+ All Categories
Home > Documents > Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

Date post: 26-Dec-2015
Category:
Upload: silvia-elliott
View: 221 times
Download: 0 times
Share this document with a friend
21
Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP taggin g JJ ↓
Transcript
Page 1: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

Robert Hass

CIS 630

April 14, 2010

NP

NP↓

Super

NP

tagging

JJ ↓

Page 2: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

OUTLINE

1. Why

2. What

3. How

4. However

Page 3: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

WHY SUPERTAG?

If lexical items have more description associated with them, parsing is easier Only useful if the supertag space is non-huge

We can make it so

Straightforward to compile parse from accurate supertagging But impossible if there are any supertag errors

Well, we can account for some supertag errors Don’t always want a full parse anyway

Page 4: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

WHAT IS SUPERTAGGING?

Not parsing But almost Think PoS tagging…

Word-by-word

Probabilistic Etc.

super tagging

…but with more structural information

• LTAG-like trees

Page 5: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

WHAT IS SUPERTAGGING?

Systematic assignment of supertags …

Page 6: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

WHAT IS SUPERTAGGING?

Supertags are:

Statistically selected Robust Tends to work

Linguistically motivated This makes sense

Page 7: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

WHAT IS SUPERTAGGING?

Many supertags for each word Extended Domain of Locality

Each lexical item has one supertag for every context it appears in

Page 8: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

…so which of these trees do we choose? Well, some clearly don’t work in certain

sentences

Page 9: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

“Alice opened her eyes and saw.” Supertags:

Verb Transitive verb Intransitive verb Infinitive verb …

Noun Noun phrase (subject) Nominal predicative Nominal modifier Nominal predicative subject extraction …

Page 10: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

“Alice opened her eyes and saw.” Supertags:

Verb Transitive verb Intransitive verb Infinitive verb …

Noun Noun phrase (subject) Nominal predicative Nominal modifier Nominal predicative subject extraction …

VP

NP↓sawNP↓

S

Page 11: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

A supertag can be ruled out for a given word in a given input string… If the supertag’s span to the left or right is

excessively large

Page 12: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

“Alice opened her eyes and saw.” Supertags:

Verb Transitive verb Intransitive verb Infinitive verb …

Noun Noun phrase (subject) Nominal predicative Nominal modifier Nominal predicative subject extraction …

to saw…

Page 13: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

A supertag can be ruled out for a given word in a given input string… If the supertag’s span to the left or right is larger

than the input string allows If the supertag contains terminals not found in

the input string

Page 14: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

This works fairly well 50% average reduction in number of possible

supertags

Page 15: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

…but there’s more to be done Good: average number of possible supertags per

word reduced from 47 to 25 Bad: average of 25 possible supertags per word

Page 16: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

Disambiguation by dependency analysis? Lack necessary large corpora This is pretty much parsing

Page 17: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

Disambiguation by unigrams? Give each word its most frequent supertag after

PoS tagging ~75% accurate

Better results than one might expect given large number of possible supertags

Common words (determiners, etc.) usually correct This helps accuracy

Back off to PoS for unknown words Also usually correct

Page 18: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOW TO SUPERTAG

Disambiguation by n-grams?

We assume that subsequent words are independent

Trigrams plus Good-Turing smoothing Accuracy around 90%

Versus 75% from unigrams Contextual information more important than lexical

Reversal of trend for PoS tagging

Page 19: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOWEVER…

Correctly supertagged text yields a 30X parsing speedup But even one mistake can cause parsing to fail

completely This is rather likely

Solution: n-best supertags? When n=3, we get up to 96% accuracy…

Still not great for parsing

Page 20: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

HOWEVER…

But that isn’t fatal Supertagging alone is enough for many

applications Information retrieval Information extraction Text simplification Language modeling

Plus, more robust parsers could account for errors in supertagging …Maybe

Page 21: Robert Hass CIS 630 April 14, 2010 NP NP↓ Super NP tagging JJ ↓

Recommended