Transcription of names written in Farsi into English

1
Transcription of names written in Farsi into
English

• Joshua Johanson
• Inxight Software/Business Objects

2
Why Transcription?

• ?????
• mhmdi
• transliteration
• Mohammadi
• transcription

3
Standardize
Dictionary (6750 names) Mohammadi ?????
?????
?????
Mohammadi
Orthographic (9355)
Reza Ali Mohammadi
mhmdi
Mohammadi
mohammadi
Basic 8 mhmdi
Mhmdi
Phonetic (13,758) frnsisku fræns?sko? Francisco
????????
Francisco
4
Standardize Input

• Arabic letters are converted into Persian letters
• ? - ?
• ? -?
• Nonstandard encodings are standardized
• ?, ??? - ??
• Convert to UTF-16
• Spacing and non-joiners are dealt with in the
  dictionary

5
Dictionary Lookup
6
Create Dictionary

• Web sites - such as yellow pages and baby name
  lists.
• Account for spacing
• Hand translated by Shahla Fahimi
• Websites, internal lexicons, sample extractions

7
Dictionary Lookup

• Look for complete names
• Check for variance in diacritics
• Only makes a difference in 64 names
• If not one of 64 names, ignore
• Check for possible affixes

8

• Prefix
• ??? Pour
• ??? Pir
• ????? Piruz

Root ?????Ghorabi ?????Ghoraie ?????Ghorban
??????????
Ghorbannejad
Suffix ???? Nejad ??? Neyat ????? Nezhadi
9
Testing

• Took names from Persian Wikipedia
• Variety of origins
• 62 of names were found by the dictionary

10
Orthographic Comparison
11
Orthographic Comparison

• Gather lists of name (Dictionary, Wikipedia, web)
• Use names to create a Finite State Transducer
• Allow possible short vowels and doubled letters
• Allow all ambiguous combinations
• Special sequences - ??? (Kha), ?? (U), ?? (I)
• Check for joined names, like ??????? (Mohammad
  Reza).

12
n
a
y
?
?
?
o
K
Ø
Koyan, Kian
????
sh
?
i
?
a
n
a
?
?
13
Testing

• Took names from dictionary
• Gave the final answer correct 76 of time
• Found a right answer 83 of the time
• Gave wrong answer 13 of time

14
Phonetic Comparison
15
Phonetic Comparison

• Reverse transcription for names that have been
  transcribed into Farsi from another language
• Transcription can be based on pronunciation.
• Michael - ????? (maykl)
• Transcription can be based on spelling
• Graham, Holmes
• Alexander
• Rider/Writer

16
Methodology

• Use names from 1990 US Census
• Names from a variety of origins
• More resources for American names
• Use CMU pronunciation dictionary for
  pronunciation of names
• Convert to closest Persian pronunciation
• Create FST similar to Orthographic comparison

17
Persianization

• Split diphthongs
• They can be written with two vowels, even if only
  one is used in English.
• Convert non-existent sounds to closest existing
  sounds
• Like /?/ and /w/
• /?/ is not written in Persian convert to either
  t or d.
• /z?/ (voiceless z) can be /s/ or /z/
• Add ? before names beginning with an s followed
  by a consonant

18
?
l
?
?
p?l Paul pa??lPowell pæv?lPavel
Paul Powell Pavel
??
p
?
a?
l
????
??
?
?
Ø
æ
?
v
?
l
?
?
Ø
19
(No Transcript)
20
Testing

• Took list of 2190 foreign names
• Not all names were American
• Found answer for 1337 names

21
Basic Transcription
22
Standard Transcription

• Most of the work was done by Karine Megerdoomian
• I added diacritics
• Most letters and diacritics have a direct,
  unambiguous transcription
• Short vowels are not guessed
• Intelligent guess is made for ambiguous characters

23
(No Transcript)
24
(No Transcript)
25
Questions?