The AVENUE Project: Bootstrapping MT Prototypes for Languages with Limited Resources

1
The AVENUE ProjectBootstrapping MT Prototypes
for Languages with Limited Resources

• Faculty
• Alon Lavie, Jaime Carbonell, Lori Levin,
• Ralf Brown
• Students
• Katharina Probst, Erik Peterson, Christian
  Monson, Ariadna Font-Llitjos, Alison Alvarez

2
Why Machine Translation for Minority and
Indigenous Languages?

• Commercial MT economically feasible for only a
  handful of major languages with large resources
  (corpora, human developers)
• Is there hope for MT for languages with limited
  resources?
• Benefits include
• Better government access to indigenous
  communities (Epidemics, crop failures, etc.)
• Better indigenous communities participation in
  information-rich activities (health care,
  education, government) without giving up their
  languages.
• Language preservation
• Civilian and military applications (disaster
  relief)

3
MT for Minority and Indigenous Languages
Challenges

• Minimal amount of parallel text
• Possibly competing standards for
  orthography/spelling
• Often relatively few trained linguists
• Access to native informants possible
• Need to minimize development time and cost

4
AVENUE Partners
5
AVENUE Approach

• Elicitation use bilingual native informants to
  produce a small high-quality word-aligned
  bilingual corpus of translated phrases and
  sentences
• Transfer-rule Learning apply ML-based methods to
  automatically acquire syntactic transfer rules
  for translation between the two languages
• Rule Refinement refine the acquired rules via a
  process of interaction with bilingual informants
• Morphology Learning
• Word and Phrase bilingual lexicon acquisition

6
AVENUE Architecture
7
Learning Transfer-Rules for Languages with
Limited Resources

• Rationale
• Large bilingual corpora not available
• Bilingual native informant(s) can translate and
  align a small pre-designed elicitation corpus,
  using elicitation tool
• Elicitation corpus designed to be typologically
  comprehensive and compositional
• Transfer-rule engine and new learning approach
  support acquisition of generalized transfer-rules
  from the data

8
English-Chinese Example
9
English-Hindi Example
10
Spanish-Mapudungun Example
11
English-Arabic Example
12
Transfer Rule Formalism
SL the old man, TL ha-ish ha-zaqen NPNP
DET ADJ N -gt DET N DET ADJ ( (X1Y1) (X1Y3)
(X2Y4) (X3Y2) ((X1 AGR) 3-SING) ((X1 DEF
DEF) ((X3 AGR) 3-SING) ((X3 COUNT)
) ((Y1 DEF) DEF) ((Y3 DEF) DEF) ((Y2 AGR)
3-SING) ((Y2 GENDER) (Y4 GENDER)) )

• Type information
• Part-of-speech/constituent information
• Alignments
• x-side constraints
• y-side constraints
• xy-constraints,
• e.g. ((Y1 AGR) (X1 AGR))

13
Transfer Rule Formalism (II)
SL the old man, TL ha-ish ha-zaqen NPNP
DET ADJ N -gt DET N DET ADJ ( (X1Y1) (X1Y3)
(X2Y4) (X3Y2) ((X1 AGR) 3-SING) ((X1 DEF
DEF) ((X3 AGR) 3-SING) ((X3 COUNT)
) ((Y1 DEF) DEF) ((Y3 DEF) DEF) ((Y2 AGR)
3-SING) ((Y2 GENDER) (Y4 GENDER)) )

• Value constraints
• Agreement constraints

14
The Transfer Engine
15
Rule Learning - Overview

• Goal Acquire Syntactic Transfer Rules
• Use available knowledge from the source side
  (grammatical structure)
• Three steps
• Flat Seed Generation first guesses at transfer
  rules flat syntactic structure
• Compositionality use previously learned rules to
  add hierarchical structure
• Seeded Version Space Learning refine rules by
  learning appropriate feature constraints

16
Flat Seed Rule Generation
17
Compositionality
18
Seeded Version Space Learning
19
Seeded VSL Some Open Issues

• Three types of constraints
• X-side constrain applicability of rule
• Y-side assist in generation
• X-Y transfer features from SL to TL
• Which of the three types improves translation
  performance?
• Use rules without features to populate lattice,
  decoder will select the best translation
• Learn only X-Y constraints, based on list of
  universal projecting features
• Other notions of version-spaces of feature
  constraints
• Current feature learning is specific to rules
  that have identical transfer components
• Important issue during transfer is to
  disambiguate among rules that have same SL side
  but different TL side can we learn effective
  constraints for this?

