Thoughts on an ontology of evolution

 An interesting blog entry from an evolutionary scientist with a background in philosphy with some comments by me below

http://shiftingbalance.wordpress.com/2006/09/05/recent-work-oe-the-ontology-of-evolution/

Some criticisms of the Music Ontology

Music should be a paradigmatic use case for the semantic web. Despite
popular musician’s attempts to wriggle out of being “boxed in” to any
one genre, music journos and fans love to categorise and subcategorise
bands and songs in myriad ways – by style, mood, era, geographical
origins and so on. A single textbox in iTunes doesn’t do this
justice. In addition, the webs of influence between different bands
and styles of music are the subject of many a pub discussion, and are
ripe for computable representation. And for the rock music geek there
are of course the “family trees” of band members, again easily
represented in a computable artefact.

For many people the web is all about music – swapping and consuming
mp3s legally and illegally. Modern musical forms such as the mashup
are essentially internet phenomena.

One would then expect or hope that the semantic web would deliver some
kind of killer-app in this domain; this killer app could use data on
the web, perhaps available in RDF to deliver music in some new and
exciting way that would thrill music geeks and ordinary consumers
equally. It is therefore disappointing that the self-styled “music
ontology” is so lacking in many respects.

I’ll give some of my thoughts below. The analysis I am applying comes
from my perspective in scientific ontologies based on the
philosophical position known as realism. I see no reason why many of
the same techniques are not applicable to the multimedia domain.

The MO is described here:

http://pingthesemanticweb.com/ontology/mo/

POOR DEFINITIONS

Looking at the first term which is defined, “Album”, we see what is
presumably the definition in the comment field:

This is a generic term defining a package of tracks. This applies to
compact disks, venils, CD singles, EPs, etc.

First of all there’s no excuse for spelling errors in a flagship
ontology such as this one. “venil” is presumably a vinyl record.

Like a lot of current scientific ontologies, the definitions here
conflate use/mention to some extent (eg “running is healthy and has 7
letters”). The leading “This is a generic term definining” can simply
be omitted, making the definition clearer.

Most serious of all, it is completely unclear what the entities in
reality which are denoted by this class. It is not clear whether the
term denotes the physical entity (CD, vinyl record) or

This is a generic term which is applied to solo artists, groups, and
also “various artists”.

INCONSISTENCY BETWEEN TEXT DESCRIPTIONS AND RDFS REPRESENTATION

Type:

A type of Album release (mo:Album). There are several sub-classes of
this type that explicitly define all of the release types that
MusicBrainz database currently understands

The comments state that Type is a type of Album release. This is
incoherent; but in as much as it is understandable, it leads us to
think that there would be some kind of link between Type and Album in
the ontology. In fact there is none. And nowhere do we find “release”
defined.

Naming a class “Type” betrays a characteristic systematic
methodological error found in many ontologies. In a realist ontology,
all classes correspond to type (or universals), which are instantiated
by individuals in the world. Even if the MO is not explicitly realist,
calling a class Type seems odd. Are the instances/individuals of this
class themselves Types, making the MO a kind of second order ontology?

Type has many subclasses, such as Remix:

A release (mo:Album) that primarily contains remixed material.

The text definition states Remix is a release (Album), but again,
there is no link between Remix and Album outside the incomprehensible
text comments presumably intended for humans

Also, if we look the comments for Track (A track on a release
(mo:Album). The term is not limited to simply music as it could cover
a spoken word track, an audio book, etc) we would expect some kind of
link between track and album. In fact no such link is stated.

MIXING EPISTEMOLOGY WITH ONTOLOGY

The ontology commits the cardinal sin of mixing epistemology with
ontology; it actually contains a class “Other”:

Any release that does not fit or can’t decisively be placed in any
of the other categories

See
http://en.wikipedia.org/wiki/Celestial_Emporium_of_Benevolent_Recognition
for this approach taken to the extreme

There is no need for an “other” category. Simply place instances in the superclass category and have the application deal with epistemological aspects

POOR REPRESENTATION OF DOMAIN

Remix:

A release (mo:Album) that primarily contains remixed material.

This incoherent definition is at odds with how the term Remix is used
by domain experts such as DJs and so on. First of all, many remixes
(sensu music fan) are not released (they may be played live as part of
a DJ set for example). Secondly, they are typically not albums, but
rather tracks (or often serial sequences of tracks with fiat
overlapping boundaries)

Live:

A release (mo:Album) that was recorded live.

