Controlled Natural Language for User Interfaces on the Semantic Web

Controlled Natural Language forUser Interfaces on the Semantic Web

Tobias Kuhn

http://www.tkuhn.ch

@txkuhn

ETH Zurich

Protege Research MeetingStanford Center for Biomedical Informatics Research

20 March 2015

http://www.tkuhn.ch

Imagine ...

... that Wikipedia can check consistency and answerquestions about the contained knowledge, and

... that all content is instantly available in alllanguages!

Tobias Kuhn, ETH Zurich Controlled Natural Language for User Interfaces on the Semantic Web 2 / 24

• AceWiki is a semantic wiki

• Articles are written in Attempto Controlled English (ACE), aControlled Natural Language (CNL)

• These sentences are internally translated into the Semantic Weblanguage OWL

• An OWL reasoner is built in to answer questions and detectinconsistencies

• Special editor for writing ACE statements

• Extended to support multilinguality

Kuhn. AceWiki: A Natural and Expressive Semantic Wiki. Semantic Web User Interaction. 2009.


Attempto Controlled English (ACE):a Natural and Intuitive Semantic Web Language

OWL (∼ Manchester syntax) Attempto Controlled English (ACE)

Bob HasType developer Bob is a developer.developer SubTypeOf professional Every developer is a professional.developer SubTypeOf owns some cup Every developer owns a cupBob HasType owns some (not cup) Bob owns something that is not a cup.loves SubRelationOf likes If X loves Y then X likes Y.

Tools:

• AceWiki (semantic wiki)

• ACE View (Protege plugin)

• ACE-in-GF (multilingual grammar for ACE)

• and more...

Fuchs, Kaljurand, Kuhn. Attempto Controlled English for Knowledge Representation. Reasoning Web. 2008.


Attempto Controlled English (ACE)

Subset of natural English:

• Disjunction, negation, if-then, nested expressions, ...

• Anaphoric references: pronouns, definite noun phrases, variables

• Quantifiers: every, no, at least 3, ...

• Content words: proper names, nouns, verbs, adjectives, ...

Grammar is fixed, but users can change content words.

Deterministic ambiguity handling:

• Anaphora resolution (France borders Spain and it bordersPortugal.)

• Quantifier scope (Every country includes a city.)

• Attachment (Every therapy treats a disease with a drug.)

Well-defined translations to and from first-order logic, OWL, ...

Fuchs, Kaljurand, Kuhn. Attempto Controlled English for Knowledge Representation. Reasoning Web. 2008.


AceWiki: A Wiki that Understands its Content

Kuhn. AceWiki: A Natural and Expressive Semantic Wiki. Semantic Web User Interaction. 2009.


Consistency Checking and Question Answeringin AceWiki

Kuhn. How Controlled English Can Improve Semantic Wikis. Semantic Wikis, 2009.


How can we reliably evaluate theunderstandability of CNLs?

Kuhn. The Understandability of OWL Statements in Controlled English. Semantic Web 4(1). 2013.


How can we reliably evaluate theunderstandability of CNLs?

Kuhn. The Understandability of OWL Statements in Controlled English. Semantic Web 4(1). 2013.


CNLs are Easier and Faster to Understand

objective understandability (score on task):

91.4%ACE

86.3%OWL (Manchester)

50% 60% 70% 80% 90% 100%

90%ACE

61%SOUL

71%Prolog

50% 60% 70% 80% 90% 100%

subjective understandability (questionnaire):

2.59ACE

1.92OWL (Manchester)

0 1 2 3

2.65ACE

1.30SOUL

1.50Prolog

0 1 2 3

needed effort (time in seconds):

13.72ACE

18.42OWL (Manchester)

0 5 10 15 20

548ACE

712SOUL

719Prolog

0 200 400 600 800

Kuhn. The Understandability of OWL Statements in Controlled English. Semantic Web 4(1). 2013. / Kuhn, Bergel.Verifiable source code documentation in controlled natural language. Science of Computer Programming 96. 2014.


