abstract, seems to make a strong claim:

  This article demonstrates that a natural language can serve as an artificial
  language also, and that much work in AI has been reinventing a wheel
  millenia old.
If this is indeed the paper's main claim, then it is not supported by the rest of the paper. The paper starts with a description of one kind of semantic net (similar, but not identical to, a semantic net scheme proposed by Rumelhart and Norman as a model of human memory). It then describes a scheme for the formal semantic analysis of Sanskrit first proposed by Panini and his disciples in the first millenium B.C. The article then goes on to show a few cases where the Paninian semantic analysis of Sanskrit and English sentences coincides with the semantic analysis by the semantic net scheme chosen by the author (i.e. they yield the same triplets). The author finds multiple commonalities between the two kinds of analysis and concludes that they are equivalent. The paper concludes with some musings about the accomplishments of Indian linguists of Panini's line that he likens to "computer scientists without the hardware". An interesting thought, to be sure.

However, a demonstration that a semantic analysis of a scant few sentences in two natural languages by two different methods (semantic nets and Paninian grammar) has similar results can not be accepted as proof that the two methods are equivalent, which really seems as what the author is trying to claim. At best it's evidence that there is an equivalence, but since this evidence consists of just a few sentences, it is really weak evidence. There is probably an infinite number of sentences that it is possible to construct in English and in Sanskrit- and a possibly infinite number of semantic parses based on the author's own chosen semantic net scheme, of each of those sentences. There's nothing to say that if we kept analysing English and Sanskrit sentences in the way the author shows, we would keep getting the same results he obtained.

So to recap, the paper has a very vague main claim that is very loosely supported empirically by some examples that we have no way to know are representative. Like I say in another comment, I really can't understand how this passed peer review in AAAI. There is no chance that this would ever be published in AAAI today.

I think you missed the main claim (the paper is definitely poorly written, also it's easy to miss unless one is familiar with its subject already): what it's pointing out is that a particular “dialect” of Sanskrit (called “Shastric Sanskrit” in the paper) was/is basically an artificial language (that's what was intended by “a natural language can serve as an artificial language also”) that has been used for essentially something like knowledge representation. More details in my comment on this post about “Attempto Controlled English”, which is something similar in English: https://news.ycombinator.com/item?id=20950126

Of course there's no point doing any of this now, it's only a curiosity as far as AI is concerned (though interesting from the viewpoint of intellectual history), and it's not surprising that the paper has been a dead end.

Perhaps that's the paper's main claim. But it's really not clear from the paper itself. You have to apply a good amount of interpretation to extract such meaning from the paper. That's what makes it a poorly written paper and I think that's the main reason why it has not been influential.

For example, I know (of) Attempto and have done a bit of reading on CNLs in general (mainly because I've done some undergraduate and post-graduate stuff with the Magic: the Gathering CNL, "ability text" :). There is no attempt by the author to describe "Shastric sanskrit" as a CNL for knowledge representation. That would have been a clearly stated claim and it would make a lot of sense- but, it's nowhere to be seen.

This paper has not influenced NLP or AI very much. It has 26 citations on Semantic Scholar, none of which are "highly influential":


To be honest, after a cursory reading, it sounds a bit meh and I'm surprised that it was actually published by AAAI.

I was taught Sanskrit in school (for 3 years, through Junior High School) and this is completely unreadable for me, due to the vocabulary difference and the wordplay. Miles apart from what I was taught.

Sanskrit is pretty close to Hindi, so much so that most people can look up Hindi courses online and be able to read Sanskrit without issue.

In many ways, Hindi was just an optimization pass on the language.

Edit: studied Hindi and Sanskrit in school. Modulo a few things, they’re very similar. Far more so than other linked languages like Latin and its derivatives

Not sure what you mean by an optimization pass. Hindi is one of the many vernacular languages derived from Sanskrit. You will find similarities with other vernacular languages derived from it as well, for example, Marathi, Gujarati, Bengali. The Indian national anthem and the Indian national song (yeah India has one of each) are in Bengali but they almost sound indistinguishable from Sanskrit to the untrained ear.

I was speaking to written form language.

In written form, Hindi is closest IMHO to Sanskrit versus the other major Indian languages. However it drops certain aspects like explicit अ and a few less common characters. But largely if you can read Hindi, you can read Sanskrit without much issue.

Whereas other Indian languages diverged more in their written text form. Though even there, I can read Gujarati text to some degree.

I’m ethnically Bengali so I can understand some spoken Bengali but struggle to read any due to the more marked difference in how the letters are formed.

Ah! that's different. What you are talking about is the Devanagari script. https://en.wikipedia.org/wiki/Devanagari

That's not the only script that Sanskrit was written in. Devanagari wasn't even the first script that Sanskrit was written in. In fact Sanskrit did not have an 'official' script. Marathi and Gujarati also uses Devanagari with a few minor differences.

For sure, I should have expanded that I only meant to say Hindi is a good gateway for learning Sanskrit today when most Sanskrit is represented most similarly to Hindi. There’s a lot more resources for Hindi than Sanskrit and IMHO it’s the fastest path.

I wasn’t speaking for historical accuracy or commenting on other languages.

Reminds me of "Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo":


I'm not sure if this was one of Neal Stephenson's inspirations for his novel Snow Crash, but if it wasn't, then I expect he would be pleasantly surprised by this real-world demo of grammar hackery.

Even though I had Sanskrit as the third language from 6-8 grades, the way it was taught was mere memorization. I still remember some till today.

I wish I was taught better.

Are there any sources to relearn Sanskrit from the ground up.

I recommend this: https://www.youtube.com/playlist?list=PLD4Ly-G0z5oD9zlmftgeK... — it may seem slow-going but stick with it. Ignore the Hindi-language parts if you don't understand them; they are an annoyance anyway. You can repeat along to what you hear, to reinforce the language in your head.

Otherwise if you prefer text there is http://learnsanskrit.org/ etc.

I grew up in Sri Lanka, where you can go to Buddhist schools from grade 1-12 every Sundays, and they teach a fairly good amount of Paali and Sanskrit. Sanskrit itself is not something you learn, but Paali resembles a lot of it to a point that you can walk up to a document from a thousand years ago and still grasp what it means.

I studied only for 4 years, but I'm native Sinhalese, which isn't really that far from Sanskrit.

> I studied only for 4 years, but I'm native Sinhalese, which isn't really that far from Sanskrit.

It's interesting you say that. My native tongue is Hindi, but growing up I had a friend whose mother tongue was Malayalam, and according to him Malayalam was closer to Sanskrit than Hindi.

Do you know if there are similarities between Sinhalese and Malyalam?

A lot! The script is similar (circles and more circles), and when I watch Malayalam movies, I pick quite a few words.

My Indian work colleagues do this a lot. Sanskrit+NASA+India+AI = super power. How is this hacker news?

How is this not linguistic hackery?

... hackery of epic scale.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact