Exploring Word Embeddings and Syntax Encoding
How much do word embeddings
encode about syntax?
Jacob Andreas and Dan Klein
UC Berkeley
Everybody loves word embeddings
few
most
that
the
a
each
this
every
[Collobert 2011]
[Collobert 2011, Mikolov 2013, Freitag 2004, Schuetze 1995, Turian 2010]
What might embeddings bring?
Cathleen complained about the magazine’s shoddy editorial quality .
Mary
executive
average
Today’s question
Can word embeddings
trained with surface context
improve a state-of-the-art
constituency parser?
(no)
Embeddings and parsing
•
Pre-trained word embeddings are useful for a
variety of NLP tasks
•
Can they improve a constituency parser?
–
(not very much)
[Cite XX, Cite XX, Cite XX]
Three hypotheses
Vocabulary expansion
(good for OOV words)
Statistic pooling
(good for medium-frequency words)
Embedding structure
(good for features)
Vocabulary expansion:
Embeddings help handling of
out-of-vocabulary words
Vocabulary expansion
John
Mary
Pierre
yellow
enormous
hungry
Cathleen
Vocabulary expansion
John
Mary
Pierre
yellow
enormous
hungry
Cathleen
complained about the magazine’s shoddy editorial quality
.
Cathleen
Mary
Vocab. expansion results
Baseline
+OOV
Vocab. expansion results
Baseline
+OOV
Vocab. expansion results
Baseline
+OOV
(300 sentences)
Statistic pooling hypothesis:
Embeddings help handling of
medium-frequency words
Statistic pooling
executive
kind
giant
editorial
average
{NN, JJ}{NN}{NN, JJ}{JJ}{NN}Statistic pooling
executive
kind
giant
editorial
average
{NN, JJ}{NN, JJ
}
{NN, JJ}{JJ, NN
}
{NN, JJ
}
Statistic pooling
executive
kind
giant
editorial
average
{NN, JJ}{NN}{NN, JJ}{JJ}{NN}editorial
NN
editorial
NN
Statistic pooling results
Baseline
+Pooling
Vocab. expansion results
Baseline
+Pooling
(300 sentences)
Embedding structure hypothesis:
The organization of the embedding space
directly encodes useful features
Embedding structure
vanished
dined
vanishing
dining
devoured
assassinated
devouring
assassinating
“transitivity”
“tense”
VBD
[Huang 2011]
Embedding structure
vanished
dined
vanishing
dining
devoured
assassinated
devouring
assassinating
“transitivity”
“tense”
[Huang XX]
Embedding structure results
Baseline
+Features
Embedding structure results
Baseline
+Features
(300 sentences)
To summarize
(300 sentences)
Combined results
Baseline
+OOV
+Pooling
Vocab. expansion results
Baseline
(300 sentences)
+OOV
+Pooling
What about…
•
Domain adaptation?
(no significant gain)
•
French?
(no significant gain)
•
Other kinds of embeddings?
(no significant gain)
Why didn’t it work?
•
Context clues often provide enough information
to reason around words with incomplete /
incorrect statistics
•
Parser already has a robust OOV, small count
models
•
Sometimes “help” from embeddings is worse
than nothing:
bifurcate
Soap
homered
Paschi
tuning
unrecognized
What about other parsers?
•
Dependency parsers
(continuous repr. as syntactic abstraction)
•
Neural networks
(continuous repr. as structural requirement)
[Henderson 2004, Socher 2013]
[Henderson 2004, Socher 2013, Koo 2008, Bansal 2014]
What didn’t we try?
•
Hard clustering
(some evidence that this is useful for
morphologically rich languages)
•
A nonlinear feature-based model
•
Embeddings in higher constituents
(e.g. in a CRF parser)
[Candito 09]
Conclusion
•
Embeddings provide no apparent benefit to
state-of-the-art parser for:
–
OOV handling
–
Parameter pooling
–
Lexicon features
•
Code online at
http://cs.berkeley.edu/~jda
Word embeddings play a crucial role in natural language processing, offering insights into syntax encoding. Jacob Andreas and Dan Klein from UC Berkeley delve into the impact of embeddings on various linguistic aspects like vocabulary expansion and statistic pooling. Through different hypotheses, they showcase how embeddings aid in handling out-of-vocabulary words and medium-frequency words, ultimately enhancing features like tense and transitivity.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
How much do word embeddings encode about syntax? Jacob Andreas and Dan Klein UC Berkeley
Everybody loves word embeddings fewmost that the aeach every this [Collobert 2011] [Collobert 2011, Mikolov 2013, Freitag 2004, Schuetze 1995, Turian 2010]
What might embeddings bring? Mary Cathleen complained about the magazine s shoddy editorial quality . average executive
Three hypotheses Cathleen Vocabulary expansion (good for OOV words) Mary Statistic pooling (good for medium-frequency words) average editorial executive Embedding structure (good for features) tense transitivity
Vocabulary expansion: Embeddings help handling of out-of-vocabulary words Cathleen Mary
Vocabulary expansion Cathleen yellow Mary John enormous Pierre hungry
Vocabulary expansion Mary Cathleen complained about the magazine s shoddy editorial quality. Cathleen yellow Mary John enormous Pierre hungry
Vocab. expansion results 100 95 91.22 91.13 90 85 80 75 70 65 60 +OOV Baseline
Vocab. expansion results 75 (300 sentences) 74 73 72.20 71.88 72 71 70 +OOV Baseline
Statistic pooling hypothesis: Embeddings help handling of medium-frequency words average editorial executive
Statistic pooling {NN} {NN, JJ} editorial {JJ} executive giant {NN} kind {NN, JJ} average
Statistic pooling {NN, JJ} {NN, JJ} editorial {JJ, NN} executive giant {NN, JJ} kind {NN, JJ} average
Statistic pooling editorial NN {NN} {NN, JJ} editorial {JJ} executive giant {NN} kind {NN, JJ} average editorial NN
Statistic pooling results 100 95 91.13 91.11 90 85 80 75 70 65 60 +Pooling Baseline
Vocab. expansion results 75 (300 sentences) 74 73 72.21 71.88 72 71 70 +Pooling Baseline
Embedding structure hypothesis: The organization of the embedding space directly encodes useful features tense transitivity
Embedding structure transitivity vanishing dining dined vanished tense devoured devouring assassinated assassinating dined VBD dined VBD [Huang 2011]
Embedding structure results 100 95 91.13 91.08 90 85 80 75 70 65 60 +Features Baseline
Embedding structure results 75 (300 sentences) 74 73 71.88 72 71 70.32 70 +Features Baseline
To summarize 100 Baseline 95 +OOV 90 +Pooling +Features 85 80 (300 sentences) 75 70 65 60
Combined results 100 95 90.70 90.11 90 85 80 75 70 65 60 +OOV +Pooling Baseline
Vocab. expansion results 75 (300 sentences) 74 73 72.21 71.88 72 71 70 +OOV +Pooling Baseline
What about Domain adaptation? (no significant gain) French? (no significant gain) Other kinds of embeddings? (no significant gain)
Why didnt it work? Context clues often provide enough information to reason around words with incomplete / incorrect statistics Parser already has a robust OOV, small count models Sometimes help from embeddings is worse than nothing: bifurcate homered tuning Soap Paschi unrecognized
What about other parsers? Dependency parsers (continuous repr. as syntactic abstraction) Neural networks (continuous repr. as structural requirement) [Henderson 2004, Socher 2013] [Henderson 2004, Socher 2013, Koo 2008, Bansal 2014]
Conclusion Embeddings provide no apparent benefit to state-of-the-art parser for: OOV handling Parameter pooling Lexicon features Code online at http://cs.berkeley.edu/~jda
