3 ms·
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- wha
by YeGoblynQueenne 7y ago
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.
- svat 7y agoI 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 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.
- YeGoblynQueenne 7y agoPerhaps 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.