Re: discussion about meaning and senses

Hi all,

Just back from holiday myself I have read the discussion with interest, and
would like to provide my 2p while still fresh.
I see there are two threads now, I hook into this one because some of my
comments are based on Guido's mail.

I agree with the requirement of interoperability and reasonability for
lexical meaning/ontological classification, and a liberal, inclusive
approach to capturing lexical senses and concepts.

I am not sure about the assumption that there are aspects of meaning that
should be kept in the lexicon, and others which should be reified into an
ontology.
This dichotomy seems rather artificial, and maintaining it would lead to
the first of the group aims Guido puts forward.
I would advocate not to define a formal definition of meaning according to
which new resources should be created and existing resources should be
restructured, and lean towards Guido's second group aim, the integration of
resources and practices.
This would shift our activities towards linking and incremental conceptual
integration rather than towards full formal analysis.

Some thoughts:

1. In LOD, lexical, terminological and ontological meaning has already been
mixed up because meaning has been formalized in ontologies such as
Ontowordnet, DUL, and DbPedia.
Moreover, the reification of linguistic structure, be it mediated through
LMF, OntoWordNet, DbPedia, Gold etc., provides in my view all the
ingredients that enable lexical meaning to be interoperable and reasonable,
at least to the extent that meaning is covered in the ever expanding sets
of LOD nodes.
LMF is also available in owl format, so any lexical sense encoded in LMF
already has ontological status as a class. All LMF's additions regarding
other linguistic aspects from Piek's list that define lexical meaning
(syntax, morphology etc.) will in principle be available if the LMF-style
ontology template has been properly populated.
There are various descriptive systems that provide an alternative to LMF.
For instance, the further development of SKOS-XL aims to cover lexical
aspects as opposed to the ontological classification the label lexicalizes.
I think that SKOS-XL provides a model that is closest to separating lexical
meaning from conceptual information.

2. As Aldo's tool shows, the resolution to these LOD nodes is the central
SW activity we need to take into account for any meaning standardization
effort.
I don't think we should not be prescriptive and try to standardize this
rather anarchic SW expansion and evolution of LOD semantic description by
imposing semantic rigor by means of an ideal descriptive system of
axiomatization (not that we could or would want to stem this tide anyway).
Many standardization processes have suffered in their uptake because they
are too specific or prescriptive.

3. There are various systems of meaning description out there whose linking
will not only establish semantic interoperability, but also practical
interoperability with ongoing activities of linguistic comparison and
linking such as those on the Open Linguistics Workgoup (
http://linguistics.okfn.org/).

3. Concentrating on this interoperability entails trying to formalize links
between semantic web resources. The resulting network can be seen as a
formalization of lexical meaning may enrich LOD resolution by e.g. using
atomic meanings from one resource as constituents of a semantically more
complex concept from another resource. This ties - at least to a certain
extent - in with Aldo's "degree of formalization".
SW meaning instances are mostly atomic and heterogeneous. Picking senses
apart according to a prescriptive system does not seem the right way
forward.
It seems to me that what we need for the semantic web to work with meaning
is a standardized linking mechanism.
This could take the form of:

a) The adoption or definition of a set of mapping relations in the form of
object properties, for instance:
- owl:sameAs
- EuroWordNet/Cornetto relations
- Lemon relations
- the SKOS mapping relations broadMatch, close-Match, exactMatch,
narrowMatch and relatedMatch.

b) The full reification of mapping relations within a mapping model. This
option promotes mappings to first class citizens (i.e. models them as
classes rather than properties). The advantage of this reification of
mapping relations into classes is that it allows us to describe them as
ontological objects and model the relations in a fine-grained and
extendable fashion at the cost of a higher level of complexity. It should
also enable reasonability.

I would favour b).
Using such a model we will be able to capture the nature of subtle semantic
differences between any pairs of lexical meanings and ontological
conceptual classes. We can resolve and formalize these differences
incrementally, and query them in a uniform fashion.
I would hope that linking will incrementally promote the level of
formalization of the meanings/concepts involved, because an increasing
number of semantic aspects will be resolved.

Sorry for this long comment, hope hope at least some of it makes sense.

