1. The Linguistic-Core Approach to Structured Translation and Analysis of Low-Resource Languages
November 2014 Project Review at ARL
Jaime Carbonell (CMU) & Team
MURI via ARO (PM: Joseph Myers)

2. The Faculty CMU: USC-ISI: MIT: UT Austin: Jaime Carbonell Kevin Knight
David Chiang
(Notre Dame)
MIT: UT Austin:
Regina Barzilay Jason Baldridge
Lori Levin
Noah Smith
Chris Dyer
Supporting roles: 2 other PhDs, 8 Grad Students, 3 Postdocs, N UGs,

3. LCMT: The Elevator Pitch
The fundamental challenge
“Modern” MT requires massive parallel data
There are L’s with scant ||-data
Rule-based MT requires extensive trained-linguist efforts
The linguistic-core approach
Goal: 90% linguistic benefit with 10% linguist effort
Annotation deep and light, linguistics  “lay” bilinguals
Augmented with machine learning from bi & mono-L text
Accomplishments to date
Theory: GFL, graph-semantics, AMR & other parsers, sparse ML training, linguistically-anchored models, …  40+ papers
Tool suites: GFL, TurboParser, MT-in-works, Morph, SuperTag,…
Languages: Kiriwanda, Malagasy, Swhhili, Yoruba

4. The Setting MURI Languages Swahili Morpho-syntactics Anamwona
Kinyarwanda
Bantu (7.5M speakers)
Malagasy
Malayo-Polynesian (14.5M)
Swahili
Bantu (5M native, 150M 2nd/3rd)
Yoruba
Niger-Congo (22+M)
Swahili
Anamwona
“he is seeing him/her”
Morpho-syntactics

5. Which MT Paradigms are Best? Towards Filling the Table
Target
Large T
Med T
Small T
Large S
SMT
LCMT
Med S
???
Small S
Source
“Old” DARPA MT: Large S  Large T
Arabic  English; Chinese  English

6. Evolutionary Tree of MT Paradigms Leading up to LCMT
Large-scale TMT
Large-scale TMT
Transfer MT
Transfer MT w stat phrases
Interlingua MT
Context-Based MT
LCMT
Analogy MT
Example-based MT
SMT with syntax
DecodingMT
Statistical MT
Phrasal SMT
1950
1980
2014

7. Linguistically omnivorous parsing
GFL annotated corpus
CMU, Texas
Parsers
Linguistic
universals
He has been writing a letter.
Dependencies
(j / join :ARG0 (p / person :name (p2 / name :op1 "Pierre" :op2 "Vinken") :age (t / temporal-quantity :quant :unit (y / year))) :ARG1 (b / board) :prep-as (d2 / director :mod (e / executive :polarity -)) :time (d / date-entity :month 11 :day 29))
Abstract Meaning Reps
ISI
CMU
MIT
Texas, MIT
Small CCG
Lexicon
Texas
Unannotated
corpus

8. Original Vision Linguistic Core Team Data:
Data selection for annotation
MT Visualizations and logs
Linguistic Core Team
(LL, JB, SV, JC)
MT Error Analysis
Data:
Parallel
Monolingual
Elicited
Related language
Multi-parallel
Comparable
Elicitation corpus
Hand-built Linguistic Core
Parser, Taggers, Morph. Analyzers
MT Features
MT Systems Team (KK, DC, SV, JC)
Triple Gold Data
MT Systems
Triple Ungold Data
Linguistic Analyzers Team (NS, RB, JB)
Inference Algorithms

9. Current Vision Linguistic Core Team Data:
Data selection for annotation
Complex Morph Analyzers
Linguistic Core Team
(LL, JB, CD, JC)
MT/TA Error Analysis
Data:
Parallel
Monolingual
Elicited
Related language
Multi-parallel
Comparable
Elicitation corpus
Hand-built Linguistic Core
Parser, Taggers, Semantic analyzers
Dependency parses
MT Systems Team (KK, DC, CD, JC)
Triple Gold and GFL annotated
MT Systems and TA modules
String/tree/graph transducers
Linguistic Analyzers Team (NS, RB, JB)
Semantic Parsing Algorithms
Definiteness/
Discourse

10. PFA Node Alignment Algorithm Example
Tree-tree aligner enforces equivalence constraints and optimizes over terminal alignment scores (words/phrases)
Resulting aligned nodes are highlighted in figure
Transfer rules are partially lexicalized and read off tree.
Proprietary and Confidential 10

