Constraint Based Hindi Parser

LTRC, IIIT Hyderabad

Introduction

 Broad coverage parser   Very crucial IL-IL MT systems, IE, co-reference resolution, etc.

Why Dependency ?

 Phrase Structures    Intrinsically presumes order Context Free Grammar (CFG) not well-suited for free-word order languages (Shieber, 1985) Particularly ill suited to Indian Languages  Dependency Structures    Gives flexibility Common structures With appropriate labels, closer to Semantics

Computational Paninian Grammar (CPG)

 Based on Panini’s Grammar (500 BC)  Inspired by Inflectionally rich language (Sanskrit)  A dependency based analysis

Computational Paninian Grammar (The Basic Framework)

 Treats a sentence as a set of modifier modified relations  Sentence has a primary modified or the root (which is generally a verb)  Gives us the framework to identify these relations    Relations between noun constituent and verb called ‘

karaka’ karakas

are syntactico-semantic in nature Syntactic cues help us in identifying the

karakas

karta

– karma karaka

 The boy opened the lock   k1 –

karta

k2 –

karma



karta, karma

usually correspond to agent, theme  But not always k1

boy open

k2

lock



karakas

are direct participants in the activity denoted by the verb

Basic karaka relations

     

karta –

agent/doer/force  Relation label – k1

karma –

object/patient  Relation label – k2

karana –

instrument  Relation label – k3

sampradaan –

 beneficiary Relation label – k4

apaadaan –

source  Relation label – k5

adhikarana –

location in place/time/other  Relation label – k7p/k7t/k7  For complete list of dependency relations: (Begum et al., 2008)

Basic karaka relations

raama phala khaataa hai

‘Ram eats fruit’

Basic karaka relations

raama chaaku se saiv kaatataa hai

‘Ram cuts the apple with knife’

Basic karaka relations

raama ne mohana ko pustaka dii

‘Ram gave a book to Mohan’

Why Paninian Labels

 Other choices for labels could be   Grammatical relations  Subject, Object, etc.

 Behavioral tests (Mohanan, 1994) Thematic roles  Agent, patient, etc.

 No concrete cues   Difficult to extract them automatically Karakas can be computationally exploited  Syntactically grounded, Semantically loaded  Gives a level of interface

Levels of Language Analysis

 Morphological analysis ( Morph Info.

)  Analysis in local context ( POS tagging )  Sentence analysis ( Chunking , Parsing )  Semantic analysis ( Word sense disambiguation, etc.

)  Discourse processing ( Anaphora resolution, Informational Structure, etc.

)

Example

 rAma ne mohana ko puswaka xI |

Example – Parsed Output

k1 rAma xI ‘give’ k4 mohana k2 puswaka ‘book’

Parser

 Two stage strategy  Appropriate constraints formed  Stage I (Intra-clausal relations)  Dependency relations marked  Relations such as k1, k2, k3, etc. for each verb  Stage II (Inter-clausal relations & conjunct relations)  Conjuncts, relative clauses, kriya mula, etc

Demand Frame for Verb

 A demand frame or karaka frame for a verb indicates the demands the verb makes  It depends on the verb and its tense, aspect and modality (TAM) label.

 A mapping is specified between karaka relations and vibhaktis (post-positions, suffix).

Karaka Frame

 It specifies what karakas are mandatory or optional for the verb and what vibhaktis (post positions) they take respectively  Each verb belongs to a specific verb class  Each class has a basic karaka frame  Each TAM specifies a transformation rule

Example

 rAma mohana ko puswaka xewA hE |

xewA hE ‘give is’ k1 k2 k4 rAma mohana

Parsed Dependency Tree

puswaka ‘book’

Transformations

 Based on the TAM of the verb 

rAma ne mohana ko KilOnA xi yA

| 

rAma ko mohana ko KilOnA xe nA padZA

|  Appropriate transformation applied

Example

 rAma ne mohana ko puswaka xI |

Karaka Frame – xe (give)

Transformation Rule – yA (TAM)

Karaka Frame

rAma ne mohana ko KilOnA xi yA | yA TAM

Transformed frame for

xe

after applying the

yA

trasformation --------------------------------------------------------------------------------------- arc-label necessity vibhakti lextype src-pos arc-dir --------------------------------------------------------------------------------------- k1 m ne n l c k2 m 0|ko n l c k3 d se n l c k4 d ko n l c ----------------------------------------------------------------------------------------

0



ne

Parsed Output

k1 rAma xI ‘give’ k4 mohana k2 puswaka ‘book’

Other frames

 Adjectives

Steps in Parsing

SENTENCE Morph, POS tagging, Chunking Identify Demand Groups Load Frames & Transform Find Candidates Apply Constraints & Solve Final Parse

Example:

 rAma ne mohana ko KilOnA xiyA |

Identify the demand group, Load and Transform DF

 xiyA  Only verb  Transformed frame  Use ‘yA’ TAM info.

--------------------------------------------------------------------------------------- arc-label necessity vibhakti lextype src-pos arc-dir --------------------------------------------------------------------------------------- k1 m ne n l c k2 m 0|ko n l c k3 d se n l c k4 d ko n l c ----------------------------------------------------------------------------------------

Candidates

k1

main  rAma ne mohana ko KilOnA

k2 k2

xiyA _ROOT_ |

k4

Constraints

 C1: demand frame for each demand group, there should be For each of the mandatory demands exactly one outgoing edge in a labeled by the demand from the demand group.

 C2: For each of the optional demands in a demand frame for each demand group, there should be at most one outgoing edge from the demand group.

labeled by the demand  C3: There should be each source group .

exactly one incoming arc into

Constraints

 A parse of a sentence is obtained by satisfying all the above constraints  Ambiguous sentences have multiple parses  Ill formed sentences have no parse.

Parse - I

k1

main  rAma ne mohana ko KilOnA xiyA _ROOT_ |

k2 k4

Parse - I

k1 rAma _ROOT_ main xiyA k2 k4 mohana KilOnA

Integer Programming Constraints



X ijk

represents a possible arc from word group

i

to

j

with karaka label

k

 It takes a value 1 and 0 if the solution has that arc otherwise. It cannot take any other values.

 The constraint rules are formulated into constraint equations.

Constraint Equations

C1: For each demand group i, for each of its mandatory demands k, the following equalities must hold: M

ik

: S

j x ikj

= 1 C2: For each demand group i, for each of its optional or desirable demands k, the following inequalities must hold: O

ik

: S

j x ikj

< = 1 C3: For each of the source groups j, the following equalities must hold: S

j

: S

ik x ikj

= 1

Multiple Frames

 If more than one karaka frame for a verb  Call Integer Programming package for each frame  If more than one demand groups (e.g., multiple verbs) in the sentence with multiple demand frames  Call Integer Programming package for each combination of such frames

Other frames

 Common karaka frame  Attached to each karaka frame  Preference given to main frame if there are clashes  Fallback karaka frame  required karaka frame is missing  Graceful degradation

Stage I: Types being handled

 Simple Verbs  Non-finite verbs  wA_huA  wA_hI  nA  kara  0_rahe, etc.

 Copula  Genitive

Example (Complex Sentence)

 rAma ne phala khaakara mohana ko

Ram ‘ERG’ fruit ‘having eaten’ Mohan ‘DAT’

KilOnA xiyA

toy gave

‘Having eaten the fruit Ram gave the toy to Mohan’

Candidates

X1: k1 X4: k2

X8: main

X7: vmod

 rAma ne phala khaakara mohana ko KilOnA xiyA _ROOT_ |

X6: k2 X3: k2 X2: k2 X5: k4

Constraint Equations

 Verb ‘xe’   Mandatory Demands (C1)   k1  k2  x1 = 1 x2 + x3 + x4 = 1 Optional Demands (C2)  k4  x5 <= 1  Verb ‘khaa’  Mandatory Demands (C1)  k2  x6 = 1  vmod  x7 = 1  _ROOT_  C1  Main  x8 = 1

Constraint Equations (contd.)

 Incoming Arcs into Source (C3)    rAma  x1 = 1 phala  x4 + x6 = 1 khaa  x7 = 1    mohana  x3 + x5 = 1 KilOnA  x2 = 1 xe  x8 = 1

Solution Graph

_ROOT_

main

xiyA rAma

k1 k2 vmod k4

khaakara

k2

mohana phala KilOnA

References

 Akshar Bharati and Rajeev Sangal. 1993. Parsing free word order languages in Paninian Framework.

USA.

ACL:93, Proc.of Annual Meeting of Association of Computational Linguistics, Association of Computational Linguistics, New Jersey.

 Akshar Bharati, Rajeev Sangal, T Papi Reddy. 2002. A Constraint Based Parser Using Integer Programming

In Proc. of ICON-2002: International Conference on Natural Language Processing.

 Rafiya Begum, Samar Husain, Arun Dhwaj, Dipti Misra Sharma, Lakshmi Bai and Rajeev Sangal. 2008. Dependency Annotation Scheme for Indian Languages.

In Proceedings of The Third International Joint Conference on Natural Language Processing (IJCNLP).

Hyderabad, India.

 S. M. Shieber. 1985. Evidence against the context-freeness of natural language. In

Linguistics and Philosophy

, p. 8, 334 –343.

 Tara Mohanan, 1994.

Arguments in Hindi

. CSLI Publications.

THANKS!!