Proiect MappingBooks

8/13/2019 Proiect MappingBooks

1/60

Cursuri 5-6

Proiectul:

MappingBooksLet me jump in the book!


2/60

Linking Book Characters

Building A Corpus Encoding

Relations Between Entities

Dan Cristea1,2, Eugen Ignat1

1 Alexandru Ioan Cuza University of IaiFaculty of Computer Science

2 RomanianAcademy, the Iai branchInstitute for Computer Science

{dcristea, eugen.ignat}@info.uaic.ro

SpeDCluj-Napoca, 15-17 October2013
http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/http://www.sped2013.ro/


3/60

I like to read books and to travel

SpeDCluj-Napoca, 15-17 October 2013


4/60

Going out of the book



5/60

I need help to

remember all

kinship

relationsbetween

characters



6/60

Characters in Forsyte Saga The old Forsytes

Ann, the eldest of the family

Old Jolyon, the patriarch of the family, having made a fortune in tea

James, a solicitor, married to Emily, a most tranquil womanSwithin, James's twin brother with aristocratic pretensions; a bachelor

Roger, "the original Forsyte"

Julia (Juley), a fluttery dowager; Mrs. Septimus Small

Hester, an old maid

Nicholas, the wealthiest in the family

Timothy, the most cautious man in England

Susan, the married sister

The young Forsytes

Young Jolyon, Old Jolyon's artistic and free-thinking son, married three times

Soames, James and Emily's son, an intense, unimaginative and possessive solicitor, married to the unhappy Irene, who later marries Young Jolyon

Winifred, Soames's sister, one of the three daughters of James and Emily, married to the foppish and lethargic Montague Dartie

George, Roger's son, a dyed-in-the-wool mocker

Francie, George's sister and Roger's daughter, emancipated from God

Their children

June, Young Jolyon's defiant daughter from his first marriage; engaged to an architect, Philip Bosinney, who becomes Irene's lover

Jolly, Young Jolyon's son from his second marriage; dies of enteric fever during the Boer Wars

Holly, Young Jolyon's daughter from his second marriage, to June's governessJon, Young Jolyon's son from his third marriage, to Irene, Soames's first wife

Fleur, Soames's daughter from his second marriage, to a French Soho shopgirl Annette; Jon's lover; later marries a baronet, Michael Mont

Val, Winifred and Montague's son; fights in the Boer Wars; marries his cousin Holly

Imogen, Winifred and Montague's daughter

Others

Parfitt, Old Jolyon's butler

Smither, Aunts Ann, Juley and Hester's housekeeper

Warmson, James and Emily's butler

Bilson, Soames's housemaid

Prosper Profond, Winifred's admirer and Annette's lover SpeDCluj-Napoca, 15-17 October 2013


7/60



8/60

This presentation

The entity linking problem

The MappingBooks project proposal

Design conventions of a corpus

Preliminary statistics and what next



9/60


10/60

The art

Classes studied in classification schemes:

Container-Contained, Time-Event, Product-

Producer; classes like Tool-Object are more

vaguely defined

SemEval-07 Task 4: 7 important nominal relations

Patterns: use context information for

classification and extraction of relationships

Supervised machine learning: match entity

mentions onto their correspondent KB records


11/60

Examples

Syntactic structures of the kind N1 N2

dog food=> foot consumed by dogs

summer morning=> a morning that happens in

the summer

Patterns of the type:

[Prefix] CW1 [Infix] CW2 [Postfix]

with [Prefix] [Infix] [Postfix]frequent words and

CW1, CW2content words



12/60

MappingBooksa project proposal

A MappedBook is a book connected with

locations/events in the virtual and real world

and sensitive to the instantaneous location (as

seized by the mobile/tablet) of a reader.

The information made available could possibly

be different depending on the moment and

the place of the reader.



13/60

MappingBooksa project proposal

multi-dimensional mash-ups combining textual,

geographical and temporal data

adequate presentation to the reader

links sensitive to:

the context of the mentions in the book,

the moment the user initiates an access

the current location of the user

make heavy use of entity linking techniques

spot the book mentions (persons and locations)

in the real and virtual world



14/60

MappingBooksan architecture



15/60

Our aims

1) connect entities mentions in the form ofnominals (noun phrases) => one coreferential

chain corresponds to each entity;

2) no preliminary records about linked entities =>the knowledge base evolves from scratch;

3) look for 4 types of relations => referential,

kinship, affective, social;4) texts under investigation: fiction => reference

area: well delimited, BUT form of expression:

totally unrestricted. SpeDCluj-Napoca, 15-17 October 2013


16/60

Entities

Types PERSON, LOCATION

Individuals (Marcus Vinicius, the emperor), groups (his

mother and father, the soldiers) and classes (emperor) Syntactic realisation: NPs (determiners, adjectives,

complement PPs included; but NO relative clauses)

Characterised by distinctive, person type, heads

If intersectedimbricated ([[the young lady]s mother])



17/60

Entities

Types PERSON, LOCATION

Only identification, not also characterisation descriptions

([the man with a straw hat], but only: [that man],corrupted to the marrow of his bones)

included entities (in Romanian in the position of subject

(dar l iubea din tot sufletul => but (3rdpers sing)-loved

him with the whole soul)



18/60

Relations

Anaphoric

Non-anaphoric:

kinship

affection

social



19/60

Anaphoric relations

coref member-of, has-as-member (inverse)

isa, class-of(inverse)

part-of, has-as-part(inverse)

subgroup-of, has-as-subgroup(inverse)

has-name, name-of (inverse)



