Controlled Natural Languages for Knowledge Capture

Introduction
Controlled Natural Languages
Conclusion

Controlled Natural Languages for Knowledge
Capture
Intelligence Augmentation Forum

R. Denaux

School of Computing
University of Leeds

Leeds, 7th of June 2010

Denaux Controlled Natural Languages

Introduction
Conclusion

Outline

1 Introduction
Context
Problem

2 Controlled Natural Languages
Deﬁnition
Design Issues
Implementation
Demo

3 Conclusion


Introduction
Context
Problem
Conclusion

Outline

1 Introduction
Context
Problem

Deﬁnition
Design Issues
Implementation
Demo

3 Conclusion


Introduction
Context
Problem
Conclusion

Knowledge Capture
What are we capturing?


Introduction
Context
Problem
Conclusion

Knowledge Capture
What are we capturing in OWL terminology?


Introduction
Context
Problem
Conclusion

Why do we need it?
Intelligence Augmentation


Introduction
Context
Problem
Conclusion

Formalisation of Knowledge

People have the real knowledge (sometimes encoded in
documents, wikis, databases, etc.)
Knowledge formalisation is not trivial and needs understanding
of logical formalism: 1st order logic, OWL, etc.


Introduction
Context
Problem
Conclusion

Wikis

Conceptual Knowledge: Yes
Factual Knowledge: Yes
However:
Not formal
Knowledge extraction results are limited, despite
semi-structured format


Introduction
Context
Problem
Conclusion

Semantic Wikis

Conceptual Knowledge: Yes, but cumbersome


Introduction
Context
Problem
Conclusion

Ontology Engineering

However:
Requires knowledge elicitation phase
Results are difﬁcult to understand by domain experts


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Outline

1 Introduction
Context
Problem

Deﬁnition
Design Issues
Implementation
Demo

3 Conclusion


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

What is a Controlled Natural Language(CNL)?
Unfortunately...


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo


Constraints
Every CNL is a kind of
Engineerd Language


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo


Constraints
Engineerd Language

Optional Constraints
A CNL usually can be
translated into a Logical
Formalism


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo


Constraints
Engineerd Language

A CNL Sentence usually has
an Unambiguous Logical
Translation


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo


Constraints
Engineerd Language

A CNL Sentence usually is an
Easy-to-Learn Language


Definition
Introduction
Design Issues
Implementation
Conclusion
Demo

Broad subdivision of CNLs
Simplified NL vs Formally Underpinned CNLs

Every Simplified NL is a kind of CNL.
cannot be directly translated into a Logical Formalism
is used to increase quality of documents
easier to understand by humans (eg non-natives)
easier to process by machines
defined by a set of restrictions on the language:
limited set of words to use
disallowed constructs (eg: passive voice)
eg: Simple English is a Simplified NL. Caterpillar Technical
English is a Simplified NL.


Definition
Introduction
Design Issues
Implementation
Conclusion
Demo

Broad subdivision of CNLs
Simplified NL vs Formally Underpinned CNLs

Every Formally Underpinned CNL is a kind of CNL.
has a semantic mapping into a Logical Formalism
is used to formalise knowledge
easier (than logic formalism) to understand by humans
directly processable by machines
defined by a formal grammar
eg: ACE, PENG, CPL(By BOEING), Common Logic Controlled
English and Rabbit are all Formally Underpinned CNLs.


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Balancing Expressivity and Parseability

Limiting statements per sentence
NL: To describe in an unambiguous manner the inland
hydrology feature classes surveyed by Ordnance Survey
with the intention of improving the use of the surveyed data
by our customers and enabling semi-automatic processing
of these data.
CNL: Ontology describes the OS Inland Hydrology Feature
Classes.
CNL: Ontology aims to improve Data Usage Of OS Data.
CNL: Ontology aims to enable Semi-automatic Processing
of OS Data.
etc.


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Balancing Expressivity and Parseability

Anaphoric Reference
1 A pilot does not have a valid license.
2 It is expired.


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Capturing the formal semantics

Subjunction
An Actor is a Person. (Actor == Person?)
Every Actor is a kind of Person.


Definition
Introduction
Design Issues
Implementation
Conclusion
Demo


Relation vs Definition
Every River flows into a Sea
A River is anything that: is a kind of Body of Water; flows
into a Sea.


Definition
Introduction
Design Issues
Implementation
Conclusion
Demo


Property Domain
The relationship “flows into” must have the subject River.
Everything that “flows into” something is a River.


Definition
Introduction
Design Issues
Implementation
Conclusion
Demo

Limiting Ambiguity

Lists
Every River flows into a Sea or a Lake.
Every River flows into a Sea or flows into a Lake


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Limiting Ambiguity

Lists with cardinality restrictions
Every River ﬂows into exactly 1 Sea or Lake.


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Limiting Ambiguity

Concept or Relationship?
Every River ﬂows into a Sea.
Every River Flow has a direction.


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Discourse Representation Structure

First Order Logic
PENG ACE


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo

Lightweight NLP

CLoNE Rabbit


Deﬁnition
Introduction
Design Issues
Implementation
Conclusion
Demo


Introduction
Conclusion

Typical Usage of Novice User

See example sentences to get a feeling for the language
Write sentence
Get feedback and improve sentence until correctly parsed


Introduction
Conclusion

How to Evaluate a CNL

Use a made up world with made up concepts
Use a “Controlled Visual Language”


Introduction
Conclusion

Knowledge Capture with CNLs

However:
No guarantee that conceptual knowledge is directly usable
Is OWL(or 1st Order Logics) correctly understood?


Introduction
Conclusion

ROO: Rabbit to OWL Ontology Authoring.
Example of adapting to ontology contributors

Domain experts:
Good knowledge of the domain to be represented
Limited or no Ontology Engineering experience
Limited or no knowledge of OWL, Protégé, etc.
ROO provides tool support for domain experts:
Guidance through ontology construction methodology
Controlled Natural Language interface
No OWL speciﬁc terminology
Adaptation at design time, not at runtime
Re-use techniques from User Modelling and
Personalisation


Introduction
Conclusion

ROO: Rabbit to OWL Ontology Authoring.
Example of adapting to ontology contributors


Introduction
Conclusion

Acknowledgements

3rd Party Images used:
http://shlomit.deviantart.com/art/
Keanu-Reeves-portrait-118051274
http://owen-c.deviantart.com/art/
Paillard-Bolex-H16-48443994
http://eliskan.deviantart.com/art/
World-Map-127501551
http://so-aesthetic.deviantart.com/art/
Drama-Masks-49646555


Controlled Natural Languages for Knowledge Capture

More Related Content

Recently uploaded (20)

Featured (20)

Controlled Natural Languages for Knowledge Capture