A Semantic Web for Linguistics

1
A Semantic Web for Linguistics

• Scott Farrar
• farrar_at_uni-bremen.de
• D. Terence Langendoen
• langendt_at_u.arizona.edu

2
Acknowledgments

• We particularly thank Gary Simons, who cant be
  here because hes doing ISO work on language
  codes as we speak.
• Baden Hughes and Dafydd Gibbon, for putting the
  fear of god into us.
• Laura Buszard-Welcher for FIELD testing our
  nascent ontology.

3
Purposes of this talk

• To introduce the problem of data interoperability
  in the context of the EMELD project.
• To show how the semantics of markup may be
  derived using a metaschema and an ontology,
  resulting in a Semantic Web for linguistics.

4
The EMELD Project

• Electronic Metastructures for Endangered Language
  Data
• Five year grant from NSF
• Eastern Michigan, Wayne State, Arizona, LDC
  (Penn), Endangered Languages Fund, SIL
• A major objective
• The "formulation and promulgation of best
  practice in linguistic markup of texts and
  lexicon"

5
Problem Statement

• Three points of community consensus
• XML markup provides the best format for the
  interchange and archiving of language data.
• No single system of XML markup can be imposed on
  all language resources.
• Linguists need to be able to perform queries
  across data sets.
• The problem
• How do we interoperate when resources use
  different markup schemas?

6
Smart searches need smart data

• XML can be used to represent linguistic analyses
  to any desired degree of refinement.
• TEI feature system recommendations demonstrate
  this. Now being proposed as ISO standard.
• Analyses in other formats can be migrated to XML
  for both archiving, and smart web searching.

7
Smart markup isnt enough

• Meaning and use of structural markup varies from
  site to site.
• Same term used with different meanings.
• Different terms used with the same meaning.
• Markup element and attribute names and values,
  and structural content may be in different
  natural languages.
• Sites are encoded at different levels of
  granularity.

8
Say what you mean!

• Markup is syntax its meaning can only be
  inferred for individual sites, or groups of sites
  that use a common markup scheme (e.g. TEI).
• The element ltdeletedgt is deleted in the
  Wittgenstein archive, but cant be guaranteed
  across archives.
• The attribute Number plural is gt 1 in a
  hypothetical markup of SAE, but is gt 2 in such
  a markup of Kiowa.

9
Passing the baton
10
Paths to interoperability

• Possibility 1 simply impose standards on
  individual data sources.
• Possibility 2 allow freedom in the construction
  of data, but map to a shared semantic resource
  (Semantic Web idea).

11
Mapping to a shared resource

• Develop markup with the semantic resource in mind
  (develop the SemWeb directly).
• Allow for full freedom in the markup. Map to
  semantic resource later (migrating legacy data).
• metaschema

12
Defining the semantics of markup

• markup schema
• A formal definition (as with XML DTD or XML
  Schema) of the permitted vocabulary and syntax of
  markup for a class of source documents.
• semantic schema
• A formal definition (as with RDF Schema or OWL)
  of the concepts in a particular domain.
• metaschema
• A formal definition of how the elements and
  attributes of a markup schema are interpreted in
  terms of the concepts of a semantic schema.

13
A metaschema language

• lt!ELEMENT metaschema (interpret ignore) gt
• lt!ELEMENT interpret (resource literal
  property) gtlt!ATTLIST interpret markup
  CDATA REQUIREDgt
• lt!ELEMENT resource (literal property embed)gt
  
• lt!ATTLIST resource concept CDATA REQUIREDgt
• lt!ELEMENT literal (text-content) gt lt!ATTLIST
  literal concept CDATA REQUIREDgt
• lt!ELEMENT property (resource resourceRef
  embed)gt lt!ATTLIST property concept CDATA
  REQUIREDgt
• ...

