EMPIRICAL INVESTIGATIONS OF ANAPHORA AND SALIENCE

Plan of the series

Today’s lecture

Preliminary corpus study (Poesio and Vieira, 1998)

Poesio and Vieira, 1998

The Vieira / Poesio system for robust definite description resolution

Methods for resolving direct anaphors

Methods for resolving discourse-new definite descriptions

A `knowledge-based’ classification of bridging descriptions (Vieira, 1998)

… continued
(cases NOT attempted by our system)

Distribution of bridging descriptions

The (hand-coded) decision tree

Overall Results

Overall Results

Questions raised by the Vieira / Poesio work

Mereological bridging references

PREVIOUS RESULTS

FINDING MERONYMICAL RELATIONS USING SYNTACTIC INFORMATION

LEXICAL RESOURCES FOR BRIDGING: A SUMMARY

FOCUSING  AND MEREOLOGICAL BRIDGES

FOCUS (CB) TRACKING + GOOGLE SEARCH (POESIO, 2003)

FINDING MEREOLOGICAL RELATIONS USING GOOGLE

NEW EXPERIMENTS
(Poesio et al, 2004)

More (and reliably annotated) data: the GNOME corpus

METHODS

Lexical Distance 1 (WordNet)

Lexical Distance, 1: WordNet

Lexical Distance 2 (Google)

WN vs GOOGLE

BASELINES

RESULTS
 (58 THE-NPs, 50:50)

MORE RESULTS

MEREOLOGICAL BDS  REALIZED WITH BARE-NPS

HARDER TEST

DISCUSSION

Back to discourse-new detection

GUITAR (Kabadjov, to appear)

DISCOURSE-NEW DEFINITE DESCRIPTIONS

WOULD DNEW RECOGNITION HELP?

SPURIOUS MATCHES

GOALS OF THE WORK

DN CLASSIFIER:
THE UPPER BOUND

VIEIRA AND POESIO’S
DN DETECTORS

VIEIRA AND POESIO’S
DN DETECTORS, II

VIEIRA AND POESIO’S DECISION TREES

VIEIRA AND POESIO’S RESULTS

BEAN AND RILOFF (1999)

BEAN AND RILOFF’S ALGORITHM

BEAN AND RILOFF’S RESULTS

NG AND CARDIE
(2002)

NG AND CARDIE’S METHODS

NG AND CARDIE’S RESULTS

NG AND CARDIE’S RESULTS

URYUPINA’S
 METHODS

URYUPINA’S FEATURES:
WEB-BASED DEFINITE PROBABILITY

URYUPINA’S RESULTS
(DNEW CLASSIFIER)

URYUPINA’S RESULTS
(DNEW CLASSIFIER)

PRELIMINARY CONCLUSIONS

PRELIMINARY CONCLUSIONS, II

A NEW SET OF EXPERIMENTS

ARCHITECTURE

A NEW SET OF FEATURES

A REVISED SET OF FEATURES (II)

LEARNING A DN CLASSIFIER

3 EVALUATIONS

RESULTS:
DNEW CLASSIFICATION

RESULTS:
OVERALL

RESULTS:
DIRECT ANAPHORA RESOLUTION

ERROR ANALYSIS

THE DECISION TREE

RESULTS:
THE VIEIRA/POESIO CORPUS

RESULTS:
AUTOMATIC PARSING

CONCLUSIONS AND DISCUSSION

Empirical studies of anaphora: some conclusions

URLs