Best wishes,

Wim

-- 
Dr. W. Peters
Research Fellow
Natural Language Processing group
Department of Computer Science
University of Sheffield
Regent Court
211 Portobello Street
Sheffield S1 4DP
tel: 00-44-114-2221902
fax: 00-44-114-2221810
email: w.peters@dcs.shef.ac.uk

On Mon, Aug 20, 2012 at 5:26 PM, Guido Vetere <gvetere@it.ibm.com> wrote:

> Philipp,
>
> thanks for your summary. Actually, there may be different visions on
> fundamental aspects of our work; although we can get along with this in a
> pragmatic way, I think that is worth discussing them a little bit more, to
> see if we can get to an agreement, or at least to agree on the nature of
> our disagreement :-)
>
> I think that we have to agree on the meaning of 'ontology', or at least be
> aware of the polysemy of this term. Whereas philosophers have mostly
> intended 'ontology' as a theory of what exists in the reality (including
> 'socially constructed reality') independently of linguistic habits,
> computer scientists, in general, use 'ontology' to refer to any formalised
> conceptual schema, whether it contain common sense notions (e.g. 'bald') or
> metaphysical ones (e.g. 'substance'). Of course, in the latter sense,
> 'ontology' may be considered as a kind of formal representation of
> linguistic meanings, hence there is obviously a continuum from lexicons to
> ontologies.
>
> Sorting this continuum based on the 'degree of formalisation', may help in
> the integration of different onto-lexicons, but it would be quite difficult
> to find out a criterion to estimate the formality of the representation of
> a linguistic sense. Pick up 'bald', for instance. How many axioms would you
> need to formalise this concept? Would a fuzzy logic provide a better
> formalisation than a typicality logic? Who knows? From a 'normative'
> perspective, on the other hand, we could tell people how to better organise
> their onto lexical content on the Web. As you know, we experimented an
> approach where 'ontology' is taken 'philosophically', hence the issue of
> distinguishing common sense, linguistic (i.e. mostly vague) concepts from
> 'metaphysical' categories is taken into account. This is achieved by
> implementing a semiotic approach, i.e. pushing the 'sign' upfront, much in
> the line of what Aldo says. Of course, I think that there are benefits in
> adopting a 'native semiotic' kind of modelling, and we can discuss it, but
> as a matter of facts most of the (so called) ontologies published on the
> Web provide a mix of (what we consider as) categories, linguistic senses,
> whatever.
>
> So, if the aim of this group is that of providing guidelines for building
> new resources or restructuring existing ones, then we should reach an
> agreement on whether there is a distinction between ontological categories
> and linguistic senses. If, on the contrary, we focus on integrating
> existing resources and practices, then we must assume that such a
> distinction is just irrelevant. Which one is the case?
>
> Regards,
>
> Guido Vetere
> Manager, Center for Advanced Studies IBM Italia
> _________________________________________________
> Rome                                     Trento
> Via Sciangai 53                       Via Sommarive 18
> 00144 Roma, Italy                   38123 Povo in Trento, Italy
> +39 06 59662137                     +39 0461 314031
>
> Mobile: +39 3357454658
> _________________________________________________
>
> Philipp Cimiano <cimiano@cit-ec.uni-bielefeld.de> wrote on 14/08/2012
> 08:59:59:
>
> > Philipp Cimiano <cimiano@cit-ec.uni-bielefeld.de>
> > 14/08/2012 08:59
> >
> > To
> >
> > "public-ontolex@w3.org" <public-ontolex@w3.org>
> >
> > cc
> >
> > Subject
> >
> > Re: discussion about meaning and senses
> >
> > Dear Guido, Aldo, Piek
>
> > I have been following the discussion in silence so far as I have
> > been traveling and not found the time to summarize the current state
> > of the discussion.
> >
> > First of all, apologies for my misconception of what bald means ;-)
> >
> > But aisde from this misconception, my main point actually was that I
> > do not think that we need to strictly separate the ontological from
> > the 'ideal' or 'linguistic' meaning as Guido is advocating.
> > In fact, I like the point raised by Aldo that we could see lexicons
> > and ontologies not as opposites but rather as points along
> > a continuum where there can be more or less formalization (I hope I
> > got Aldo right, did I?)
> > So this means that meanings can be highly formalized or not
> > formalized at all. "Bald" would be a very vague concept along these
> > lines which would be possibly not axiomatized at all other than
> > saying sth. like $\forall x Bald(x) \rightarrow Human(x)$ which is
> > more or less
> > what in Senso Comune is expressed by the "characterizes Humans".
> > So I like the perspective that in principle we *can* formalize any
> > aspect of the meaning of words, but for pragmatic reasons we might