20
Examples of Learned Rules (Hindi-to-English)
21
XFER MT for Hebrew-to-English

• Two month intensive effort to apply our XFER
  approach to the development of a
  Hebrew-to-English MT system
• Challenges
• No large parallel corpus
• Limited coverage translation lexicon
• Rich Morphology incomplete analyzer available
• Accomplished
• Collected available resources, establish
  methodology for processing Hebrew input
• Translated and aligned Elicitation Corpus
• Learned XFER rules
• Developed (small) manual XFER grammar as a point
  of comparison
• System debugging and development
• Evaluated performance on unseen test data using
  automatic evaluation metrics

22
(No Transcript)
23
Morphology Example

• Input word BWRH
• 0 1 2 3 4
• --------BWRH--------
• -----B-----WR--H--
• --B---H----WRH---

24
Morphology Example

• Y0 ((SPANSTART 0) Y1 ((SPANSTART 0)
  Y2 ((SPANSTART 1)
• (SPANEND 4) (SPANEND
  2) (SPANEND 3)
• (LEX BWRH) (LEX B)
  (LEX WR)
• (POS N) (POS
  PREP)) (POS N)
• (GEN F)
  (GEN M)
• (NUM S)
  (NUM S)
• (STATUS ABSOLUTE))
  (STATUS ABSOLUTE))
• Y3 ((SPANSTART 3) Y4 ((SPANSTART 0)
  Y5 ((SPANSTART 1)
• (SPANEND 4) (SPANEND
  1) (SPANEND 2)
• (LEX LH) (LEX
  B) (LEX H)
• (POS POSS)) (POS
  PREP)) (POS DET))
• Y6 ((SPANSTART 2) Y7 ((SPANSTART 0)
• (SPANEND 4) (SPANEND
  4)
• (LEX WRH) (LEX
  BWRH)
• (POS N) (POS
  LEX))
• (GEN F)
• (NUM S)

25
Sample Output (dev-data)

• maxwell anurpung comes from ghana for israel four
  years ago and since worked in cleaning in hotels
  in eilat
• a few weeks ago announced if management club
  hotel that for him to leave israel according to
  the government instructions and immigration
  police
• in a letter in broken english which spread among
  the foreign workers thanks to them hotel for
  their hard work and announced that will purchase
  for hm flight tickets for their countries from
  their money

26
Evaluation Results

• Test set of 62 sentences from Haaretz newspaper,
  2 reference translations

27
Future Directions

• Continued work on automatic rule learning
  (especially Seeded Version Space Learning)
• Use Hebrew and Hindi systems as test platforms
  for experimenting with advanced learning research
• Rule Refinement via interaction with bilingual
  speakers
• Developing a well-founded model for assigning
  scores (probabilities) to transfer rules
• Redesigning and improving decoder to better fit
  the specific characteristics of the XFER model
• Improved leveraging from manual grammar resources
• MEMT with improved
• Combination of output from different translation
  engines with different confidence scores
• strong decoding capabilities

28
Flat Seed Generation

• Create a transfer rule that is specific to the
  sentence pair, but abstracted to the POS level.
  No syntactic structure.

