Transcript Machine Translation marazI to UNL

Machine Translation
marazI to UNL
Presented by
Ashwini, Salil
Center for Indian Language Technology Solutions
CSE, IIT Powai
Characteristics of marazI
a. Syntactic structure
–
Subject-object-verb
e.g. rama Baat Katao.
–
Similarity with Hindi
b. Morphology
– P`a%yaya
–
Differences with Hindi
Main tasks
1. Marathi-UW dictionary building
2. Rulebase building for converting Marathi
language phenomenon to UNL expressions
3. Testing using corpus sentences
4. Verification with Hindi and Marathi
deconverters.
Analysis consists of
•
•
•
•
Morphology
Syntax
Semantics
Pragmatics
Marathi analysis done so far
We focus on Marathi morphology
•
•
•
•
•
Noun morphology
Pronoun morphology click
Verb morphology click
Relation label morphology click
Adjective morphology click
Types of adjectives in Marathi
1. Pronounic adjectives
1.1 Pronoun adjectives: The nine pronouns being
used as adjectives.
1.2 Adjectives derived from the nine pronouns
2. Qualitative adjectives
2.1 Adjectives ending with vowel +É
2.2 Adjectives ending with vowels other than +É
2.3 Postposition adjectives
Type of adjectives [contd.]
3. Numerical adjectives
• 3.1 Cardinal
3.1.1 (whole number)
3.1.2 (fractional number)
3.1.3 (entirety, totality, completeness)
• 3.2 Ordinal
• 3.3 Occurrencial
6 types
• 3.4 Distinctive
[pAvaNedonashe] means 175 or 199.75?
- There is no word assigned to 199.75, 299.75, etc.
- the problems with paun, pauvane and savva.
- (pAvaNedon) times 100 (she). she and shambhar,
both mean 100. pAUNashe means 75.
pAvaNeshambhar means 99.75.
- The powers of ten for which there is a distinct
word in Marathi need to be stored separately.
- pronunciation is not
pAvaNedona-[pause]-she but
pAvaNe
-[pause]-donashe
Tables of numbers:
continous and random access.
• Some forms of numbers are used for verbalizing
the tables of numbers: ºÉÉiÉ / ºÉÉiÉÉ / ºÉÉiÉä / ºÉÉiÉÒä /
ºÉiiÉä.
• Marathi: A, B times, (is C), occurring in the table
for A. English: B A’s (are C).
• Usage of forms: 1. only for the expression ‘A’
2. only for ‘B times’ 3. only while recalling the
number directly without going through the table.
• Some forms occur especially for square. The
repetition is emphasized.
words used to familiarise a child
with numbers
• Some words are used mostly to familiarise a
child with numbers: BEÒ BE, nÖEÔ nÉäxÉ, ÊiÉEÔ
iÉÒxÉ, etc. The similarity of each word with
the number is used to help a child remember
the number. The words used as familiarisers
are: BEÒ, nÖEÔ, ÊiÉEÔ, SÉÉèEÒ, {ÉÉSÉÒ, ºÉɽÒ,
ºÉÉiÉÒä, +É`Ò, xÉ´Éä, nɽÒ.
playing cards and game of
cricket
1. playing cards:
ekka, durri / durra, tirri / tirra, chavvi / chouka,
panji / panja, chhakki / chhakka, satti / satta,
atthi / attha, navvi / nashsha, dashshi / dashsha.
2. shots scoring multiple runs in the game of cricket:
SÉÉèEÉ®, ¹É]EÉ®.
The current status of dictionary
Number of entries 375
•Dictionary click
•Nouns
•Noun morphology suffixes
•Verbs
•Verb morphology suffixes
The current status of rulebase
Number of rules is 1050.
• Verb morphology (Simple and conjunct verbs)
– Tense (Past, Present, Future)
– Aspect of tense (Progress, complete, custom)
– Voice (Passive voice)
– +lÉÇ (imperative, should, negative)
– Ability, intention etc. for conjunct verbs only.
The current status of rulebase
[contd.]
• Noun morphology
– Number
– With case marker (ºÉɨÉÉxªÉ° {É)
• Case when penultimate vowel is either
> or <Ç
e.g. ¨ÉÚ±É - ¨ÉÖ±Éä (Plural)
The current status of rulebase
[contd.]
• Relation labels used so far
agt, obj, gol, aoj, and, or
e.g.
¨ÉÖ±ÉÉÆxÉÒ +ÉƤÉä JÉɱ±Éä xÉ´½iÉäÃ.
obj(eat(icl>do).@entry.@pred.@past.@not.
@complete, mango(icl>fruit):08.@pl)
agt(eat(icl>do).@entry.@pred.@past.@not.
@complete, child(icl>person):00.@pl)
Plans
• Adjective morphology
• Pronoun morphology
• Relation labels handling for corpus
sentences.
For simple sentence only.
THANK YOU
References:
•Damle, Moro Keshav (1970). Shastriya marathi
vyakarana. [SaswrIya marATI vyAkaraNa]. (Ed: K.
S. Arjunwadkar). Pune: Deshmukh & Co.
•Meying, Zhu (2000) EnConverter specifications,
version 2.1. Tokyo: UNU/IAS/UNL Center.
• Meying, Zhu (2002) UNL specifications, version
3 edition 1. Tokyo: UNU/IAS/UNL Center.
•Valambe, M. R. (2001) Sugam marathi vyakaran
lekhan [sugama marATI vyAkaraNa leKana]. Pune:
Nitin Prakashan.