STS: - PowerPoint Presentation

Slide1

STS: Temporal and Spatial Constraints on Text Similarity

James PustejovskyBrandeis University

March

13,

2012Slide2

Measuring SimilarityObjectsEventsSlide3

Object similarity is a function of:Sortal

correlationTemporal proximitySpatial proximity

the Latin Quarter of the 1920s

t

he 5

th

Arrondissement in 1929

Paris i

n 1925

The Left Bank in the early 20

th

CenturySlide4

Event similarity is a function of:Predicative similarity

Participant correlationTemporal proximitySpatial proximity

cf. Kim (1993), Davidson (1980), Lewis (1986)Slide5

Event Similaritya. Mary visited John in Boston on Tuesday.

b. The woman/she saw her husband in Copley Square yesterday.Sim(P1,P2

):

visit

vs.

see

Sim

(Subj

1

,Subj

2): Mary vs. the womanSim(Obj1,Obj2): John vs. her husbandSim(Loc1,Loc2): Boston vs. Copley SquareSim(Time1,Time2): Tuesday vs. yesterday

Brandeis CS114-2012 PustejovskySlide6

Predicative SimilarityLexical resourcesLSAVector-based models

Brandeis CS114-2012 PustejovskySlide7

Argument AlignmentSemantic Role Labeling +Sortal Similarity

Brandeis CS114-2012 PustejovskySlide8

Temporal SimilarityNormalizationMap to standardized ISO-TimeML formatReferencingReference relative to local temporal values

Val(Tuesday) = Val(yesterday)Brandeis CS114-2012 PustejovskySlide9

Spatial SimilarityNormalizationMap to standardized ISO-Space formatReferencingReference relative to accessible spatial values

Val(Copley_Sq) Spatial-IN Val(Boston)Brandeis CS114-2012 PustejovskySlide10

Temporal IssuesSubsumption in anchoringThe bombing occurred Monday morning.The bombing occurred Monday.

The bombing occurred last week. Brandeis CS114-2012 PustejovskySlide11

Motivation for time and event markupNatural language is filled with references to past and future events, as well as planned activities and goals; Without a robust ability to identify and temporally situate events of interest from language, the real importance of the information can be missed;

A Robust Annotation standard can help leverage this information from natural language text.Slide12

Temporal Awareness in Real TextThe bridge collapsed during the storm but after

traffic was rerouted to the Bay Bridge. President Roosevelt died in April 1945 beforethe war

ended

.

(event happened)

he

dropped

the bomb.

(event didn’t happen)

The CEO plans to retire next month. Last week Bill was running the marathon when he twisted his ankle. Someone had tripped him. He fell and didn't finish the race.Slide13

Current Time Analysis TechnologyDocument Time LinkingFind the document creation time and link that to all events in the text;Local Time Stamping

find an event and a “local temporal expression”, and link it to that time;Slide14

Document Time Stamping April 25, 2010President Obama paid tribute Sunday to 29 workers killed

in an explosion at a West Virginia coal mine earlier this month, saying they died "in pursuit of the American dream." The blast at the Upper Big Branch Mine was the worst U.S. mine disaster in nearly 40 years.Obama ordered

a

review

earlier this month and

blamed

mine officials for lax

regulation

.Slide15

Document Time Stamping: April 25, 2010President Obama paid tribute Sunday to 29 workers

killed in an explosion at a West Virginia coal mine earlier this month, saying they died "in pursuit of the American dream." The blast at the Upper Big Branch Mine was the worst U.S. mine disaster in nearly 40

years.Obama

ordered

a

review

earlier this month and

blamed

mine officials for lax

regulation.Slide16

Identify which Events Should be OrderedThe annotation specification should specify a kernel

of events and time expressions to be annotated. Anchoring relations between events and times depend on genre,

style

, and

register

.