29
Compositionality - Overview

• Traverse the c-structure of the English sentence,
  add compositional structure for translatable
  chunks
• Adjust constituent sequences, alignments
• Remove unnecessary constraints, i.e. those that
  are contained in the lower-level rule

30
Seeded Version Space Learning Overview

• Goal add appropriate feature constraints to the
  acquired rules
• Methodology
• Preserve general structural transfer
• Learn specific feature constraints from example
  set
• Seed rules are grouped into clusters of similar
  transfer structure (type, constituent sequences,
  alignments)
• Each cluster forms a version space a partially
  ordered hypothesis space with a specific and a
  general boundary
• The seed rules in a group form the specific
  boundary of a version space
• The general boundary is the (implicit) transfer
  rule with the same type, constituent sequences,
  and alignments, but no feature constraints

31
Seeded Version Space Learning Generalization

• The partial order of the version space
• Definition A transfer rule tr1 is strictly more
  general than another transfer rule tr2 if all
  f-structures that are satisfied by tr2 are also
  satisfied by tr1.
• Generalize rules by merging them
• Deletion of constraint
• Raising two value constraints to an agreement
  constraint, e.g.
• ((x1 num) pl), ((x3 num) pl) ?
• ((x1 num) (x3 num))

32
Seeded Version Space Learning

• NP v det n NP VP
• Group seed rules into version spaces as above.
• Make use of partial order of rules in version
  space. Partial order is defined
• via the f-structures satisfying the constraints.
• Generalize in the space by repeated merging of
  rules
• Deletion of constraint
• Moving value constraints to agreement
  constraints, e.g.
• ((x1 num) pl), ((x3 num) pl) ?
• ((x1 num) (x3 num)
• 4. Check translation power of generalized rules
  against sentence pairs

33
Seeded Version Space LearningThe Search

• The Seeded Version Space algorithm itself is the
  repeated generalization of rules by merging
• A merge is successful if the set of sentences
  that can correctly be translated with the merged
  rule is a superset of the union of sets that can
  be translated with the unmerged rules, i.e. check
  power of rule
• Merge until no more successful merges

34
Conclusions

• Transfer rules (both manual and learned) offer
  significant contributions that can complement
  existing data-driven approaches
• Also in medium and large data settings?
• Initial steps to development of a statistically
  grounded transfer-based MT system with
• Rules that are scored based on a well-founded
  probability model
• Strong and effective decoding that incorporates
  the most advanced techniques used in SMT decoding
• Working from the opposite end of research on
  incorporating models of syntax into standard
  SMT systems Knight et al
• Our direction makes sense in the limited data
  scenario

35
AVENUE Architecture
Run-Time Module
Learning Module
SL Input
SL Parser
Morphology Pre-proc
Elicitation Process
Transfer Rule Learning
Transfer Rules
Transfer Engine
TL Output
TL Generator
Decoder
User
36
Learning Transfer-Rules for Languages with
Limited Resources

• Rationale
• Large bilingual corpora not available
• Bilingual native informant(s) can translate and
  align a small pre-designed elicitation corpus,
  using elicitation tool
• Elicitation corpus designed to be typologically
  comprehensive and compositional
• Transfer-rule engine and new learning approach
  support acquisition of generalized transfer-rules
  from the data

37
The Elicitation Corpus

• Translated, aligned by bilingual informant
• Corpus consists of linguistically diverse
  constructions
• Based on elicitation and documentation work of
  field linguists (e.g. Comrie 1977, Bouquiaux
  1992)
• Organized compositionally elicit simple
  structures first, then use them as building
  blocks
• Goal minimize size, maximize linguistic coverage

38
The Transfer Engine
39
Transfer Rule Formalism
SL the man, TL der Mann NPNP DET N -gt
DET N ( (X1Y1) (X2Y2) ((X1 AGR)
3-SING) ((X1 DEF DEF) ((X2 AGR)
3-SING) ((X2 COUNT) ) ((Y1 AGR)
3-SING) ((Y1 DEF) DEF) ((Y2 AGR)
3-SING) ((Y2 GENDER) (Y1 GENDER)) )

• Type information
• Part-of-speech/constituent information
• Alignments
• x-side constraints
• y-side constraints
• xy-constraints,
• e.g. ((Y1 AGR) (X1 AGR))

