How do archaeologists, cryptoanalysts and linguists decipher extinct languages? Has there been a case in history where this was successfully accomplished, without the means of something like the Rosetta stone or people that still speak a similar variant of the language in question?

Obviously you start out by collecting statistics. How often do characters appear, how often do certain characters appear together, etc. But what do you then do with those statistics? How does it help knowing that A E I O U are vowels, or that u almost always follows q and that h frequently follows t in English?

If you were to discover English, without knowing anything about it (except for maybe that it is an alphabetical language), or any related languages, how would you go about deciphering it? To simplify things let's assume a fairly simplified version of English with only abstract concepts like squares or triangles. If you had books full of sentences like "A square consists of two triangles", "A triangle consists of three lines intersecting with each other", etc. how could one extract the meaning of those sentences without prior knowledge of the language?

  • Are you more interested in the deciphering of obscure languages in general or more interested in (possible/hypothetical) languages like what you described?
    – Louis Rhys
    Nov 10, 2011 at 6:56
  • Not sure what the right answer to that question is. I'm more interested in the general process itself rather than specifics about different languages. My example above was just so as not to get in a debate about knowledge, lexicography, character sets, etc. and cut directly to the core of how you can "extract" the meaning out of sentences.
    – Dexter
    Nov 10, 2011 at 9:20
  • This is a fascinating question. It’s somewhat related to the (possibly future) study of alien languages. Say we intercept radio excerpts of an alien language. This may be even harder than trying to figure out the meaning of unknown characters. Also, your example of mathematical texts reminds me of Carl Sagan's reasoning about communication with extraterrestrial civilizations. He says (in Pale blue dot, I believe) that science must be fairly universal, and it will provide the basis to understanding each other.
    – rberaldo
    Nov 11, 2011 at 21:43

2 Answers 2


You should read Peter T. Daniels' 1996 article "Methods of decipherment" in an excellent book, The world's writing systems, eds. Daniels and Bright. New York: Oxford University Press. He has a really nice table, "Typology of decipherment" (adapted from Gelb 1973, 1975):

enter image description here

  • Type O: (e.g., Punic texts written in Greek letters, Indian names in Chinese Buddhist compositions)
  • Type IA: (e.g. Phoenician, Ugaritic)
  • Type IB: familiar languages in unknown scripts (e.g., Linear B, Maya glyphs)
  • Type II: codes and ciphers (cryptanalysis); pronounceable but unintelligible languages (e.g., Sumerian, Elamite, Hittite)
  • Type III: (e.g., Mesopotamian cuneiform)

In the above-mentioned article, you can also read about steps in decipherment (cataloging, distributional analysis etc.)

The number of different characters is also important:

about 30 => abjad (alphabet);

about 100 => abugida (syllabry);

about 300 => logosyllabry (logography)

Re your last question, at first, most, if not all, languages are in contact with each other, i.e. there are borrowings. Secondly, there are extra-linguistic cues, e.g. a place where your text was found (thus, you may expect words for certain concepts, like certain types of trees or animals). People who spoke that unknown language should have contacts with other civilizations, thus, you may find certain proper names or concepts mentioned in an already known language.

In your example, there might be drawings of triangles and squares etc. All that might help make educated guesses and it could be a good start.

Indeed, there are some (still) undeciphered languages, because either there isn't enough material to (credibly) decipher them or nothing is known about their linguistic affinity.

  • The table you provided in your answer is ill-formatted. I was going to edit it, but I'm not sure how exactly it is supposed to be laid out. As it turns out, there is no real solution to the problem of table formatting in SE. So, you might use the "code sample" formatting, or convert the table to an image or any other creative solution you come up with ;-) Nov 10, 2011 at 12:45
  • Fixed. There were more than four spaces before the image reference, which made Markdown interpret it as code sample. Nov 10, 2011 at 23:26
  • @AlekStorm: I thought the answer needed more improvements so I improved your edit. :)
    – Alenanno
    Nov 11, 2011 at 10:45
  • @OtavioMacedo I made a simple table and took a screenshot.
    – Alenanno
    Nov 11, 2011 at 10:46
  • 2
    Great, @Alenanno! Looks much better now. Nov 11, 2011 at 11:30

The short answer is that, in the scenario you give, It's currently not possible.

All of the decipherments to date have depended on some secondary reference, be it a version of the text in a known language, some illustrations, or some other clues as to the content of the text.

In the example you give, just plain text describing abstract geometric figures, would need diagrams included to be deciphered.

  • 3
    Can you cite your references or at least elaborate more?
    – Louis Rhys
    Nov 10, 2011 at 8:40
  • Why do you say it's "currently" not possible? Is there new promising research in this area?
    – Dexter
    Nov 10, 2011 at 9:15
  • @Dexter I was just hedging. IMO It's an intractable problem, I'll have to elaborate when I have a few moments
    – Dan Milway
    Nov 10, 2011 at 11:14
  • Dexter's question suggest a plausible case: mathematical texts are fairly universal, there's a realistic chance of working one out from scratch. Especially a geometry treatise where illustrations would indicate the nature of the document and perhaps provide some starting vocabulary. Nov 11, 2011 at 18:38
  • @Gilles: Don't illustrations count as a sort of Rosetta Stone? Nov 17, 2011 at 17:47

Not the answer you're looking for? Browse other questions tagged or ask your own question.