We can distinguish two families of approaches to the building of ontologies -- corresponding roughly to the contrast between 'neats' and 'scruffies' in artificial intelligence research. We describe the implications of each approach for the building of an ontology of philosophy, focusing especially on the Indiana Philosophy Ontology (InPhO) project led by Colin Allen.
A video presentation based on these slides is available here: https://www.youtube.com/watch?v=5HV3M0NvyPM
5. The problem these ontologies
were built to solve
You have a lot of data / literature
The data is described in heterogeneous ways
You need to access and reason with the data
in a uniform way
1. Create a controlled vocabulary of preferred
labels for describing the data
2. Provide logical (computable) definitions
3. Tag (‘semantically enhance’) the data with
ontology term URIs
15. answer: by tagging data with terms from a
controlled vocabulary such as the Gene Ontology
15
sphingolipid transporter activity
Holliday junction helicase complex
age-dependent behavioral decline
18. Figure 3.
Shotton D, Portwin K, Klyne G, Miles A (2009) Adventures in Semantic Publishing: Exemplar Semantic Enhancements of a
Research Article. PLoS Comput Biol 5(4): e1000361. doi:10.1371/journal.pcbi.1000361
http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1000361
… allows tagging of literature
RB Reis, GS Ribeiro, RDM Felzemburgh, et al., Impact of Environ-
ment and Social Gradient n Leptospira Infection in Urban Slums
27. Example: Emotional personality trait
An emotional personality trait
=def. a stable enduring characteristic of a
person
which involves a predisposition (i.e. a
disposition which gives rise to an increased risk)
to undergo emotions of a particular sort, both
occurrents and dispositions.
Saturday, April 4, 2015 27
28. all terms provided with definitions
Saturday, April 4, 2015
28
The Emotion Ontology
33. built by downward population from MF (which is in turn
built from BFO)
MFO-EM affective representation is_a
MFO:cognitive representation
MFO:cognitive representation is_a
BFO:specifically dependent continuant
33
37. Example: biochemical basis of emotion
Emotions are effected in part by neurotransmitters
such as dopamine, tryptophan
with thanks to Janna Hastings, European Bioinformatics Institute
Saturday, April 4, 2015 37
dopamine
(CHEBI:25375)
molecular entity
(CHEBI:25375)
biological role
(CHEBI:24432)
neurotransmitter
(CHEBI:25512)
has role
neurotransmitter
receptor activity
(GO:0030594)
Molecular function
(GO:0003674)
realized in
happiness
(MFOEM:42)
part of
emotion
(MFOEM:1)
subtype
38. Is-a overloading
Toronto is a city
capital city is a city
It is a disgrace to the human race that it has
chosen to employ the same word ‘is’ for these
two entirely different ideas (predication and
identity) – a disgrace which a symbolic logic
language of course remedies. (Russell 1919:172)
38/
39. Three kinds of Relations
39
Relations between types (or ‘classes’)
is_a (= is a subtype of)
Relations between instances (or ‘individuals’)
author_of, teacher_of
Relations connecting instances to types
is_an_expert_on
is_allergic_to
is_an_instance_of
40. An ontology is a representation of types
of entities and of the relations between
them
The result of applying an ontology to a
body of data about instances is a
knowledge base
Gene Ontology (GO) vs.
Gene Ontology Annotation Database (GOA)
40
41. Manual ontology building vs. NLP
41
Natural language processing and machine
reasoning more generally are making progress
But (so far) only ontologies built by manual
experts have proven value
42. Ontology of Philosophy
42
- text vs. structured data
- conflicts of interpretation affecting the goals
of ontology itself
- no neutral perspective
- for GO and other scientific ontologies science
itself provides a neutral perspective
- what can provide the neutral perspective here?