40
Transfer Rule Formalism (II)
SL the man, TL der Mann NPNP DET N -gt
DET N ( (X1Y1) (X2Y2) ((X1 AGR)
3-SING) ((X1 DEF DEF) ((X2 AGR)
3-SING) ((X2 COUNT) ) ((Y1 AGR)
3-SING) ((Y1 DEF) DEF) ((Y2 AGR)
3-SING) ((Y2 GENDER) (Y1 GENDER)) )

• Value constraints
• Agreement constraints

41
Rule Learning - Overview

• Goal Acquire Syntactic Transfer Rules
• Use available knowledge from the source side
  (grammatical structure)
• Three steps
• Flat Seed Generation first guesses at transfer
  rules flat syntactic structure
• Compositionality use previously learned rules to
  add hierarchical structure
• Seeded Version Space Learning refine rules by
  generalizing with validation (learn appropriate
  feature constraints)

42
Examples of Learned Rules (I)
43
A Limited Data Scenario for Hindi-to-English

• Put together a scenario with miserly data
  resources
• Elicited Data corpus 17589 phrases
• Cleaned portion (top 12) of LDC dictionary
  2725 Hindi words (23612 translation pairs)
• Manually acquired resources during the SLE
• 500 manual bigram translations
• 72 manually written phrase transfer rules
• 105 manually written postposition rules
• 48 manually written time expression rules
• No additional parallel text!!

44
Manual Grammar Development

• Covers mostly NPs, PPs and VPs (verb complexes)
• 70 grammar rules, covering basic and recursive
  NPs and PPs, verb complexes of main tenses in
  Hindi (developed in two weeks)

45
Manual Transfer Rules Example
PASSIVE OF SIMPLE PAST (NO AUX) WITH LIGHT
VERB passive of 43 (7b) VP,28 VPVP V V
V -gt Aux V ( (X1Y2) ((x1 form) root)
((x2 type) c light) ((x2 form) part) ((x2
aspect) perf) ((x3 lexwx) 'jAnA') ((x3
form) part) ((x3 aspect) perf) (x0 x1)
((y1 lex) be) ((y1 tense) past) ((y1 agr
num) (x3 agr num)) ((y1 agr pers) (x3 agr
pers)) ((y2 form) part) )
46
Manual Transfer Rules Example
NP PP NP1 NP P Adj N
N1 ke eka aXyAya N
jIvana
NP NP1 PP Adj N
P NP one chapter of N1
N life
NP1 ke NP2 -gt NP2 of NP1 Ex jIvana ke
eka aXyAya life of (one) chapter
gt a chapter of life NP,12 NPNP PP
NP1 -gt NP1 PP ( (X1Y2) (X2Y1) ((x2
lexwx) 'kA') ) NP,13 NPNP NP1 -gt
NP1 ( (X1Y1) ) PP,12 PPPP NP Postp
-gt Prep NP ( (X1Y2) (X2Y1) )
47
Adding a Strong Decoder

• XFER system produces a full lattice
• Edges are scored using word-to-word translation
  probabilities, trained from the limited bilingual
  data
• Decoder uses an English LM (70m words)
• Decoder can also reorder words or phrases (up to
  4 positions ahead)
• For XFER(strong) , ONLY edges from basic XFER
  system are used!

48
Testing Conditions

• Tested on section of JHU provided data 258
  sentences with four reference translations
• SMT system (stand-alone)
• EBMT system (stand-alone)
• XFER system (naïve decoding)
• XFER system with strong decoder
• No grammar rules (baseline)
• Manually developed grammar rules
• Automatically learned grammar rules
• XFERSMT with strong decoder (MEMT)

49
Results on JHU Test Set (very miserly training
data)
50
Effect of Reordering in the Decoder

51
Observations and Lessons (I)

• XFER with strong decoder outperformed SMT even
  without any grammar rules in the miserly data
  scenario
• SMT Trained on elicited phrases that are very
  short
• SMT has insufficient data to train more
  discriminative translation probabilities
• XFER takes advantage of Morphology
• Token coverage without morphology 0.6989
• Token coverage with morphology 0.7892
• Manual grammar currently somewhat better than
  automatically learned grammar
• Learned rules did not yet use version-space
  learning
• Large room for improvement on learning rules
• Importance of effective well-founded scoring of
  learned rules