Writing in CNL: Predictive Editor

Kuhn. A Principled Approach to Grammars for Controlled Natural Languages and Predictive Editors. Journal of Logic,Language and Information 22(1). 2013.


Determining Possible Anaphoric References ina Predictive Editor

s ∼

vp

vp ∼

np

ref

...

v

tv

destroy

aux

does not

conj

and

vp

pp

np

>n

enemy

det

every

prep

from

np

n

house

det

a

v

tv

protects

np

n

man

det

Every

∼

�

�

�

( ( ()

>

>

<

1Kuhn. A Principled Approach to Grammars for Controlled Natural Languages and Predictive Editors. Journal of Logic,

Language and Information, 22(1), 2013.


CNLs are Easier and Faster to Write

objective usability (score):

64%CNL

33%non-CNL

0% 20% 40% 60% 80% 100%

subjective usability (questionnaire):

2.33CNL

1.42non-CNL

0 1 2 3

needed effort (time in seconds):

86CNL

108non-CNL

0 30 60 90 120

Kuhn, Hofler. Coral: Corpus Access in Controlled Language. Corpora 7(2). 2012.


Multilingual AceWiki: AceWiki-GF

• Making wiki content available in different languages• Automatically translated content using high-quality rule-based

machine translation: Grammatical Framework (GF)• Language switching like in Wikipedia• Localization of the user interface

Kaljurand, Kuhn. A Multilingual Semantic Wiki Based on Attempto Controlled English and Grammatical Framework.ESWC 2013. / Kaljurand, Kuhn, Canedo. Collaborative multilingual knowledge management based on controllednatural language. Semantic Web. 2015.


Grammatical Framework (GF)

GF is a framework for multilingual grammar engineering:

• Rule-based

• Functional programming language (based on Haskell) optimizedto handle natural language

• Modular architecture for problem-specific languages

• Resource Grammar Library implementing common morphologicaland syntactic structures

• Mildly context sensitive

• Bidirectional translations: concrete languages ⇔ abstract syntax

http://www.grammaticalframework.org/



Multilingual AceWiki


Evaluation of AceWiki-GF:(Almost) Nothing is Lost in Translation

• 30 participants created in total 316 sentences in differentlanguages and tagged them as true or false.

• They then had to classify sentences others had written as true orfalse.

Results:

82.2%without translation

84.0%with translation

0% 25% 50% 75% 100%

agreement level

Kaljurand, Kuhn. A Multilingual Semantic Wiki Based on Attempto Controlled English and Grammatical Framework.ESWC 2013. / Kaljurand, Kuhn, Canedo. Collaborative multilingual knowledge management based on controllednatural language. Semantic Web. 2015.


There are Many Other Kinds ofControlled Natural Language (CNL)

CNLs have been used for: knowledge representation, query interfaces,mathematical texts, technical specifications, technical documentation,international communication, government documents, business/policyrules, personal scripts, emergency instructions, ...

• Goal: comprehensibility / translation / formal representation

• Written / spoken

• Domain-specific / general-purpose

• Origin: academia / industry / government

Kuhn. A Survey and Classification of Controlled Natural Languages. Computational Linguistics 40(1). 2014.


CNLs for the Semantic Web

Languages:

• Ginseng’s Guided English

• AIDA

• ClearTalk

• Controlled Language forOntology Editing(CLOnE)

• Rabbit

• OWL ACE (used in ACEView)