44. The Philosophy Family Tree
An academic genealogy of philosophers
Only one type of link: is_Doktorvater_of
• as wiki
• as indented list
• as linked graph
140,000 entries
The largest (and longest) chain of links begins
with Leibniz
44/
45. as wiki (still working)
45/
http://philosophyfamilytree.wikispaces.com
58. 58
Subject branch
• Place
– Instances: Skjolden; Cambridge
• Date
– Instances: 11 May 1936
• Issue
– Instances: philosophy; logical analysis
• Point
– Example of instance: Logical analysis is essential to philosophy
• Field (a field of philosophical discussion)
– Has subclasses:
• Epistemology
– Scepticism
» Rule-FollowingScepticism
• Perspective
– Has subclasses: APichler_Course_TLP; APichler_Course_PI
– Instances: contradiction; state_of_affairs …
59. 59
Examples of Relations
isArguedForIn
– [Philosophical analysis is essential to philosophy]
isArguedForIn [W-TLP]
isPublishedInWork
− [Ms-114,48v[5]et49r[1]] isPublishedIn [W-
PG1969:PartI:II:sect19]
isReferredToIn
– [Augustinus, Aurelius: Confessiones]
isReferredToIn [Ms-114,48v[5]et49r[1]]
60. Alois Pichler (WAB). CCPL BY-
NC-SA
60
Interlinked browsing of texts (data) and
relations (metadata)
69. Simple ontological traffic rules
1. avoid is_a overloading
2. use exclusively singular nouns and noun
phrases
3. do not suppose that A is a kind of A & B
4. true path rule (asserting A is_a B is to assert
something that is grammatical, and
universally true)
Principal lesson of scientific ontologies:
reasoning power depends on rule 4
69/
71. Breaking traffic rules
• moral rationalism is_a the a priori
• the a priori is_a epistemological sources
• epistemological sources is_a epistemology
• epistemology is_a metaphysics and
epistemology
The first generation of scientific ontologies
broke these rules too. But they have learned
since then to do it right.
71/
72. Another ontological traffic rule
• Do not populate an ontology through multiple
unmonitored human sources
• Do not create an ontology on the basis of a
single source of data
– the principal value of a well-built ontology is in its
secondary uses, uses which were not anticipated
when the ontology was first developed
72/
73. PhilOnto
An example of an Ontology of
Philosophy that tries to do it right
http://ontology.buffalo.edu/philosophome/pdcp
hilontology-v1.owl
73/
80. Features of PhilOnto
80
PRO
• Built on the basis of tested best practice
principles for ontology development
• Built to be extendible through an
evolutionary process
• Built manually, on the basis of careful
thinking about structure and definitions
93. is_a and subclass
93
change is_a metaphysics
metaphysics is_a Idea
are not helped if we read ‘subclass of’ in place of
‘is_a’
since ‘subclass of’ is to be understood set-
theoretically
what would every member of the class change is
a member of the class metaphysics mean?
95. What does ‘instance’ mean?
Colin: [it is a] kind of meaning in use, i.e., a
specification of how instances are assigned and a
contextual interpretation, supplied by end users, in
which it makes sense to say that ideas
about Japanese Zen Buddhist Philosophy are
instances of ideas about Japanese Philosophy more
generally. It is this latter, more pragmatist approach
to meaning that I prefer …
96. 6 Put more precisely, we take a computational
ontology to be a directed acyclic graph where nodes
represent concepts and the links between concepts
represent the taxonomic “isa” relation … everything
that “is a” instance of Red Wine “is a” instance of
Wine …
everything that “is a” instance of Racism “is a” instance
of African and African-American philosophy
97. Further mysteries
How is it decided what gets listed under ‘Instances’ of feminist
philosophy and what gets listed under ‘Related Terms’.
Is there any right and wrong for any of this?
100. Features of InPhO
PRO
• Impressive tooling
• Authoritative data sources such as the Philosophy
Family Tree being used to populate the InPhO
knowledge base
• Secondary uses being explored (e.g. as part of a
robotics application to try to detect contexts in
which there are ethically significant issues in play)
101. Features of InPhO
CON
• full of mysteries
• does not follow established best practices
• no concern for interoperability with other
ontologies
• no concern for correctness of is_a hierarchies and
• no concern for logical definitions (as far as I can
see)
• thus many opportunities for reasoning with the ontology
are foreclosed
102. Challenges for InPhO
• OWL provides reasoners to check consistency
• Were inconsistencies ever found when building InPhO?
• One secondary use for ontologies is to detect errors in
databases
• Can InPhO be used to detect errors in the SEP?
• One secondary use for ontologies is to enhance existing
classification and tagging systems
• Can InPhO be used to improve the classifications in
PhilPapers?
• by finding redundancies?
• by aiding more coherent classification by identifying subsumption
relations?
• via semantic enhancement?