52
Observations and Lessons (II)

• MEMT (XFER and SMT) based on strong decoder
  produced best results in the miserly scenario.
• Reordering within the decoder provided very
  significant score improvements
• Much room for more sophisticated grammar rules
• Strong decoder can carry some of the reordering
  burden

53
Conclusions

• Transfer rules (both manual and learned) offer
  significant contributions that can complement
  existing data-driven approaches
• Also in medium and large data settings?
• Initial steps to development of a statistically
  grounded transfer-based MT system with
• Rules that are scored based on a well-founded
  probability model
• Strong and effective decoding that incorporates
  the most advanced techniques used in SMT decoding
• Working from the opposite end of research on
  incorporating models of syntax into standard
  SMT systems Knight et al
• Our direction makes sense in the limited data
  scenario

54
Future Directions

• Continued work on automatic rule learning
  (especially Seeded Version Space Learning)
• Improved leveraging from manual grammar
  resources, interaction with bilingual speakers
• Developing a well-founded model for assigning
  scores (probabilities) to transfer rules
• Improving the strong decoder to better fit the
  specific characteristics of the XFER model
• MEMT with improved
• Combination of output from different translation
  engines with different scorings
• strong decoding capabilities

55
Rule Learning - Overview

• Goal Acquire Syntactic Transfer Rules
• Use available knowledge from the source side
  (grammatical structure)
• Three steps
• Flat Seed Generation first guesses at transfer
  rules no syntactic structure
• Compositionality use previously learned rules to
  add structure
• Seeded Version Space Learning refine rules by
  generalizing with validation

56
Flat Seed Generation

• Create a transfer rule that is specific to the
  sentence pair, but abstracted to the POS level.
  No syntactic structure.

57
Flat Seed Generation - Example

• The highly qualified applicant did not accept the
  offer.
• Der äußerst qualifizierte Bewerber nahm das
  Angebot nicht an.
• ((1,1),(2,2),(3,3),(4,4),(6,8),(7,5),(7,9),(8,6),(
  9,7))

SS det adv adj n aux neg v det n -gt det adv
adj n v det n neg vpart (alignments (x1y1)(x2
y2)(x3y3)(x4y4)(x6y8)(x7y5)(x7y9)(x8
y6)(x9y7)) constraints ((x1 def) ) ((x4
agr) 3-sing) ((x5 tense) past) . ((y1
def) ) ((y3 case) nom) ((y4 agr)
3-sing) . )
58
Compositionality - Overview

• Traverse the c-structure of the English sentence,
  add compositional structure for translatable
  chunks
• Adjust constituent sequences, alignments
• Remove unnecessary constraints, i.e. those that
  are contained in the lower-level rule
• Adjust constraints use f-structure of correct
  translation vs. f-structure of incorrect
  translations to introduce context constraints

59
Compositionality - Example

SS det adv adj n aux neg v det n -gt det adv
adj n v det n neg vpart (alignments (x1y1)(x2
y2)(x3y3)(x4y4)(x6y8)(x7y5)(x7y9)(x8
y6)(x9y7)) constraints ((x1 def) ) ((x4
agr) 3-sing) ((x5 tense) past) . ((y1
def) ) ((y3 case) nom) ((y4 agr)
3-sing) . )
NPNP det AJDP n -gt det ADJP
n ((x1y1) ((y3 agr) 3-sing) ((x3 agr
3-sing) .)
SS NP aux neg v det n -gt NP v det n neg
vpart (alignments (x1y1)(x3y5)(x4y2)(x4
y6)(x5y3)(x6y4) constraints ((x2 tense)
past) . ((y1 def) ) ((y1 case) nom) .
)
60
Seeded Version Space Learning Overview

• Goal further generalize the acquired rules
• Methodology
• Preserve general structural transfer
• Consider relaxing specific feature constraints
• Seed rules are grouped into clusters of similar
  transfer structure (type, constituent sequences,
  alignments)
• Each cluster forms a version space a partially
  ordered hypothesis space with a specific and a
  general boundary
• The seed rules in a group form the specific
  boundary of a version space
• The general boundary is the (implicit) transfer
  rule with the same type, constituent sequences,
  and alignments, but no feature constraints