• OWL Simplified English

• and several others



AECMA Simplified English AIDA Airbus Warning Language ALCOGRAM ASD Simplified TechnicalEnglish Atomate Language Attempto Controlled English Avaya Controlled English Basic EnglishBioQuery-CNL Boeing Technical English Bull Global English CAA Phraseology Caterpillar Fun-damental English Caterpillar Technical English Clear And Simple English ClearTalk CLEF QueryLanguage COGRAM Common Logic Controlled English Computer Processable English ComputerProcessable Language Controlled Automotive Service Language Controlled English at Clark Con-trolled English at Douglas Controlled English at IBM Controlled English at Rockwell ControlledEnglish to Logic Translation Controlled Language for Crisis Management Controlled Language forInference Purposes Controlled Language for Ontology Editing Controlled Language Optimized forUniform Translation Controlled Language of Mathematics Coral’s Controlled English Diebold Con-trolled English DL-English Drafter Language E-Prime E2V IBM’s EasyEnglish Wycliffe Associates’EasyEnglish Ericsson English FAA Air Traffic Control Phraseology First Order English Formalized-English ForTheL Gellish English General Motors Global English Gherkin GINO’s Guided English Gin-seng’s Guided English Hyster Easy Language Program ICAO Phraseology ICONOCLAST LanguageiHelp Controlled English iLastic Controlled English International Language of Service and Mainte-nance ITA Controlled English KANT Controlled English Kodak International Service Language LiteNatural Language Massachusetts Legislative Drafting Language MILE Query Language Multina-tional Customized English Nortel Standard English Naproche CNL NCR Fundamental English OceControlled English OWL ACE OWLPath’s Guided English OWL Simplified English PathOnt CNLPENG PENG-D PENG Light Perkins Approved Clear English PERMIS Controlled Natural LanguagePILLS Language Plain Language PoliceSpeak PROSPER Controlled English Pseudo Natural Lan-guage Quelo Controlled English Rabbit Restricted English for Constructing Ontologies RestrictedNatural Language Statements RuleSpeak SBVR Structured English SEASPEAK SMART ControlledEnglish SMART Plain English Sowa’s syllogisms Special English SQUALL Standard Language SunProof Sydney OWL Syntax Template Based Natural Language Specification ucsCNL Voice Actions



Survey of 100 CNLs for English

1930 1940 1950 1960 1970 1980 1990 2000 2010

FAA Air Traffic Control Phraseology →Basic English

Caterpillar Fundamental English (CFE) →

ASD Simplified Technical English (ASD-STE) →

SBVR Structured English →AIDA →

Drafter Language →

Attempto Controlled English (ACE) →

E2V →

Sowa’s Syllogisms →

1

Kuhn. A Survey and Classification of Controlled Natural Languages. Computational Linguistics 40(1). 2014.Tobias Kuhn, ETH Zurich Controlled Natural Language for User Interfaces on the Semantic Web 21 / 24

Classification Scheme for CNLs: PENS

Four dimensions (PENS):

• Precision: from very imprecise (e.g. English) to maximallyprecise (e.g. propositional logic)

• Expressiveness: from very inexpressive (e.g. propositional logic)to maximally expressive (e.g. English)

• Naturalness: from very unnatural (e.g. propositional logic) tofully natural (e.g. English)

• Simplicity: from extremely complex (e.g. English) to very simple(e.g. propositional logic)



CNLs: Between Natural and Formal Languages

PENS classes of CNLs (blue) in comparison to natural (white) andformal (black) languages:

Precision1 2 3 4 5

Expressiveness

1

2

3

4

5

1

Naturalness1 2 3 4 5

Sim

plicity

1

2

3

4

5

1Kuhn. A Survey and Classification of Controlled Natural Languages. Computational Linguistics 40(1). 2014.


Thank you for your attention!

Questions?

Some links:

• AceWiki: http://attempto.ifi.uzh.ch/acewiki/

• AceWiki-GF: http://attempto.ifi.uzh.ch/acewiki-gf/

• Attempto project: http://attempto.ifi.uzh.ch/

• CNL community: http://www.sigcnl.org/

• Grammatical Framework: http://www.grammaticalframework.org/


http://attempto.ifi.uzh.ch/acewiki/

http://attempto.ifi.uzh.ch/acewiki-gf/

http://attempto.ifi.uzh.ch/

http://www.sigcnl.org/


Controlled Natural Language for User Interfaces on the Semantic Web

Science

semantic web user interaction

attempto controlled

semantic wikis

expressive semantic

semantic wiki articles

semantic weblanguage

eth zurich

reasoning web