22 August 2003 CLEF 2003
The 2003 TIDES Surprise Language Exercise
Douglas W. Oard
University of Maryland
Outline
• Thinking out of the box
• Some results
• Lesson Learned
Surprise Language Framework
• Zero-resource start (treasure hunt)
• Time constrained (10 or 29 days)
• English Users / Documents in language X
• Character-coded text
• Research-oriented
• Intensely collaborative (team-based)
Schedule
Cebuano• Announce: Mar 5• Test Data: • Stop Work: Mar 14• Newsletter: April• Talks: May 30
(HLT)• Papers:
Hindi
Jun 1
Jun 27
Jun 30
August
Aug 5 (TIDES PI)
Aug 15 (TALIP)
16 Participating TeamsCebuano and Hindi
ISI
Maryland
NYU
Johns Hopkins
Sheffield
LDC
CMU
UC Berkeley
MITRE
Hindi Only
U Mass
Alias-i
BBN
IBM
CUNY
KAT
SPAWAR
• Five evaluated tasks– Automatic CLIR (English queries)– Topic tracking (English examples, event-based)– Machine translation into English– English “Headline” generation– Entity tagging (five MUC types)
• Several useful components– POS tags, morphology, time expressions, parsing
• Several demonstration systems– Interactive CLIR (two systems)– Cross-language QA (English Q, Translated A)– Machine translation (+ Translation elicitation)– Cross-document entity tracking
Hindi Participants
Alias-I
UC
Berkeley
BB
N
CM
U
CU
NY
Johns Hopkins
IBM ISI
LDC
MIT
RE
NY
U
SP
AW
AR
U. S
heffield
U. M
assachusetts
U. M
aryland
ResourceGeneration
Detection
Extraction
Summarization
Translation
TranslationDetection
Extraction
Summarization
BooksWeb
Books
WebPeople
Lexicons
Corpora
Time
ResourceHarvesting
Systems
ResearchResults
CaptureProcess Knowledge
Innovation Cycle
Coordination
StrategyPushOrganizeTalk
The Synchronization Challenge
Cebuano MT Results
0 2 4 6 8 10 12
DDC
DCNDCNB
DBNDB
D5CN5BMDCNBM
DCMDBC
D5CN10BMDCNMDBCMDBNM
DNDBMDNM
DMDCN5BM
BLEU (%)
BibleCebuano bookDictMelamedNews
Cebuano Interactive CLIR
• Starting Point: iCLEF 2002 system (German)– Interface: “synonyms”/examples (parallel)/MT– Back end: InQuery/Pirkola’s method
• 3-day porting effort– Cebuano indexing (no stemming)– One-best gloss translation (bilingual term list)
• Informal Evaluation– 2 Cebuano native speakers (at ISI)
Hindi syntax is generally very “regular”• Subject – Object – Verb is the preferred order
– John saw Mary. = जॉ�न न� मे�री� को दे�खा ।• Presence of (occasionally deleted) case markers
often permit reordering– John saw Mary. = मे�री� को जॉ�न न� दे�खा ।
• English (or western) punctuation is pervasive in many modern texts– John said, “ I am here ” = जॉ�न न� कोहा , “ मे� यहा � हूँ�
”
• The subject may be omitted in some contexts– A: Where is John? B: [He] went home.– अ: जॉ�न कोहा � हा�? ब: [वहा] घरी चला गय ।
Hindi Encoding• Text encoding for storage and transmission and text
rendering for display and printing are separated
• Which syllable constituents get their own code-points?– Several 8-bit encodings:
• After assigning a code point to each stand-alone vowel and full consonant, and to half-consonants and vowels within a syllable, spare code-points get used for assorted/frequent CC clusters.
– Unicode UTF-16: Only stand-alone vowels, full consonants and vowels within syllables have their own code-points. All half consonants are realized by a `full consonant + halant’ sequence
• Choice of the “grammar” for syllable construction and rendering?– Several 8-bit encodings write the code-points in display order,
simplifying the rendering program– Unicode writes it in pronunciation order, making for a
considerably more complex display program
Hindi Week 1: Porting• Monday
– 2,973 BBC documents (UTF-8)– Batch CLIR (no stem, 2/3 known items rank 1)
• Tuesday– MIRACLE (“ITRANS”, gloss)– Stemmer (implemented from a paper)
• Wednesday– BBC CLIR collection (19 topic, known item)
• Friday:– Parallel text (Bible: 900k words, Web: 4k words) – Devanagari OCR system
Hindi Weeks 2/3/4: Exploration• N-grams (trigrams best for UTF-8)• Relative Average Term Frequency (Kwok)• Scanned bilingual dictionary (Oxford)• More topics for test collection (29)• Weighted structured queries (IBM lexicon)• Alternative stemmers (U Mass, Berkeley)• Blind relevance feedback• Transliteration• Noun phrase translation • MIRACLE integration (ISI MT, BBN headlines)
Formative Evaluation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 5 10 15 20 25 30
Day (=Date-1)
Mea
n R
ecip
roca
l R
ank
Transliteration
• Importance: Names, loan words– देक्षि�ण कोरिरीय (Dakshin Korea)
• Pronunciation crosswalk English->Hindi– English pronunciation (Festival)– Overgenerate Hindi characters (hand-built rules)
• Doctor => d aa k t ax r OR d ao k t ax r
– Rank n-best using bigrams (Hindi name list)
• Treat as alternate translations for CLIR– Pirkola’s method
Some Challenges
• Formative evaluation
• Synchronize variable-rate efforts– Soccer, not football
• Integration
• Capturing lessons learned– See the forest, not just the trees
For More Information
• TIDES Newsletter– Cebuano: April– Hindi: August
• Papers– NAACL/HLT Short paper– MT Summit (late Sep)– ACM TALIP Special Issue
• Demonstration systems– Contact individual sites