Ordering relations between events depend largely on

discourse relations

in the text. Slide17

Creation vs. Narrative TimeDocument Creation Timewhen the utterance is made (speech time)Narrative Timewhen the event occursSlide18

Genre, Style, and RegisterParticipantsRelations among participants

ChannelProduction CircumstancesSettingCommunicative PurposeTopicSlide19

Genre, Register, and StyleHelp distinguish text types in order to better characterize the information structure of the text Example, news wire vs. news articlenarrative time (NT) is a function of publication/creation frequency. Slide20

Narrative TimeIdentifies the temporal interval of the events being described in the text. Document Narrative Time: set by text-genreCurrent Narrative Time: shifts through the textSlide21

Document Time Stamping: for real April 25, 2010President Obama paid tribute

Sunday to 29 workers killed in an explosion at a West Virginia coal mine earlier this month, saying

they

died

"in pursuit of the American dream." The 

blast

at the Upper Big Branch Mine was the worst U.S. mine disaster in nearly 40

years.Obama

ordered

a review earlier this month and blamed mine officials for lax regulation.Slide22

Narrative Container April 25, 2010President Obama paid tribute

Sunday to 29 workers killed in an explosion at a West Virginia coal mine earlier this month,

saying

they

died

"in pursuit of the American dream." The 

blast

at the Upper Big Branch Mine was the worst U.S. mine

disaster

in nearly 40 years

. Obama ordered a review earlier this month and blamed mine officials for lax regulation.Slide23

Time Stamping: the good, bad, … ✓ ☺Set up a meeting

on Tuesday with EMC. ✓ ☺Franklin arrives tomorrow from London. ✗ ☹

Franklin

arrives

on

the afternoon

flight from London

tomorrow

.

✗

☹ ☹ Most people drive today while talking on the phone. Slide24

ISO-TimeML Enables Temporal ParsingA new generation of language analysis tools that are able to temporally organize events in terms of their ordering and time of occurrence

These tools can be integrated with visualization, summarization, question answering, and link analysis systems to help analyze large event-rich information spaces.Slide25

ISO-TimeML Provides elements to:Find all events and times in newswire textLink events to the document time and to local timesOrder event relative to other eventsEnsure consistency of the the

temporal relations Slide26

ISO-SpaceCapture the complex constructions of spatial language in textProvide an inventory of how spatial information is presented in natural languageISO-Space is not designed to provide a formalism that fully represents the complexity of spatial languageSlide27

Applications of ISO-SpaceBuilding a spatial map of objects relative to one another.Reconstructing spatial information associated with a sequence of events.Determining object location given a verbal description.

Translating viewer-centric verbal descriptions into other relative descriptions or absolute coordinate descriptions.Constructing a route given a route description.Constructing a spatial model of an interior or exterior space given a verbal description.Integrating spatial descriptions with information from other media.Slide28

Semantic Requirements for AnnotationFundamental distinction between the concepts of annotation and representationBased on

ISO CD 24612 Language resource management - Linguistic Annotation Framework (Ide and Romary, 2004)Distinguish between abstract syntax and concrete syntaxConcrete Syntax  XML encoding

Abstract Syntax  Conceptual inventory and a set of syntactic rules defining the combination of these elementsSlide29

Spatial ExpressionsConstructions that make explicit reference to the spatial attributes of an object or spatial relations between objectsFour grammatically defined classes:Spatial Prepositions and Particles: on, in, under, over, up, down, left ofVerbs of Position and Movement:

lean over, sit, run, swim, arriveSpatial Attributes: tall, long, wide, deepSpatial Nominals: area, room, center, corner, front, hallwaySlide30

Spatial RelationsTopological:In, inside, touching, outsideOrientational (with frame of reference):Behind, left of, in front ofTopo-metric:Near, close by

Topological-orientational:On, over, belowMetric:20 miles awaySlide31

Frames of Reference (Levinson

, 2003)

Absolute

Th

e lake is north of the city.

Relative

The book is to your left.

The tree is between the

Pru

and the Monitor.

IntrinsicThere’s a ball in front of the car.The tree is behind the bench.Slide32

Frames of reference

The tree to the left of the entranceThe steps in front of me/the entranceSlide33

ISO-Space 1.4Spatial Relations are split into 4 types:Topological (QSLink

)Relational (OrientLink)Movement (MoveLink)Measurement (MLINK, from TimeML)Spatial Relations are

identified

with role labels, include Figure and Ground

SPATIAL_NAMED-ENTITYSlide34

Conclusion: Measuring Semantic SimilarityNormalizing temporal and spatial expressionsDeveloping standardized specifications contribute towards corpora for training and evaluation for such normalization

Cases in point:ISO-TimeML (ISO adopted)ISO-Space (in development)