This definition is circular, and it’s not clear what this means,
especially in the context of a lot of current music. Recorded in front
of a concert audience? Recorded in a continuous segment, without
remixing etc?

The choice of label is also poor. “Live” would be best reserved for a quality that can inhere in entities such as a musical recording. “Live Album” would be a far better choice of label. Labels are extremely important for the human users of an ontology.

If this is to be a serious ontology within its domain then the maintainers should ensure that domain experts from all across the domain have input such that the ontology is not slanted to the ontology maintainers favourite portions of the domain

DOMAIM NOT CLEARLY SPECIFIED

The blurb on the website states:

The Music Ontology is an attempt to link all the information about
musical Artists, Albums and Tracks together: from MusicBrainz to
MySpace. The goal is to express all relations between musical
information to help people finding anything about music and
musicians. It is based around the use of machine readable
information provided by any web site or web service on the Web.

In fact the ontology contains Audiobook, “An audiobook is a book read
by a narrator without music.”, as well as Interview and Spokenword.

Of course, audiobook is close to music album in semantic space. They
are frequently distributed on similar media, they are audio
content. But the consequences of expanding the domain in this way are
not clear. Are the authors and narrators both “Artists”?

One would expect there to be some kind of representation of style or
genre in a music ontology. Of course, even a stuffy formalist such as
myself would advocate letting a thousand flowers bloom here. I’d love
to see different fans producing different music style ontologies and
try out various different mapping and alignment ontologies on them -
this could be a really fun SW app.

What is curious is the MO does not include a single recommendation of
how style should be captured. One would expect at least an open-ended
has_genre rdfs property. If the MO purports to link “all information
about Artists, Albums and Tracks together” or to “help people finding
anything about music and musicians” then surely style/genre must come
into this?

It would seem that the MO is in fact more of a Music *Business*
Ontology combined with an Audio *Media* Ontology, as there is very
little about the domain of “music” as a musician would presumably
understand the term. I would expect a music ontology to represent
entities such as pitches, frequencies, instruments, timbres, sequences
of sounds, keys and so on.

INCORRECT USE OF RDFS SEMANTICS

We see a classic mistake in the domains and ranges of remix_of:

<rdf:Property rdf:about=”http://purl.org/ontology/mo/remix_of”
rdfs:label=”remix_of”>
<rdf:type rdf:resource=”http://www.w3.org/2002/07/owl#ObjectProperty”/>
<rdfs:comment>Used to relate the remix of a work in a substantially altered version produced by mixing together individual tracks or segments of an original source work.</rdfs:comment>
<rdfs:domain rdf:resource=”http://purl.org/ontology/mo/Album”/>
<rdfs:domain rdf:resource=”http://purl.org/ontology/mo/Track”/>
<rdfs:range rdf:resource=”http://purl.org/ontology/mo/Album”/>
<rdfs:range rdf:resource=”http://purl.org/ontology/mo/Track”/>
<rdfs:isDefinedBy rdf:resource=”http://purl.org/ontology/mo/”/>
</rdf:Property>

This means that anything which is the domain or range of remix_of is
*both* an Album and a Track. If these two classes are declared
disjoint (it is not clear if they should be), then this will be an
“empty class”

DOES NOT TAKE TIME INTO ACCOUNT

Like most semantic web resources (ontologies and ontology
representation languages included), time is not treated seriously.

member_of is a binary relation (which is forced when using rdf/owl
properties to represent relations). The MO homepage gives as an
example of use:

<mo:Artist rdf:about=”http://mm.Music.org/artist/2f58d07c-4ed6-4f29-8b10-95266e16fe1b”>
<rdfs:label>Dave Mustaine</rdfs:label>
<mo:member_of rdf:resource=”http://mm.Music.org/artist/65f4f0c5-ef9e-490c-aee3-909e7ae6b2ab”/>

The URL is enresolvable but from the other examples on the web page
appears to reference Metallica. In fact, as every heavy metal fan
knows, Dave Mustaine was one of the original members of Metallica,
before his legendary falling out and subsequent formation of
Megadeth. It is not clear how this vital fact is to be represented
with MO.

