[Top] [All Lists]

Re: [ontolog-forum] FW: FW: CfP 11/16/2015: Knowledge-Based AI Track at

To: Thomas Johnston <tmj44p@xxxxxxx>, "[ontolog-forum] " <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "Obrst, Leo J." <lobrst@xxxxxxxxx>
Date: Tue, 22 Sep 2015 14:03:10 +0000
Message-id: <CY1PR09MB082669186F787ADA96B7BC72DD450@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>



The notion of semantic fields goes back to earlier generations of linguists. John Lyons [1] devotes a chapter (ch. 8) in the first volume to that subject, providing a history that primarily points to Trier’s work in the 1930s, but which built on others.


Often there is some confusion between “semantic fields” and “lexical fields”, with the former being primarily conceptual (dealing with meaning) and the latter lexical (as one might think, dealing with form).


Until recently linguistic semantics primarily focused on “externalist” issues, as you call them, i.e., word, sense, and reference. And most realist ontologists are externalist. But with the rise of corpus linguistics and statistical models of language, distributional semantics has emerged. In addition, of course, for some time, there has been “cognitive” linguistics (Lakoff, Fauconnier, and many more, up to and beyond Gardenfors) and “cognitive” semantics. These latter trends are mostly “internalist”, hence often more concerned with lexical semantics and cognitive “concepts”. In the literature I often find cognitivists as kind of reinventing the notions and theories of semantics primarily derived from Frege, Russell, and the formal strain, leading today to that built on model-theory: see, e.g., [2, 3], mentioned previously in this forum.





[1] Lyons, John.  1979.  Semantics 1 & 2.  Cambridge: Cambridge University Press.

[2] Margolis, Eric; Stephen Laurence. 1999. Concepts: Core Readings. Cambridge, MA; London, England: The MIT Press.

[3] Concepts. http://plato.stanford.edu/entries/concepts/.


From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx [mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx] On Behalf Of Thomas Johnston
Sent: Monday, September 21, 2015 1:54 PM
To: Thomas Johnston <tmj44p@xxxxxxx>; [ontolog-forum] <ontolog-forum@xxxxxxxxxxxxxxxx>; [ontolog-forum] <ontolog-forum@xxxxxxxxxxxxxxxx>
Subject: Re: [ontolog-forum] FW: FW: CfP 11/16/2015: Knowledge-Based AI Track at 2016 FLAIRS




On the 18th, Leo asked me: "Do you mean you are interested in “semantic fields”, e.g., https://en.wikipedia.org/wiki/Semantic_field. Or developing a lexical theory that would account for such?"


To begin with, what do I understand a semantic field to be? I need to clarify this because the _expression_ "semantic field" is new to me, and so my understanding of it will not be the understanding that the current linguistics community has (or, at least, not yet).


I'll begun with the Wikipedia reference Leo provided. (After that, I want to move on to Jackendoff.)


That reference says: "In linguistics, a semantic field is a set of words grouped semantically (that is, by meaning), referring to a specific subject."


This is ok to get started with, but it's a pretty loose definition, and for two reasons. First, it uses "meaning" (in "grouped ... by meaning") in an _expression_ intended to help define an _expression_ which contains "semantic" (namely, "semantic field"). Second, "specific subject" is vague. For any collection of words, semantically-related or not, we can always find a "specific subject" for them by "ontological ascent". For any noun phrase, isn't "thing" (or, perhaps, "thing" or "stuff") a "specific subject" (or, at least, a shared subject)?


I'd like to come up with a definition of "semantic field" that would be clear enough to bear significant theoretical weight. My current stumbling point is that I can't find a way to draw the boundaries of semantic fields. That would mean a criterion that would distinguish between (i) a given set of semantically-related lexical items ("words" for short) constituting a semantic field, from (ii) other words related to those words, but not part of that semantic field. In other words, the definition I want would answer the question "Why are some semantically-related words 'inside' a given semantic field, whereas others, related to them, are not?"


This, perhaps, is what "specific subject" is supposed to do, to make this distinction. Perhaps the idea is that, for a selected group of semantically-related words, there is a lowest-level (most specific) ontological category that they can all be predicated of. Those words, plus that category, are a semantic field relative to that category, and a subset of the (unspecified) maximal collection of words also predicable of that category.