14
Metaschema schematic
Metaschema
SourceDocument
SemanticInterpretations
DocumentInterpreter
SourceDocument
SourceDocument
15
Implementation

• The document interpreter has been implemented in
  XSLT as a two-stage process
• Input a metaschema documentStylesheet the
  metaschema compiler (XSLT)Output interpreter
  for that metaschema (XSLT)
• Input a source documentStylesheet interpreter
  for the metaschema (XSLT) Output the semantic
  interpretation (RDF/XML)

16
For example

• Source document
• ltmorph id"aba"gt lt!-- Content --gt lt/morphgt
• Metaschema directive
• ltinterpret markup"morph"gt ltresource
  concept"goldmorpheme"/gt lt/interpretgt
• Interpretation of document
• ltgoldmorpheme rdfabout"element(aba)"gt
  lt!-- Interpretation of content --gt
  lt/goldmorphemegt

17
Example 2

• Source document
• ltorth typevariantgtabbalt/orthgt
• Metaschema directive
• ltinterpret markuporth_at_typevariant"gt
  ltliteral concept"goldOrthoWord"/gt
  lt/interpretgt
• Interpretation of document
• ltgoldOrthoWordgtabbalt/goldOrthoWordgt

18
The full power

• The full XPath expression language is available
  to specify _at_markup.
• lttext-contentgt allows literal values to be
  composed (with optional before and after labels)
  from multiple markup sources.
• ltembedgt allows explicit control of embedding
• partition of source child elements into separate
  semantic substructures
• movement of source elements

19
Ontological support

• For a definition, refer to other presentations
• There can be many ontologies for a given domain.
• The same domain can be modeled by more than one
  ontology.
• Some domain may require the use of more than one
  ontology.

20
Some examples of upper ontologies

• Practical ontologies
• CYC (Cycorp)
• SUMO (Teknowledge)
• GUM (Bateman et al.)
• Almost practical ontologies
• DOLCE (Trento, Rome),
• BFO (Leipzig--IFOMIS)

21
GOLD

• General Ontology for Linguistic Description
• Based on SUMO
• Currently in OWL-DL
• http//emeld.org/gold
• Not a theory of language, but a metalanguage to
  talk about theory specific constructs.

22
Major Conceptual Categories

• started with morphosyntax (morphemes, words,
  meanings)

• linguistic expressions
• linguistic units
• features/values
• linguistic relations
• NL semantics

23
Partial SUMO Taxonomy
Entity
is-a
is-a
Physical
Abstract
24
Linguistic Expressions

• Orthographic units of language

Physical
Object
SelfConnectedObject
ContentBearingObject
WrittenExpression
25
Linguistic Units

• Abstract aspect of language

Abstract
LinguisticUnit
MorphosyntacticUnit
PhrasalUnit
LexicalUnit
SublexicalUnit
26
Semantic Units

• Cognitive containers for meanings of lexemes
  and bound forms (like synsets in WordNet)

SelfConnectedObject
Artifact
HuntingImplement
(function Arrow Hunting)
27
Linguistic Features

• Attributes which inhere to linguistic units (in
  the spirit of HPSG)

Abstract
Attribute
InternalAttribute
LinguisticFeature
MorphosyntacticFeature
28
GOLD Relations

• Formal means of relating linguistic entities to
  one another

BinaryPredicate
LinguisticRelation
SyntacticRelation
RhetoricalRelation
SemanticRelation
29
GOLD Relations Basic

• relations between two entities of the same type

part
Eingang
gang
constituent
/in the middle/
/the middle/
antonym
Happy
Sad
30
GOLD Relations Semiotic

• Special kinds of linguistic relation that cut
  across linguistic types

realizes
designates
dog
/dog/
Dog
31
Summary

• Achieving semantic interoperability among
  disparate data sets is a non-trivial problem, as
  shown by the EMELD project.
• Interoperability may be achieved using a
  metaschema and an ontology, resulting in a
  Semantic Web for linguistics.
• Any kind of semantic interoperability requires a
  shared semantic resourceGOLD.

32
Contact Info

• D. Terence Langendoen
• langendt_at_u.arizona.edu
• Scott Farrar
• farrar_at_uni-bremen.de