程序代写代做代考 Slide 1

Slide 1

COMPUTATIONAL

LINGUISTICS

Copyright © 2017

Suzanne Stevenson,

Graeme Hirst and Gerald

Penn. All rights reserved.

9B

9B. Supertagging

Gerald Penn

Department of Computer Science, University of Toronto

CSC 2501 / 485

Fall 2018

Based upon slides by Michael Auli, Rober Hass and Aravind Joshi

WHY SUPERTAG?

 If lexical items have more description associated

with them, parsing is easier

 Only useful if the supertag space is not huge

 Straightforward to compile parse from accurate

supertagging

 But impossible if there are any supertag errors

 We can account for some supertag errors

 Don’t always want a full parse anyway

WHAT IS SUPERTAGGING?

 Systematic assignment of supertags

 Supertags are:

 Statistically selected

 Robust

 Tends to work

 Linguistically motivated

 This makes sense

WHAT IS SUPERTAGGING?

 Many supertags for each word

 Extended Domain of Locality

 Each lexical item has one supertag for every syntactic

environment it appears in

 Inspiration comes from LTAG, lexicalized tree-adjoining

grammars, in which all dependencies are localized.

 Generally, agreement features such as number and tense,

are not part of the supertag.

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

 …

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

HOW TO SUPERTAG

 A supertag can be ruled out for a given word in a

given input string…

 Left and/or right context is too long/short for the

input

 If the supertag contains other terminals not found in

the input

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…

HOW TO SUPERTAG

 This works fairly well

 50% average reduction in number of possible

supertags

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

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

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

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…

 Not bad at all for such a simple method

 425 lexical categories (PTB-CFG: ~50)

 12 combinatory rules (PTB-CFG: > 500,000)