This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Dependency Relations for Sanskrit Parsing andTreebank
1Department of Sanskrit StudiesUniversity of Hyderabad
2School of Linguistics & Literary StudiesChinmaya Vishwavidyapeeth
Ernakulam
19th International Workshop on Treebanks and LinguisticTheories 2020
1 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Introduction
1 Pānini’s Grammar
2 Samsādhanī Tagset
3 Enhanced Tagset
4 Samsādhanı Parser
5 Sanskrit Treebank
6 Evaluation
7 Conclusion
2 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Pāṇini’s Grammar
Around 400 BCEFull-fledged grammar of the then prevalent SanskritCovering both Vedic and Non-Vedic textsOnly full-fledged, manually written, computational grammarfor any natural language
3 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Samsādhanī Tagset
The first tagset for Sanskrit (Ramakrishnamacharyulu 2009)Approximately 100 relationsInter-sentential relationsIntra-sentential relations
4 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Samsādhanī Tagset
Examined from computational view pointFound to be fine-grainedFine-grain relations - 100
5 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Samsādhanī Tagset
Kartāanubhavī(experiencer)amūrtaḥ(abstract)prayojakaḥ(causative agent)prayojyaḥ(causee)abhiprerakaḥ(cause of temptation)karmakartṛkaraṇakartṛṣaṣṭīkartā
6 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Samsādhanī Tagset
Skt: Rāmaḥ(Kartā) pacatiEng: Rama cooks
Skt: Ghaṭaḥ(anubhavī kartā) naśyatiEng: Pot is experiencing the destruction
Coarse-grain relations - 31
7 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Lacunae in the Samsādhanī Tagset
Relations without any associated semanticsRelations due to morphological requirementComplementiser
Inconsistent with Pāṇinian GrammarCoordinating conjuncts
8 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Relations due to morphological requirement
grāmaṃ paritaḥ vṛkṣāḥ santi.village surrounding trees are.
upapadasambandhaḥlocation
Kartā
Figure: Old annotation
grāmaṃ paritaḥ vṛkṣāḥ santi.village surrounding trees are.
reference_pointlocation
Kartā
Figure: New annotation
9 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Complementiser
ahaṃ gṛhaṃ gacchāmi iti rāmaḥ avadat.I home goes that Rama said
anuyogī
agentpratiyogī
agent
karma
Figure: Complementiser: old version
ahaṃ gṛhaṃ gacchāmi iti rāmaḥ avadat.I home goes that Rama said
agentvākyakarma
vākyakarmadyotakaḥ
agent
karma
Figure: Complementiser: new version
10 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Coordinating Conjuncts
Rāmaḥ Sītā ca vanaṃ gacchati.Rama Sita and forest go.
kartākarma
conjunctconjunct
Figure: Conjuncts: Old version
Rāmaḥ Sītā ca vanaṃ gacchati.Rama Sita and forest go.
kartā
karmaconjunct conjunction
Figure: Conjuncts: New version
11 / 23
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
...
.
Samsādhanī Enhanced Tagset
The current version has 54 relations and classified into followingcategories.
3 Implemented as an edge-centric binary join to build adependency tree, in bottom-up approach, with local andglobal constraints on the edges and the edge labels.