20/60

Kinship relations

parent-of, child-of(inverse)

grandparent-of, grandchild-of(inverse)

sibling(symmetrical)

ant-uncle-of, nephew-of(inverse relations)

cousin-of(symmetrical)

spouse-of(symmetrical unknown



21/60

Affective relations

friend-of, enemy-of(inverse)

love,hate(inverse)

worship



22/60

Social relations

inferior-of, superior-of(inverse)

colleague-with



23/60

Poles and directionality

In case of referential:

from anaphor towards antecedent

In case of non-referential: from sourcetowards destination


24/60

How to identify poles

Imbricated:

the anaphor/source is larger than the

antecedent/destination

Non-imbricated: referential:

the anaphoris to the right of the antecedent

non-referential: from sourceto destination, as the trigger reads


25/60

Poles and triggers

Referentiality: coref

John met Maria on the ski slope. He raced her.

anaforantecedent



26/60

Poles and triggers

Referentiality: coref

John met Maria on the ski slope. He raced her .

anaforantecedent



27/60

Poles and trigger

Kinship: parent-of

their father

trigger

destination

source



28/60

Poles and trigger

Social: inferior-of

Cesar s principal courtiers

trigger

destination source



29/60

Poles and trigger

Affective: worship

Lygia dropped on her knees to implore someone else .

trigger

destinationsource



30/60

EntitiesPetroniu

Vinicius was

the son of his oldest sister ,

who years before had married his father ,

a man of consular dignity from the time of Tiberius .



31/60

Referential relations: coref

Petroniu

Vinicius was






32/60

Petroniu

Vinicius was




Referential relations: coref



33/60


34/60

Referential relations: class-of

Petroniu

Vinicius was






35/60

Kinship relations: sibling

Petroniu

Vinicius was






36/60


37/60

Kinship relations: parent-of

Petroniu

Vinicius was






38/60

Kinship relations: spouse-of

Petroniu

Vinicius was






39/60

Social relations: inferior-of

Petroniu

Vinicius was



a man of consulardignity from the time of Tiberius .



40/60

The Quo Vadis RO-EntLink corpus

words ent ref aff kin soc

Now 25314 3663 2045 39 29 15Whole 137510 19483 9523 301 127 103

annotators: 12 master students, first year in CL

3 experts



41/60


42/60

We do not mark

Negated relations

1:[Lygia]could not become 2:[the concubine of any man]

Characterisations addressing subjects, expressed by

predicative nouns

in 1:[her]there was 2:[something uncommon]

Interpreted relations, besides coreferential

Petronius felt that beneath a statue of 1:[that maiden]

one might write 2:["Spring."] (name-of-interpret)



43/60

We do not mark

More than the minimum number of relations Se repezi la1:[Petru]i, lundu-2:[i] 3:[minile], ncepu

s 4:[i]5:[le]srute.

[2] coref [1],

[3] part-of [1] (or [3] part-of [2]),

[4] coref [1] (or [4] coref [2]),

[5] coref [3]

[5] part-of [1] superfluous; by transitivity, from [5] coref

[3] and [3] part-of [1]



44/60

Our thanks go to

the class of first year master students in

Computational Linguistics, from the Faculty ofComputer Science, Alexandru Ioan Cuza

University of Iai, second term 2012-2013



45/60

And my thanks to

YOU!



46/60

And now

Lets build the Mapped Books!

F t t t h


47/60

Features we want to have

Understand what a text says about

Know who is who

Recognise real world entities

Know where I am

What real world entities are in my proximity

Trace on Google Maps a path described in the book

Fetch, process and make use of geo-data

Mix images with generated info

Attractive user interfaces

Client-server

F t t t h


48/60


Understand what a text says about

the capacity to see a text different than a string ofletters

TEXT ANALYTICS


49/60

F t t t h


50/60


What real world entities are mentioned in the book

virtual world vs real world

link textual mentions of entities in the virtual world

decide on relevant info from virtual to be presented to

Master use multiple sources

ENTITY CROWLING

F t t t h


51/60


Know where I am

What real world entities are in my proximity

detection of my position

computation of distances from the mentioned places

signalling interesting locations in proximity

LOCALISATION



52/60


Trace on Google Maps a path described in the book

path detection in text

use Google Maps APIs

trace locations and paths on Google Maps

PATH DETECTION

MAPS&TRAJECTORIES


53/60



54/60


Mix images with generated info

process images => segment, contours, recognition

sense orientation of the camera

decide info to be displayed

use info provided by the camera focus

AUGMENTED REALITY


55/60



56/60


Client-server

users Portrait

the database

the architecture

standards and communication protocols

CLIENT-SERVER


57/60


58/60

Modules

1. TEXT ANALYTICS

2. NAME ENTITY

RECOGNITION3. ENTITY CROWLING

4. LOCALISATION

5. PATH DETECTION

6. MAPS&TRAJECTORIES

7. GEOGRAPHY

8. AUGMENTED REALITY

9. INTERFACES

10. CLIENT-SERVER

11. RESOURCES

12. MANAGEMENT AND

EVALUATION


59/60

Organisational details

2 projects!

Two groups: lead by Ionu Pistol and Mdlina Rschip

Each project includes a number of modules (12?)

Each module is built by a team

Each team has aprox. 10-15 members

Only one Management & Organisation team


60/60

Immediate deadline

Teams organised next week!

Proiect MappingBooks

Documents

Transcript of Proiect MappingBooks