A presentation by Martin Haase at the 24th Chaos Communication Congress, Berlin, December 2007. http://events.ccc.de/congress/2007/Fahrplan/events/2284.en.html http://www.youtube.com/watch?v=M3TYMmKxbqA http://lanyrd.com/scgypk

1. Linguistic Hacking
Martin.Haase@uni-bamberg.de
maha@jabber.ccc.de
How to know what a text in an unknown
language is about?
1

2. Contents
• how to identify the language of a written text
• in traditional ways,
• with the help of computer technology.
• how to get at least some information out of
an unknown text.
2

3. “the intellectual challenge of creatively overcoming or
circumventing limitations.”
Hacking
Eric Raymond (1996):
The New Hackerʼs
Dictionary.
3

4. Spoken texts?
multi-language corpus of telephone calls
4

5. Writing Systems
• Roman (thousands of languages)
• Cyrillic (> 60 languages)
• Arabic (> 20 languages)
• Devana̅garī (> 10 languages, not
counting derivative writing systems)
• Hebrew (~ 3 – 5 languages)
5

6. Devanāgarī
!वनागरी
6

7. 7

8. 8

9. 9

10. 10

11. Hebrew
• Old and Modern Hebrew,
• Ladino (with different varieties),
• Judeo-Arabic,
• Yiddish.
11

12. 12

13. 13

14. Norman C. Ingle (1980):
Language Identification
Table. London: Technical
Translation International.
14

15. 15

16. Computer-aided
identification
• frequencies of unique characters and
character strings
• common words recognition
• n-gram analysis
16

17. “Text”
17

18. ( TE), (TEX), (EXT), (XT )
18

19. 19

20. variant of the unique character string approach
20

21. compression efficiency
21

22. reference
model
22

23. reference
model
text in language
to be identified
+
23

24. reference
model
text in language
to be identified
+
gzip
24

25. reference
model
text in language
to be identified
+
gzip
compression efficiency
25

26. Interesting
applications
• measuring linguistic difference
> language families
• determining types of text
• spam detection?
26

27. • TextCat (http://odur.let.rug.nl/vannoord/TextCat/Demo/), n-gram
based, 76 languages, usable as a web application,
• Languid (http://languid.cantbedone.org/), downloadable
program, web application not running,
• Langid (http://complingone.georgetown.edu/∼langid/), n-gram
based, 65 languages, web application,
• LanguageGuesser (http://www.xrce.xerox.com/cgi-bin/mltt/
LanguageGuesser), frequency tests on characters and character
sequences, about 40 languages, web application,
• Polyglot 3000 (http://www.polyglot3000.com/), corpora, method
unknown, currently 441 languages, closed-source Windows
freeware. :-(
27

28. approaching “content analysis”
28

29. Hackerʼs approach
• numbers, dates, words from another
language
• typographic hints:
• bold or italic print,
• colored or underlined text chunks,
• capital letters
29

30. Zipfʼs law
Very frequent words are shorter and contain less
lexical information, whereas infrequent words are
longer and contain more lexical information.
30

31. less lexical information implies more grammatical
information and vice versa
31

32. most interesting for us:
words with more specific lexical information
32

33. Ignore all short words!
(even if they reiterate throughout the text)
33

34. Ua salalau lenei gagana i le lalolagi atoa. ‘O lenei fo‘i
gagana, ‘ua ‘avea ma gagana lona lua a le tele o
tagata ‘o le vasa Pasefika, e pei ‘o Samoa. E iai le
manatu, ‘o le gagana fa‘aperetania, ‘ua matua̅ talitonu
i ai le tele o tagata Samoa e fa‘apea ‘o le gagana e
maua ai le atamai ma le poto. ‘E talitonu fo‘i nisi o i
latou, ‘e le̅ aoga la latou gagana. E le̅ sa‘o lea ta̅ofi,
‘aua̅ e ‘avatu le gagana fa‘aperetania i Samoa, ‘ua
leva ona atamamai ma popoto tagata Samoa e fai lo
latou soifua ma lo latou lalolagi.
34

35. Ua salalau lenei gagana i le lalolagi atoa. ‘O lenei fo‘i
gagana, ‘ua ‘avea ma gagana lona lua a le tele o
tagata ‘o le vasa Pasefika, e pei ‘o Samoa. E iai le
manatu, ‘o le gagana fa‘aperetania, ‘ua matua̅ talitonu
i ai le tele o tagata Samoa e fa‘apea ‘o le gagana e
maua ai le atamai ma le poto. ‘E talitonu fo‘i nisi o i
latou, ‘e le̅ aoga la latou gagana. E le̅ sa‘o lea ta̅ofi,
‘aua̅ e ‘avatu le gagana fa‘aperetania i Samoa, ‘ua
leva ona atamamai ma popoto tagata Samoa e fai lo
latou soifua ma lo latou lalolagi.
35

36. Ua salalau lenei gagana i le lalolagi atoa. ‘O lenei fo‘i
gagana, ‘ua ‘avea ma gagana lona lua a le tele o
tagata ‘o le vasa Pasefika, e pei ‘o Samoa. E iai le
manatu, ‘o le gagana fa‘aperetania, ‘ua matua̅ talitonu
i ai le tele o tagata Samoa e fa‘apea ‘o le gagana e
maua ai le atamai ma le poto. ‘E talitonu fo‘i nisi o i
latou, ‘e le̅ aoga la latou gagana. E le̅ sa‘o lea ta̅ofi,
‘aua̅ e ‘avatu le gagana fa‘aperetania i Samoa, ‘ua
leva ona atamamai ma popoto tagata Samoa e fai lo
latou soifua ma lo latou lalolagi.
36

37. Ua salalau lenei gagana i le lalolagi atoa. ‘O lenei fo‘i
gagana, ‘ua ‘avea ma gagana lona lua a le tele o
tagata ‘o le vasa Pasefika, e pei ‘o Samoa. E iai le
manatu, ‘o le gagana fa‘aperetania, ‘ua matua̅ talitonu
i ai le tele o tagata Samoa e fa‘apea ‘o le gagana e
maua ai le atamai ma le poto. ‘E talitonu fo‘i nisi o i
latou, ‘e le̅ aoga la latou gagana. E le̅ sa‘o lea ta̅ofi,
‘aua̅ e ‘avatu le gagana fa‘aperetania i Samoa, ‘ua
leva ona atamamai ma popoto tagata Samoa e fai lo
latou soifua ma lo latou lalolagi.
37