61
Seeded Version Space Learning

• NP v det n NP VP
• Group seed rules into version spaces as above.
• Make use of partial order of rules in version
  space. Partial order is defined
• via the f-structures satisfying the constraints.
• Generalize in the space by repeated merging of
  rules
• Deletion of constraint
• Moving value constraints to agreement
  constraints, e.g.
• ((x1 num) pl), ((x3 num) pl) ?
• ((x1 num) (x3 num)
• 4. Check translation power of generalized rules
  against sentence pairs

62
Seeded Version Space Learning Example
SS NP aux neg v det n -gt NP v det n neg
vpart (alignments (x1y1)(x3y5)(x4y2)(x4
y6)(x5y3)(x6y4) constraints ((x2 tense)
past) . ((y1 def) ) ((y1 case) nom)
((y1 agr) 3-sing) ) ((y3 agr) 3-sing)
((y4 agr) 3-sing) )
SS NP aux neg v det n -gt NP n det n neg
vpart ( alignments (x1y1)(x3y5) (x4y2)(x
4y6) (x5y3)(x6y4) constraints ((x2
tense) past) ((y1 def) ) ((y1 case)
nom) ((y4 agr) (y3 agr)) )
SS NP aux neg v det n -gt NP v det n neg
vpart (alignments (x1y1)(x3y5)(x4y2)(x4
y6)(x5y3)(x6y4) constraints ((x2 tense)
past) ((y1 def) ) ((y1 case) nom) ((y1
agr) 3-plu) ((y3 agr) 3-plu) ((y4 agr)
3-plu) )
63
Preliminary Evaluation

• English to German
• Corpus of 141 ADJPs, simple NPs and sentences
• 10-fold cross-validation experiment
• Goals
• Do we learn useful transfer rules?
• Does Compositionality improve generalization?
• Does VS-learning improve generalization?

64
Summary of Results

• Average translation accuracy on cross-validation
  test set was 62
• Without VS-learning 43
• Without Compositionality 57
• Average number of VSs 24
• Average number of sents per VS 3.8
• Average number of merges per VS 1.6
• Percent of compositional rules 34

65
Conclusions

• New paradigm for learning transfer rules from
  pre-designed elicitation corpus
• Geared toward languages with very limited
  resources
• Preliminary experiments validate approach
  compositionality and VS-learning improve
  generalization

66
Future Work

• Larger, more diverse elicitation corpus
• Additional languages (Mapudungun)
• Less information on TL side
• Reverse translation direction
• Refine the various algorithms
• Operators for VS generalization
• Generalization VS search
• Layers for compositionality
• User interactive verification

67
Seeded Version Space Learning Generalization

• The partial order of the version space
• Definition A transfer rule tr1 is strictly more
  general than another transfer rule tr2 if all
  f-structures that are satisfied by tr2 are also
  satisfied by tr1.
• Generalize rules by merging them
• Deletion of constraint
• Raising two value constraints to an agreement
  constraint, e.g.
• ((x1 num) pl), ((x3 num) pl) ?
• ((x1 num) (x3 num))

68
Seeded Version Space Learning Merging Two Rules

• Merging algorithm proceeds in three steps.
• To merge tr1 and tr2 into trmerged
• Copy all constraints that are both in tr1 and tr2
  into trmerged
• Consider tr1 and tr2 separately. For the
  remaining constraints in tr1 and tr2 , perform
  all possible instances of raising value
  constraints to agreement constraints.
• Repeat step 1.

69
Seeded Version Space LearningThe Search

• The Seeded Version Space algorithm itself is the
  repeated generalization of rules by merging
• A merge is successful if the set of sentences
  that can correctly be translated with the merged
  rule is a superset of the union of sets that can
  be translated with the unmerged rules, i.e. check
  power of rule
• Merge until no more successful merges