11. LCMT: NLP Workflow and Tools
GFL annotator Framework
CMU (current)
CMU
Supervised POS Taggers
Annotated Data
CMU
CMU + Texas
ISI
Supervised Dependency & AMR Parsers
Tree-Graph Syn/Sem trx
Texas
MIT
CMU
Unannotated Data
Semisupervised POS Taggers
Semisupervised Dependency Parsers
Texas
MIT
Unsupervised Dependency Parsers
= Toolsuite software (more to come)

12. Machine Translation Paradigms
Phrase-based MT (LCMT 20+% of effort)
Morph-Syntax-based MT (LCMT 30+%)
Meaning-based MT (LCMT 40+%)
source
string
target
string
NIST 2009 c2e
source
string
source
tree
target
tree
target
string
source
string
source
tree
meaning
representation
target
tree
target
string

13. Some Key Results to Date
Theory of transducers (string, tree, graph)
Massive Lexical borrowing across diverse languages
Linguistic universals
Dependencies, semantic roles, conservation, AMR, discourse, …
Statistical learning over strings, trees, graphs
Bayesian, HMM/CRF, active sampling  model parameters
Parsing into deep semantics (AMR)
MT demonstrations: Focus on M, K, S Y, but also across ~20 languages (WMT honors, synthetic phrases)
A suite of 11 serious software modules and tools (morphology, variable-depth linguistic annotation, dependency parsing, MT, …)
Current scientific challenges
Is general graph topology induction possible?
Bridging structural divergences via semi-universals?
Semantic invariance: lexical, structural, non-propositional?

14. List of “Firsts” for the Linguistic Core
First use of models incorporating linguistic knowledge in the form of hand-written morpho- grammatical rules combined with limited-volume corpus statistics First use of models of lexical "borrowing" from other (major) languages to improve translation and analysis of low resource languages (publication in prep). First efficient and exact probabilistic model for structured prediction with arbitrary syntactic and semantic dependencies derived from the input language. First exploitation of large monolingual foreign text collections (vs bilingually translated collections) to improve low-density MT, via treating foreign text as a mapped/encoded version of English. First application of formal graph transduction theory to natural language analysis; earlier efforts applied to string transduction and tree transduction theory only. First substantial corpora annotated cheaply by novices used to build effective NLP tools First statistical parser to map language into abstract meaning representation of semantics First to show that for resource-impoverished languages, a multilingual parser based on language universals outperforms a target language parser target language First analyses to prove formally and empirically demonstrate that inference in dependency parsing is computationally easy on average case (despite NP-hard for the worst case).

15. External Honors for the LC Project
Best human judgments of English-Russian translations at WMT2013 Best BLEU on Hindi-English translation at WMT2014 Best student paper, ACL 2014 Low-Rank Tensors for Scoring Dependency Structures. Tao Lei, Yu Xin, Yuan Zhang, Regina Barzilay and Tommi Jaakkola. Best paper, honorable mention, ACL 2014 A Discriminative Graph-Based Parser for the Abstract Meaning Representation. Jeffrey Flanigan, Sam Thomson, Jaime Carbonell, Chris Dyer and Noah A. Smith. Best paper, runner up, EMNLP 2014 Language Modeling with Power Low Rank Ensembles. Best paper (one of four), NIPS 2014 Conditional Random Field Autoencoders for Unsupervised Structured Prediction. Waleed Ammar, Chris Dyer, and Noah A. Smith

16. Lexical Borrowing of Common Words

17. Swahili morphology using a crowdsourced lexicon
Patrick Littell, Lori Levin, Chris Dyer
No provenance: the root of the word was collected by hand.
[GUESS1]: the root is inferred from the Kamusi lexicon part of speech tag including noun class.
[GUESS2]: the root is from Kamusi, but no noun class is given.
[GUESS3]: possible English loan word
[GUESS4]: complete guess
FST written by Patrick Littell
Lexicon extracted from dictionaries and textbooks

18. Parsing Progress (F1)
On CoNNL Dataset: (CMU), (MIT)

19. CCG Supertagging the lazy dogs wander np/n n/n n n np np n/n (s\np)/np
Linguistically-
Motivated
Priors
the
lazy
dogs
wander
np/n
n/n n n
HMM np np
n/n
(s\np)/np
np/n
s\np …

20. Parsing into AMR (ACL 2014 honorable mention for best paper)
Approximately guards patrol the kilometre border between Russia and Afghanistan.
a2 / approximately
11000
p / patrol-01
g / guard
(d4 / distance-quantity :unit (k2 / kilometer)
1200
b / border
b2 / between
(c / country :name (n / name :op1 "Russia"))
(c2 / country :name (n2 / name :op1 "Afghanistan"))
Add relations
(p / patrol-01
:ARG0 (g / guard
:quant (a2 / approximately
:op ))
:ARG1 (b / border
:quant (d4 / distance-quantity
:unit (k2 / kilometer)
:quant 1200)
:location (b2 / between
:op1 (c / country
:name (n / name
:op1 "Russia"))
:op2 (c2 / country
:name (n2 / name
:op1 "Afghanistan")))))
New Results:
61% F1
CMU and ISI

21. Unsupervised Part-of-Speech Tagging
conditional random field autoencoder
featurized hidden Markov model
classic hidden Markov model
V-measure (higher is better)
Arabic Basque Danish Greek Hungarian Italian Kin. Mal. Turkish Ave.

22. Automatic Classification of the Communicative Functions of Definiteness
Predicted semantic functions of definiteness: 78.2% accuracy
Annotated Corpus
Semantics of Definiteness
Logistic regression classifier
Syntactic features extracted from dependency parser
Why Definiteness:
One instance of non-propositional semantics
Major determinant of word order
Wildly divergent in morpho-syntactic expression
Problems in word alignment and language models

23. Integrating Alignment and Decipherment for Better Low-Density MT
Small bilingual Malagasy/English text
(need to align words [Brown et al 93])
joint
Large Malagasy monolingual text
(need to decipher [Dou & Knight 13])
Decipherment helps
Word Alignment
Decipherment helps
Machine Translation
Bleu
ISI jointly with CMU/Texas/MIT

24. Graph Formalisms for Language Understanding and Generation
String Automata Algorithms
Tree Automata Algorithms
Graph Automata Algorithms
N-best answer extraction
… paths through an WFSA
(Viterbi, 1967; Eppstein, 1998)
… trees in a weighted forest (Jiménez & Marzal, 2000; Huang & Chiang, 2005)
Investigating:
Linguistically adequate
representations
Efficient algorithms
Using them in:
Text  Meaning (NLU)
Meaning  Text (NLG)
Meaning-based MT
Unsupervised EM training
Forward-backward EM (Baum/Welch, 1971; Eisner 2003)
Tree transducer EM training
(Graehl & Knight, 2004)
Determinization, minimization
… of weighted string acceptors (Mohri, 1997)
… of weighted tree acceptors (Borchardt & Vogler, 2003;
May & Knight, 2005)
Intersection
WFSA intersection
Tree acceptor intersection
Application of transducers
string  WFST  WFSA
tree  TT  weighted tree
acceptor
Composition of transducers
WFST composition
(Pereira & Riley, 1996)
Many tree transducers not closed under composition
(Maletti et al 09)
Software tools
Carmel, OpenFST
Tiburon (May & Knight 10)
ISI jointly with CMU

25. Supervision (David) Supervision (Azim)
Tajik Corpus from
Leipzig Archive
Tajik Reference Grammar
(Perry, 2005)
PerLex Persian Lexicon
(Sagot and Walther, 2010)
Tajik and Persian Wikipedias
Persian Treebank (Rasooli et al., 2013)
IPA Converter
(Kartik, Pat, Chris)
Persian-Tajik Converter
(Chris)
Supervision
(David)
Supervision
(Azim)
NE Gold Standard (native speaker)
(Azim)
NE Pyrite Standard (linguist)
(Alexa, Lori)
Morphological
Analyzers
(Swabha, Chris)
Brown
Clusters
(Kartik)
Morphology Lists
for NE
(Alexa, David)
Gazetteers
(Pat, Chris)
Tajik POS Tagger
(Chris)
Named Entity Recognizer
(Kartik, Chris)

26. New Results for Graph Automata for Mapping Between Text and Meaning
Strings
Graphs
FSA
CFG
DAG acceptor
HRG
probabilistic
yes
intersects with finite-state
EM training
transduction
O(n)
O(n3)
O(|Q|T+1n)
O((3dn)T+1)
implemented
d = graph degree for AMR, high in practice
T = treewidth complexity for AMR, low in practice (2-3)

27. Next Steps (high level overview)
Finalize MT systems: K, M, S, Y
Package and make available externally
Possibly integrate with government translator workbench
Compare with Govt systems when available and appropriate (e.g. Malagasy with Carl Rubino)
Complete scientific investigations (Graph transduction, MT with AMR, supertagging  parsing, borrowing++, …)
Document and distribute tool suites (rapid annotation, morphology, CCG supertagging, dependency parsing, AMR parsing, generation, end-to-end MT, lexicon borrowing, ML modules, …)  15 +/-
Publish, publish, publish (40+ papers and counting)
Detailed next steps at the end of each major presentation

28. THANK YOU!
Jaime Carbonell, CMU

29. Select/show as needed for discussion period
Supplementary Slides
Select/show as needed for discussion period

30. Tag Dictionary Generalization
Raw Corpus
________
Type Annotations
____________
the DT
PRE2_th
PRE1_t
SUF1_g
dog NN
TYPE_the
TYPE_thug
TYPE_dog
PREV_<b>
PREV_the
NEXT_walks
TOK_the_4
TOK_the_1
TOK_thug_5
TOK_dog_2
Token Annotations
____________
Any arbitrary features could
be added
the dog walks DT NN
VBZ . .

31. “These 7 people include astronauts coming from France and Russia”
RULE 15:
S(x0:NP, x1:VP, x2:PUNC)
 x0 , x1 , x2
RULE 14:
VP(x0:VBP, x1:NP)
 x0 , x1
“include astronauts coming from
France and Russia”
“astronauts coming from
France and Russia”
RULE 16:
NP(x0:NP, x1:VP)
 x1 , 的 , x0
“coming from France and Russia”
RULE 11:
VP(VBG(coming), PP(IN(from), x0:NP))
 来自 , x0
“France and Russia”
“these 7 people”
RULE 13:
NP(x0:NNP, x1:CC, x2:NNP)
 x0 , x1 , x2
RULE 10:
NP(x0:DT, CD(7), NNS(people)
 x0 , 7人
“these”
“include”
“France”
“&”
“Russia”
“astronauts”
“.”
RULE 1:
DT(these)
 这
RULE 2:
VBP(include)
 中包括
RULE 4:
NNP(France)
 法国
RULE 5:
CC(and)
 和
RULE 6:
NNP(Russia)
 俄罗斯
RULE 8:
NP(NNS(astronauts))
 宇航 , 员
RULE 9:
PUNC(.)
 .
这 7人 中包括 来自 法国 和 俄罗斯 的 宇航 员

32. Model Minimization
<b> The man saw the saw <b>
1.0
1.0
<b> DT NN
VBD
1.0
0.4
1.0
0.6
0.7
0.2
0.8
0.3 . .

33. Linguistically opportunistic parsing
GFL annotated corpus
CMU, Texas
Parsers
Linguistic
universals
He has been writing a letter.
Dependencies
(j / join :ARG0 (p / person :name (p2 / name :op1 "Pierre" :op2 "Vinken") :age (t / temporal-quantity :quant :unit (y / year))) :ARG1 (b / board) :prep-as (d2 / director :mod (e / executive :polarity -)) :time (d / date-entity :month 11 :day 29))
Abstract Meaning Reps
ISI
CMU
MIT
Texas, MIT
Small CCG
Lexicon
Texas
Unannotated
corpus

34. Fragmentary Unlabeled Dependency Grammar (Schneider, O’Connor, Saphra, Bamman, Faruqui, Smith, Dyer, and Baldridge, 2013)
Represents unlabeled dependencies
Special handling for:
multiword expressions
coordination
anaphora
Allows underspecification
Graph fragment language for easy annotation

35. Graph Fragment Language (GFL)
Provide a detailed analysis of coordination…
{Our three} > weapons > are < $a
$a :: {fear surprise efficiency} :: {and~1 and~2}
ruthless > efficiency
(Our three weapons*) > are < (fear surprise and ruthless efficiency)
Or focus just on the high level…
(((Ataon’ < (ny > mpanao < fihetsiketsehana)) < hoe < mpikiky < manko) < (i > Gaddafi)) Atoan’ < noho < (ny_1 > kabariny < lavareny)
Provide detailed syntactic dependency structure
The Malagasy sentence means “Gaddafi has referred to protesters as rodents in his rambling speeches.” The sentence is in the tradition (but typologically unusual) VOS order (it’s actually VOS + PP, where the last PP is “in his rambling speeches”)
Ataon’ < (ny_1 mpanao* fihetsiketsehana)
Atoan’ < (hoe* mpikiky manko)
Atoan’ < (i Gaddafi*) Atoan’ < noho < (ny_2 kabariny lavareny)
Or focus on predictate/arguments
“Gaddafi has referred to protesters as rodents in his rambling speeches.”

36. GFL (CMU/Texas) & AMR (ISI)
The classic: “Pierre Vinken , 61 years old , will join the board as a nonexecutive director Nov. 29 .”
( (S (NP-SBJ (NP (NNP Pierre) (NNP Vinken) ) (, ,) (ADJP (NP (CD 61) (NNS years) ) (JJ old) ) (, ,) ) (VP (MD will) (VP (VB join) (NP (DT the) (NN board) ) (PP-CLR (IN as) (NP (DT a) (JJ nonexecutive) (NN director) )) (NP-TMP (NNP Nov.) (CD 29) ))) (. .) ))
join < [ Pierre Vinken ]join < boardjoin < as < directorjoin < [Nov. 29]nonexecutive > director61 > years > old > [ Pierre Vinken ]
(j / join :ARG0 (p / person :name (p2 / name :op1 "Pierre" :op2 "Vinken") :age (t / temporal-quantity :quant :unit (y / year))) :ARG1 (b / board) :prep-as (d2 / director :mod (e / executive :polarity -)) :time (d / date-entity :month 11 :day 29))

37. Probabilistic Graph Grammars
WITH BASIC RULES LIKE THIS:
WE CAN DERIVE AND TRANSFORM SEMANTIC GRAPHS:
ARG1
instance
instance
ARG1
ARG0
ARG1
instance
ARG0 X
ARG1
instance
WANT
WANT
BELIEVE
ARG0
WANT
“the boy wants
something
involving himself”
instance
GIRL
“the boy wants the girl
to believe he is wanted”
instance B
BOY

38. Example Parsing into AMR
Approximately guards patrol the kilometre border between Russia and Afghanistan.
(p / patrol-01
:ARG0 (g / guard
:quant (a2 / approximately
:op ))
:ARG1 (b / border
:quant (d4 / distance-quantity
:unit (k2 / kilometer)
:quant 1200)
:location (b2 / between
:op1 (c / country
:name (n / name
:op1 "Russia"))
:op2 (c2 / country
:name (n2 / name
:op1 "Afghanistan")))))
a2 / approximately
11000
p / patrol-01
g / guard
(d4 / distance-quantity :unit (k2 / kilometer)
1200
b / border
b2 / between
(c / country :name (n / name :op1 "Russia"))
(c2 / country :name (n2 / name :op1 "Afghanistan"))
Add relations

40. Deciphering Foreign Language
Accuracy of learned
bilingual dictionary
test on Spanish
not translations
of each other
(Dou & Knight 2013)
Dependency-based:
Linguistic analysis helps substantially!
English
text
Foreign
text
Deciphering
Engine
(Dou & Knight 2012)
Ngram-based
bilingual
word-for-word
dictionary
How much foreign text (running words)

41. Constituent Structure Trees
Strength: You can use tests for constituency (movement, deletion, substitution, coordination) to get reproducible results for corpus annotation.
Weakness:
1. Tests for constituency sometimes fail to provide reproducible results.
The five trees (based on an exercise in Radford , 1988) have each been proposed in a published paper and can each be defended by tests for constituency.
2. People do not have uniform intuitions about which tree is “correct”.

42. Morpho-syntactics Iñupiaq (North Slope Alaska)
Tauqsiġñiaġviŋmuŋniaŋitchugut.
‘We won’t go to the store.’

43. Mathematical Foundations for Semantics-Based Machine Translation
Previous MT systems have been based on clean string automata and tree automata
General purpose algorithms have been worked out (in part by MT scientists), with wide applicability
software toolkits even implement those algorithms
But new models of meaning-based MT deal in semantic graph structures
Foreign string  Meaning graph  English string
QUESTION: Do efficient, general-purpose algorithms for graph automata exist to support these linguistic models?

44. General-Purpose Algorithms for Manipulating Linguistic Structures: Acceptors
String Acceptors successfully applied to speech recognition
Tree Acceptors successfully applied to syntax-based MT
Graph Acceptors now being applied to semantics-based MT
Membership checking ...
... of string (length n) in WFSA.
O(n) if WFSA is determinized.
... of tree in forest.
O(n) if determinized.
... of graph in hyperedge-replacement grammar (HERG) (Drewes 97)
New algorithm:
Chiang (forthcoming), O(2dn)k+1 : d & n properties of individual grammar
k-best …
… best k paths through an WFSA with n states and e edges (Viterbi 67; Eppstein 98)
O(e + n log n + k log k)
… trees in a weighted forest (Jiménez & Marzal 00; Huang & Chiang 05)
O(e + n k log k)
... graphs in weighted HERG.
Efficient Huang & Chiang results carry over.
EM training of probabilistic weights
Forward-backward EM (Baum/Welch 71; Eisner 03)
O(n)
Tree acceptor training (Graehl & Knight 04)
Efficient Graehl & Knight results carry over.
Intersection
WFSA intersection
O(n2) classical
Tree acceptor intersection
Graph acceptor intersection
NOT CLOSED (in general)
co-PI supported under MURI project

45. General-Purpose Algorithms for Feature Structures (Graphs)
String World
Tree World
Graph World
Acceptor
Finite-state acceptors
Tree automata
HRG
Transducer
Finite-state transducers
Tree transducers
Synchronous HRG
Membership checking
O(n)
O(n) for trees
O(n3) for strings
O(nk+1) for graphs
N-best …
… paths through an WFSA
(Viterbi, 1967; Eppstein, 1998)
… trees in a weighted forest (Jiménez & Marzal, 2000; Huang & Chiang, 2005)
… graphs in a weighted forest
EM training
Forward-backward EM (Baum/Welch, 1971; Eisner 2003)
Tree transducer EM training
(Graehl & Knight, 2004)
EM on forests of graphs
Intersection
WFSA intersection
Tree acceptor intersection
Not closed
Transducer composition
WFST composition
(Pereira & Riley, 1996)
Many tree transducers not closed under composition
(Maletti et al 09)
General tools
Carmel, OpenFST
Tiburon (May & Knight 10)
Bolinas

46. Functional Collaboration
Data selection for annotation
MT Visualizations and logs
Linguistic Core Team
(LL, JB, SV, JC)
MT Error Analysis
Data:
Parallel
Monolingual
Elicited
Related language
Multi-parallel
Comparable
Elicitation corpus
Hand-built Linguistic Core
Parser, Taggers, Morph. Analyzers
MT Features
MT Systems Team (KK, DC, SV, JC)
Triple Gold Data
MT Systems
Triple Ungold Data
Linguistic Analyzers Team (NS, RB, JB)
Inference Algorithms

47. Malagasy - English Resources
Malagasy Resources
Tokens
Types
Hapax
Bible (Year 1)
579,578
19,460
8,401
Leipzig corpus (Year 2)
618,282
41,462
23,659
CMU Global Voices (Year 2)
2,148,976
84,744
46,627
Total
3,346,836
115,172
62,517
Malagasy - English Resources
eng-Tokens
eng-Types
mlg-Tokens
mlg-Types
Bible (Year 1)
584,872
13,084
579,578
19,460
CMU Global Voices (Year 2)
1,785,472
63,357
2,148,976
84,744
Total
2,370,344
67,790
3,346,836
115,172

48. Evolutionary Tree of MT Paradigms Prior to LCMT
Large-scale TMT
Large-scale TMT
Transfer MT
Transfer MT w stat phrases
Interlingua MT
Context-Based MT
LCMT
Analogy MT
Example-based MT
SMT on syntax struct.
DecodingMT
Statistical MT
Phrasal SMT
1950
1980
2012

49. Model Parameters
Distribution over number of arguments given the parent tag
Weights for selection features, shared across all set sizes
Weights for ordering features
All parameters are shared across languages

50. Malagasy Language Modeling
Data
Seq. X-ent
Word X-ent
Total X-ent.
Perplexity
OOVs
3-gram+char
Bible
10.35
7.66
18.01
264,323
23.94% GV
7.02
1.14
8.16
286.0
3.30%
3-gram+morph
0.90
7.92
241.4
Successes
Malagasy analyzer has << 100% coverage, but we still get substantial gains
Year 3 Goals
Improve word sequence model with morphosyntactic information
Improve coverage of Malagasy morphological phenomena
Incorporation in MT system
Kinyarwanda analyzer/generator under development
Cross entropy, perplexity, and OOV rates were computed on the MT test set.

51. How CMU ISI UT and MIT collaborate
Monthly teleconference calls
Focused on management and project coordination
Technical topics follow when appropriate
Semi-annual face-to-face meetings
Last ones in Nov 2012 and March 2013
Include students/postdocs, etc. Focused on science
Much more frequent focused calls/chats/etc.
Data collection, annotations, SW APIs, brainstorming new algorithms, …
Sharing/reviewing results and papers
Website/repository + shared SW/data sets + papers + more goodies
Student exchanges (e.g. week, month, summer)
Occasional individual faculty trips
Combined research (GFL, AMR parsing, CCG parsing, decipherment,…)