> > decide not to do it (because the effort is too high and it does not
> > pay off / the inferences we could draw and not particularly
> > interesting for applications, etc. or there is no agreement on how
> > to formalized). These are all pragmatic reasons, but no principled
> > ones again the formalization of the meaning of some lexical element.
>
> > So let me try to summarize the state of play of the discussion:
>
> > Guido (1): we clearly need senses as reified objects, reifying the
> > association between lexical entries and concepts.
> > This is necessary because we need to predicate over these
> > associations. For instance, many 'senses' come with specific
> > grammatical constraints, e.g. for nouns, plural is often used to
> > mean something different from a mere collection of individuals, as
> > in Italian 'acqua' (water) and 'acque' (thermal treatments).
> > => I fully agree with this and in fact such a reification is at the
> > core of our lemon model for the same reasons as mentioned by Guido.
>
> > Guido (2): Linguistic and ontological meanings should be clearly
> > distinguished. In Senso Comune 'meanings' are regarded in most cases
> > as vague and idealized meanings that are not properly formalized, so
> > that interpretation is subjective and can vary depending on the context..
> > => I would not agree with this point and rather adhere to the
> > continuum views that Aldo has been putting forth.
> > The consequence of such a view is that there is in principle no
> > problem with what Aldo has called "the direct mapping" approach that
> > I was advocating.
>
> > Aldo (1): we should see lexicons and ontologies rather as a
> > continuum than as an opposition. If possible, and for the sake of
> > semantic interoperability and reasonability, concepts should be
> > axiomatized as far as possible, but if for pragmatic reasons such as
> > formalization is not provided, then a semi-formal definition as
> > provided in a lexicon (e.g. through a gloss or lexical
> > relationships) is better than nothing. In the long-term, such
> > "informal" meanings might be incorporated into the ontology by
> > axiomatizing them appropriately.
> > => I share this view.
>
> > Piek (1): was posing the question of whether there is any
> > fundamental line between what should be formalized and what not.
> > Clearly, I do not have a definite answer to this, but all those
> > lexical and linguistic properties that Piek mentions (morphology,
> > pronunciation, lexical relations) etc. should be definitely
> > modelled/represented at the lexical side (e.g. in OWL, which does
> > not mean that they are axiomatized along the lines of the argument
> > above). In this sense I do indeed advocate the two modular layers: a
> > lexical and a semantic/ontological one. Of course, the crucial
> > question is whether we have to model the semantic implications of
> > certain linguistic properties/distinctions. Take the word "dog" and
> > its plural "dogs". Clearly, they have different semantics, as "dog"
> > is used to refer to one element in the set of dogs, while "dogs" is
> > used to refer to a set of at least two elements in the set of dogs.
> > However, this is purely linguistic knowledge that is "systematic" in
> > the sense that it holds for all nouns. As such, it is questionable
> > whether this should be modelled in the lexicon-ontology interface.
> > But I agree that the interesting question is which aspects of the
> > meaning of lexical entries that touch the actual interface should be
> > formalized and how (one example is register).
>
> > By the way: both layers can support reasoning, but in a different
> > domain. In one layer, we reason about linguistic properties, while
> > in the ontological layer we would reason about (domain) concepts.
> >
> > So there are certain properties in the list of Piek that are purely
> > linguistic and do not touch the meaning layer at all (e.g
> > morphology, pronunciation, etc). while there are others that
> > modulate the meaning. The latter ones are the ones we should take
> > about in more detail.
> > In any case, I think it would be worth going through the bullet list
> > of Piek in our next telco to reach some consensus there.
>
> > Just my two cents for now.
> >
> > Best regards,
> > Philipp.
> > --
> > Prof. Dr. Philipp Cimiano
> > Semantic Computing Group
> > Excellence Cluster - Cognitive Interaction Technology (CITEC)
> > University of Bielefeld
> >
> > Phone: +49 521 106 12249
> > Fax: +49 521 106 12412
> > Mail: cimiano@cit-ec.uni-bielefeld.de
> >
> > Room H-127
> > Morgenbreede 39
> > 33615 Bielefeld
>
>
>
> IBM Italia S.p.A.
> Sede Legale: Circonvallazione Idroscalo - 20090 Segrate (MI)
> Cap. Soc. euro 347.256.998,80
> C. F. e Reg. Imprese MI 01442240030 - Partita IVA 10914660153
> Società con unico azionista
> Società soggetta all’attività di direzione e coordinamento di
> International Business Machines Corporation
>
> (Salvo che sia diversamente indicato sopra / Unless stated otherwise above)