1 quasi-synchronous grammars based on key observations in mt: translated sentences often have some...
Post on 19-Dec-2015
216 views
TRANSCRIPT
![Page 1: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/1.jpg)
1
Quasi-Synchronous Grammars Based on key observations in MT:
translated sentences often have some isomorphic syntactic structure, but not usually in entirety.
the strictness of the isomorphism may vary across words or syntactic rules.
Key idea: Unlike some synchronous grammars (e.g. SCFG,
which is more strict and rigid), QG defines a monolingual grammar for the target tree, “inspired” by the source tree.
![Page 2: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/2.jpg)
2
Quasi-Synchronous Grammars In other words, we model the generation of
the target tree, influenced by the source tree (and their alignment)
QA can be thought of as extremely free monolingual translation.
The linkage between question and answer trees in QA is looser than in MT, which gives a bigger edge to QG.
![Page 3: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/3.jpg)
3
Model Works on labeled dependency parse trees Learn the hidden structure (alignment between Q and
A trees) by summing out ALL possible alignments
One particular alignment tells us both the syntactic configurations and the word-to-word semantic correspondences
An example…
question answer
answerparse tree
questionparse tree
an alignment
![Page 4: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/4.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 5: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/5.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 6: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/6.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
isVB
Q: A:$
root$
root
root root
subj with
nmod
nmod
root)|P(root
noNE)|P(noNE
VBD)| P(VB
Our model makes local Markov assumptions to allow efficient computation via Dynamic Programming (details in paper)
given its parent, a word is independent of all other words (including siblings).
![Page 7: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/7.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
isVB
Q: A:$
root$
root
root
subj
root
subj with
nmod
nmod
child)-parent|P(subj
person)|P(qword
NNP)|P(WP
![Page 8: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/8.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
Q: A:$
root$
root
root
subj obj
root
subj with
nmod
nmod
child)-tgrandparen|P(obj
noNE)|P(noNE
NN)|P(NN
![Page 9: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/9.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
Q: A:$
root$
root
root
subj obj
det
root
subj with
nmod
nmod
)word-same|P(det
noNE)|P(noNE
N)|P(DT
![Page 10: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/10.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
)child-parent|P(of
location)|P(location
JJ)|P(NNP
![Page 11: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/11.jpg)
11
6 types of syntactic configurations
Parent-child
![Page 12: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/12.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 13: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/13.jpg)
Parent-child configuration
![Page 14: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/14.jpg)
14
6 types of syntactic configurations
Parent-child Same-word
![Page 15: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/15.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 16: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/16.jpg)
Same-word configuration
Parent-child configuration
![Page 17: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/17.jpg)
17
6 types of syntactic configurations
Parent-child Same-word Grandparent-child
![Page 18: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/18.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 19: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/19.jpg)
Parent-child configuration Same-word configuration
Grandparent-child configuration
![Page 20: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/20.jpg)
20
6 types of syntactic configurations
Parent-child Same-word Grandparent-child Child-parent Siblings C-command(Same as [D. Smith & Eisner ’06])
![Page 21: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/21.jpg)
Parent-child configuration Same-word configuration Grandparent-child configuration
Child-parent configuration Siblings configuration C-command configuration
![Page 22: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/22.jpg)
22
Modeling alignment Base model
)child-parent|P(of
location)|P(location
N)|P(N
![Page 23: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/23.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 24: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/24.jpg)
BushNNP
person
metVBD
FrenchJJ
location
presidentNN
Jacques ChiracNNP
person
whoWP
qword
leaderNN
isVB
theDT
FranceNNP
location
Q: A:$
root$
root
root
subj obj
det of
root
subj with
nmod
nmod
![Page 25: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/25.jpg)
25
Modeling alignment cont.
Base model
Log-linear modelLexical-semantic features from WordNet,Identity, hypernym, synonym, entailment, etc.
Mixture model
![Page 26: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/26.jpg)
26
Parameter estimation
Things to be learnt Multinomial distributions in base model Log-linear model feature weights Mixture coefficient
Training involves summing out hidden structures, thus non-convex.
Solved using conditional Expectation-Maximization
![Page 27: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/27.jpg)
27
Experiments
Trec8-12 data set for training Trec13 questions for development
and testing
![Page 28: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/28.jpg)
28
Candidate answer generation
For each question, we take all documents from the TREC doc pool, and extract sentences that contain at least one non-stop keywords from the question.
For computational reasons (parsing speed, etc.), we only took answer sentences <= 40 words.
![Page 29: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/29.jpg)
29
Dataset statistics Manually labeled 100 questions for training
Total: 348 positive Q/A pairs 84 questions for dev
Total: 1415 Q/A pairs 3.1+, 17.1-
100 questions for testing Total: 1703 Q/A pairs 3.6+, 20.0-
Automatically labeled another 2193 questions to create a noisy training set, for evaluating model robustness
![Page 30: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/30.jpg)
30
Experiments cont.
Each question and answer sentence is tokenized, POS tagged (MX-POST), parsed (MSTParser) and labeled with named-entity tags (Identifinder)
![Page 31: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/31.jpg)
31
Baseline systems (replications) [Cui et al. SIGIR ‘05]
The algorithm behind one of the best performing systems in TREC evaluations.
It uses a mutual information-inspired score computed over dependency trees and a single fixed alignment between them.
[Punyakanok et al. NLE ’04] measures the similarity between Q and A by
computing tree edit distance. Both baselines are high-performing, syntax-based,
and most straight-forward to replicate We further enhanced the algorithms by augmenting
them with WordNet.
![Page 32: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/32.jpg)
32
ResultsMean Average
PrecisionMean Reciprocal
Rank of Top 1
Statistically significantly better than the 2nd best score in each column
28.2% 23.9% 41.2% 30.3%
![Page 33: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/33.jpg)
33
Summing vs. Max
![Page 34: 1 Quasi-Synchronous Grammars Based on key observations in MT: translated sentences often have some isomorphic syntactic structure, but not usually in](https://reader030.vdocument.in/reader030/viewer/2022032800/56649d3a5503460f94a14ec4/html5/thumbnails/34.jpg)
34
Switching back
Tree-edit CRFs