5

There's a nice question I thought about, and I'd like to know more about it. I would assume it has been discussed many times, but I'm not sure what its called so I'm having trouble finding any texts that discuss it. Could someone kindly refer me to a relevant text?


The Question

Suppose one wanted to learn a natural language solely from a dictionary of that language (e.g. learn Greek from a Greek-Greek dictionary). Would that be possible? We assume that the learner is a normal person, but he doesn't know anything about the target language (e.g. there are no similar words between his language and the target language and the characters are all different).

On the one hand, it could be argued that this is impossible. If one does not know any Greek at all, a dictionary written entirely in Greek cannot add any new information, and hence is useless.

On the other hand, the dictionary might be sufficient. For example, one might argue that there exists a rather limited core vocabulary of say 2000 words, and that these words are used to define all the other words in the dictionary, and exist pretty much in any language. So one could in theory test all the possible matches between these words in the target language and the known language. Any match that is incorrect would yield definitions that are clearly nonsensical, so it would be possible to eventually settle down on the right fit.

6
  • Seems somewhat similar to this question, though I'm not sure there are any canonical references identified there
    – Joseph Weissman
    Feb 14, 2014 at 22:13
  • @JosephWeissman Yes there is similarity, but like you said, the OP there doesn't ask for a reference on this topic, and also the focus is more on what a language is and whether it "is a dictionary", and less on the learning issue.
    – π314
    Feb 14, 2014 at 22:43
  • 3
    It sounds like a machine-learning problem in translation rather than a generally recognised philosophical problem; I imagine it best looking for references there where they discuss the general features of the problem. ie strategy, computational efficiency, accuracy, testing etc. Feb 15, 2014 at 1:30
  • Perhaps you might want to restrict your question to written language, because no natural language is purely written.
    – user3164
    Feb 15, 2014 at 21:01
  • 1
    'Left' and 'right' may pose a problem if one doesn't have access to our natural world. I think.
    – user3164
    Feb 15, 2014 at 21:29

5 Answers 5

2

I don't know the answer to the internal question about whether it is possible -- but I do know the literature on it.

You want to look at:

(1) late WVO Quine

(2) Donald Davidson

(3) Richard Rorty - specifically Philosophy and the Mirror of Nature

(4) Hans Georg Gadamer

All four are discussing the meat of what you are asking by wondering whether it is possible to understand something at a distance.

For some other supporting reading,

I suggest:

(1) Some later Wittgenstein and his considerations about language (specifically the private language argument -- and the large body of interpretive literature that asks what it means).

(2) Greg Lynch

(3) There's also the Chinese room problem. and Searle looking at whether this is possible

1

I don't think it is a viable scientific hypotheses a "radical empiricst" theory of language aquisition.

You must see at least Noam Chomsky's theory about the syntax and his hypotheses of an "innate" syntactical capability (like a microcode harwired in tha brain that works like a boot-strapping software in order to drive the language aquisition process).

You can see in SEP Innateness and language and Philosophy of Linguistics.

Also : James McGilvray (editor), The Cambridge Companion to Chomsky (2005), Noam Chomsky, Language and mind (3rd ed - 2005) and Neil Smith, Language Frogs and Savants More Linguistic Problems Puzzles and Polemics (2005).

Also : Charles Yang Knowledge and Learning in Natural Language (2002), Nobuo Masataka The Onset of Language (2003) and Andrea Moro, The Boundaries of Babel The Brain and the Enigma of Impossible Languages (2008).

1

No, you couldn't, but not for some deep philosophical reason. But just because the lexicon (the store of words) is only one part of any language. You would also need to learn the grammar of the language.

Also, even if you had memorized the dictionary of some natural language, that wouldn't mean you'd be able to figure out the grammar on your own just by reading text.

Also you'd be lacking idioms, which are ubiquitous. And you probably wouldn't get a lot of the cultural references, either, unless the dictionary you memorized were a really, really good one.

TL;DR

Natural language are horrifically complex.

3
  • Couln't a really good algorithm guess the laws of grammar so it's more or less correct?
    – jinawee
    Feb 15, 2014 at 20:41
  • I don't think you'd be able to write such an algorithm to learn the grammar of the language without already knowing the language. The only way around this, maybe, would be if there were a universal grammar from which all natural languages were derived. If there were such a thing, and you knew what the features of that universal grammar were, maybe you could figure out the specific grammatical features of the particular language in question blind.
    – user5172
    Feb 16, 2014 at 12:40
  • (1) Some 'connectivity' of the language could likely be inferred. (2) Whether not it is declined and conjugated could also likely be inferred. (3) The Computational Theory of the Laws of Nature indicates that merely knowing the grammar is likely to be quite insufficient (search the text for 'grammar'). So you probably focus too strongly on grammar. (4) Your 'idioms' idea may be a much bigger problem; see Metaphors We Live By.
    – labreuer
    Feb 19, 2014 at 3:00
1

It wouldn't be possible without solving the problem of reference, or otherwise finding a way to map the learner's concepts into some minimum subset of the target lexicon, and that would involve more than just a monolingual dictionary.

A somewhat easier problem (#2) is "could you learn one target language if all you had are relevant monolingual dictionaries and a massive aligned corpus of texts translated between your target language and another language you don't know?" You could even add some bilingual dictionaries between the two unknown languages, but I think that would just be a jump starter, and not make it a qualitatively different problem.

In a sense, problem #2 is "solved" by Google Translate, which can generate a massive encoding of the grammar and lexicon on one language in terms of the grammar and lexicon of another. But it seems clear that by itself Google Translate hasn't "learned" any of its new languages, the semantics and certainly the pragmatics remain impenetrable (although the semantics is shifted into a new language). At least they seem impenetrable within what I understand to be the core algorithms of Google Translate; if one links further to data that Google has also been accumulating about the things that words refer to, or about the concepts they encode in some target-language-external representation, one can undoubtedly make more progress. But that is no longer similar to problem #1, your Greek dictionary learning project.

About relevant texts, my take on things points towards the massive literature on reference and concept mapping. There is a lot of recent discussion on the semantics-pragmatics interface that may be useful (the syntax-semantics interface seems insufficient for the problem(s), if you take semantics to be a mapping from forms of expression to symbolic structures, like set-theoretic worlds equipped with a signature of logically proper names). Emma Borg's Semantic Minimalism (2004) comes to mind, and Critical Pragmatics: An Inquiry into Reference and Communication by Kepa Korta and John Perry (2011).

0

I will make two assumptions:

  • Knowing a language is the ability to form grammatically correct sentences in that language, not necessarily the knowledge of it's idiomatic use and the culture around it

  • The dictionary is a resource, where each word in the language is either mapped to another word, or is described using grammatically correct sentences/sentence fragments in that language.

Within these constraints, I think it should be possible to learn the language using only the dictionary. After all, we have been able to learn ancient languages from millenniums ago, written in ancient alphabets. It would be possible by the dint of our ability to find patterns. Let me try to give an example. Given this dictionary of a hypothetical language

a: e b c d

b: f d

c: f e

d: f b

e: f c

can we find the meanings of some words? I'd suggest that given these descriptions, we can infer that f means not, b and d are opposite words which represent some kind of logical proposition, and c and e are opposite words that can be used to combine logical propositions. I'd suggest that it should be possible to learn a language with this kind of reasoning.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .