Owlizr

Knowledge Representation System for Bahasa Indonesia Based on Knowledge Representation System for Bahasa Indonesia Based on

Web Ontology Language Description Logic (OWL DL)Web Ontology Language Description Logic (OWL DL)

Fariz DarariAdila Alfa Krisnadhi

Hisar Maruli Manurung

Faculty of Computer ScienceUniversitas Indonesia

ICACSIS 2010Download paper: ir.cs.ui.ac.id/publication/2010/owlizr.pdf

Outline

Background Knowledge Representation Architecture Evaluation Results Conclusions

Challenge

? KNOWLEDGETEXT

Why?

Automated Knowledge Representation & Reasoning!!

We always love automated things:• Automated Door• Automated Teller Machine• Automated Motorbike

Goal

KNOWLEDGETEXT

Background

Previous Works

Larasati's Model a syntactic and semantic processing system for QA in Bahasa

Mahendra's Extend the system with the axiom addition, e.g., NLP axiom & world knowledge axiom

NLP and Event Representation

Reification

Turning non-object thing into object

Using Neo-davidsonian approach, with thematic roles (agent, patient, theme, time, location)

Description Logic

Consists of two components:

Previous research by Franconi with KODIAK, representing knowledge from natural language using Description Logic

OWLizrKnowledge Representation

Event Modelling

EventEvent

PatientPatient

AgentAgent

ActionAction

hasAgent

hasAction

hasPatient

Event Modelling (cont)

event_1event_1

carcar

budibudi

buy_actionbuy_action

hasAgent

hasAction

hasPatient

“Budi buys a car” or “Budi membeli mobil”

Background Knowledge

DL Model

Class: Thing, PhysicalObject, AbstractObject, LivingPhysicalObject, NonLivingPhysicalObject

Class Definition

DL Model (cont)

OWLizrArchitecture

Architecture

Two Modes

Knowledge-assertion Mode NLP Semantic Analyzer → KB Generator → KB Reasoner

Query Mode NLP Semantic Analyzer → SPARQL Query Generator

NLP Semantic Analyzer

Reusing Mahendra's program

Analysis using syntax-driven semantic analysis with lambda calculus

Divided into 4 parts (Lexicon, Grammar, Lexical Semantics, Semantic Attachment Rules)

Example:

“Pabrik memproduksi mobil” or “The factory produces the car” becomes

[location(x5,pabrik), event(x1,memproduksi), agent(x1,x5), patient(x1,x6), objectx(x6,mobil)]

KB Generator

Transforms semantic notations into OWL

Implemented in Java with library Protege-OWL API

Two main functions, instance and property assertions

Example

Instance Assertion:from[location(x5,pabrik), event(x1,memproduksi), agent(x1,x5), patient(x1,x6), objectx(x6,mobil)]toFactory(factory_1), Event(event_1), Car(car_1)

Property Assertion:

Instance assertion

Property Assertion

KB Reasoner

Two Main Uses: Consistency Checking Check if there is a contradiction, for example “Mobil membeli radio” or “The car buys the radio” will produce an error. But, of course people can buy the radio.

Instance Checking Function to obtain inferred knowledge. It will check whether an instance could be classified in some classes or not.

SPARQL Query Generator

Translates semantic notations into SPARQL Query

Formed by two components, SELECT and WHERE clause

Then, execute query on KB

Example

Evaluation Results

Evaluation

Serves as a proof-of-concept Using a specific, domain ontology, that is economic activity Example terms are "price" or "harga", "expensive" or "mahal", "buy" or "membeli", "sell" or "menjual", "buyer" or "pembeli", and "shop" or "toko" The evaluation tests the ontology features, such as subclass, intersection, union. So, we define each terms in the ontology using various ontology features. For example, buyer is defined as:

Knowledge Assertion Mode

Input: “Anto buys the car in the shop” or “Anto membeli mobil di toko”

NLP Semantic Analyzer:

[person(x6,anto), event(x4,membeli), agent(x4,x6), patient(x4,x3), objectx(x3,mobil), di(x4,x1), location(x1,toko)]

Asserted Knowledge (Instance):

Assertion Mode (cont)

Asserted Knowledge (Property)

Assertion Mode (cont)

Inferred Knowledge

Query Mode

Input: “Who buys the car?” or “Siapa yang membeli mobil?”

NLP Semantic Analyzer:

[ans(x7), person(x5,x7), event(x4,membeli), agent(x4,x5), patient(x4,x2), objectx(x2,mobil)]

SPARQL Query Form:

ans(x7) → SELECT ?x7

person(x5,x7), event(x4,membeli), agent(x4,x5), patient(x4,x2), objectx(x2,mobil) → WHERE { ?event :hasAction :buy_action . ?event :hasAgent ?x7 . ?event :hasPatient ?ins . ?ins rdf:type :Car }

Conclusions

This research works as a bridge between NLP (Mahendra's) and DL (Franconi's)

The research value is in the formalization attempt to natural language text

The resulting knowledge can be shared and reused across the web

Next, implementing TBox assertion and increasing OWL version to OWL 2