RECOMMENDATIONS

The developers of MO may like to look at formal ontology to guide
future development of MO. For example, mereotopology (the study of
parts and wholes) would be useful for understanding how musical wholes
can be broken into parts and subparts. The current MO partition of
musical entities into Tracks and Albums perpetuates an ontological
division forced onto use by media that are rapidly becoming
outmoded. Many forms of music, such as classical music and overblown
70s concept albums with their more recursive part-whole structures
require a less rigid breakdown.

Formal ontology recognises a distinction between bona-fide and fiat
bounadaries. A bona-fide boundary is an objective division in reality,
such as that between a table and sourrounding space, or between two
tracks separated by silence. A fiat boundary reflects a continuity in
nature, such as the fiat boundary between my arm and torso.

Soon-to-be-outdated media forces bona-fide musical divisions upon
us. In fact, with the advent of turntables as a musical instrument in
the 70s mereological divisions between musical parts in popular music
has frequently been along fiat boundaries. It is to be expected that
the internet will do more to dissolve these divisions than perpetuate
them.

Formal ontology is clear about the different kinds of entities in
reality. The MO does not seem clear on the difference between actual
physical instantiations of media or content bearing entities such as
an actual physical CD and the actual content which is carried by that
media. This is no problem for everyday linguistic usage; a human
understands the different sense of “album” in the questions “how many
albums have franza ferdinand put out” vs “how many albums have franz
ferdinand sold”. Howevers, computers are less intelligent and cannot
disambiguate in this way and thus it seems crazy not to make this
explicit in the ontology.

Music, language and living entities share certain essential characteristics. It would be interesting to try and share foundational relations. For example: bands and musical styles are continuants which endure through time, gaining and losing parts as they progress, just like a human composed of cells. Much of the work that has been done in foundational ontologies in biology (such as RO) could perhaps be applied here – eg developmental relations between different musical forms, fission and fusion between bands.

The recommendation for more formality in the semantic web may seem
curious, given that SW practitioners are keen to distance themselves
from any kind of stuffiness in order to attrack the folksonomy crowd.

In fact both can exist side by side; I want both. I’d like to tag
musical content and content-producers with ad-hoc tags like “cheesy“,
“evil“, “lively“, “mutant“, “good music for hangovers” etc. But I’d also like to
get consistent answers back in searches for musical content across the
semantic web and the only way to do that would seem to me to be using
an ontology that is built on better ontological principles

Perhaps the best place to start would be to solicit some use cases; as it stands now it’s not clear exactly what problems the MO wants to solve, and which problems it is capable of solving.

flying death-squirrel, aarrrghh

totally awesome front cover picture on this month’s nature:

http://www.nature.com/nature/journal/v444/n7121/index.html

something about this artist’s rendition reminds me of 80s heavy-metal album covers

This chap’s name is ‘Volaticotherium antiquus’, which means ‘ancient gliding beast’, which is a fairly heavy metal name too.

It’s 130 million years old, and may be representative of a new mammalian order

On a more serious note, there is a nice review section on obesity and diabetes which is nice and timely for a test pilot project I am working on with the Disease Ontology

Linnæus’s taxonomy of mental disease

I heard a reference to this at the recent workshop on the Ontology of Disease – everyone associates Linnaeus with his taxonomy of organisms, but this curiosity seems to be a little bit more obscure. I did a little googling and managed to find various pages mentioning a categorisation of mental disease. I’d love to see the original manuscript.

I cheekily added this as a slide during a recent talk to some eminent experts on fish evolution at a NESCENT workshop and was chuffed to discover that they’d never heard of it before. Some of the subspecies got a few laughs.

Here it is:

• Mental – (genus)
  • PATHETIC – (species)
    • citta — desire to eat what is not food
    • bulimia — insatiable desire for food
    • polydipsia — continuous desire for drink
    • satyriasis — enormous desire for sex
    • erotomania — indecent desire for lovers
    • nostalgia — desire for country and relatives
    • Tarantismus – desire for dancing, often caused by an insect bite
    • rabies — desire to bite and lacerate the harmless
    • hydrophobia — aversion to drink
    • cacositia — aversion to food, accompanied by horror of it
    • antipathia — aversion to a particular object
    • anxietas — aversion to ordinary things, with pain in the heart