And what kind of semantic connections count for binding words into a semantic field? I'd like to think that we can distinguish between "attractive" and "repulsive" semantic "forces". Two antonyms have a repulsive force between them; two synonyms have an attractive force between them. Two homonyms may or may not have any semantic connections. Two polysemic senses of the same word have an attractive force. Over time, though, the polysemic force between them may weaken so much that it "breaks", and the result is a pair of homonyms which may then continue to semantically drift apart.


Hyponymy may seem to be an attractive semantic force, since words in the same branch of a semantic tree are thereby related. But, by the same token, words on different "twigs" of the same branch would thereby have a repulsive semantic force. In other words, two sub-trees can be pulled out from any multi-level semantic hierarchy and, from the point of view of those sub-trees, there is a negative semantic force between any pair of words consisting of one word from each sub-tree. But conversely, any pair of sub-trees in a semantic hierarchy are part of some higher-level tree and, from that perspective, have an attractive semantic force.


But I'm still not ready to say that this discussion provides me with a theoretically well-grounded concept of a semantic field because any definition of "semantic field" still seems to me to run in too-tight a circle. So this leads me to provisionally treat "semantic field" as a heuristic, meaning that it is any collection of words that the person positing the semantic field feels are semantically related to one another, and all related to (predicable of?) a "specific subject" (which is itself a lexical item).


So: provisionally taking the concept of a semantic field to be a heuristic, my answer to Leo's question is that I am not interested in semantic fields per se, but that I am very interested in "developing a lexical theory that would account for such". Which would mean, among other things, providing a theoretical context which would support a rigorous definition of "semantic field" and which would, among other things, (i) define rigorous boundary conditions for semantic fields, and (ii) semantic relationships among semantic fields, etc.


I want to stay with semantic fields awhile longer. So I'm now moving on to Ray Jackendoff, in particular (i) Semantics and Cognition (1983), and (ii) Foundations of Language (2002). But as well as the use he makes of semantic fields (especially in the later book), Jackendoff takes a position on the internalism/externalism debate. He is, albeit in a qualified sense, an internalist about semantics, a position for which meaning is "in the head". Philosophers, on the other hand, often take an externalist stand, in which meaning is a relationship between words and things.


I'd like to develop the notion that this distinction is a (very big) tempest in a teapot. (In doing so, though, I suspect that the best I will be able to do is provide a series of footnotes to Kant.)


More to come . . . . .





On Saturday, September 19, 2015 12:20 PM, Thomas Johnston <tmj44p@xxxxxxx> wrote:


(This is my second and last catch-up post.)





The research of yours that you mention is right in my own field of interests. I would appreciate any url links you have to that work.


Concerning the "drift" of "semantic fields" that you refer to, the question that interests me is "What causes the drift?" I think that perhaps there are two things that account for it.


The second, derivative, factor is statistical. Semantics in a language is derivative on, ultimately, the state of each individual's semantic net at a point in time. Continuities in those semantic net states over time and over small to large communities of speakers are the necessary condition of the use of language to communicate. Those statistical patterns are periodically formalized in both general and special-purpose dictionaries, whose definitions are based on guesses that lexicographers make about the statistical patterns in those states over person and over time, and the aggregate state reached in a given language community at the point in time the definitions are formulated.


The first, basic, factor is the plasticity of the human brain. Over time, and under the influence of what we read and of our verbal communications with others, we evolve neurally-based dispositions to use, and to accept as valid, pairs of words/expressions.


At the weaker end of the dispositional spectrum for a given pair of expressions, we have dispositions that correspond to statements expressing supposedly factual regularities. These are Kant's synthetic a posteriori statements, empirical generalizations open to revision.


At the stronger end of the dispositional spectrum, we have dispositions that correspond to statements expressing accepted linguistic conventions. These are Kant's analytic a priori statements, ones not open to revision (except by revising the semantic rules the statements express).


This dispositional spectrum is what accounts for the analytic/synthetic continuum, and what explains why, as Quine demonstrated, there is no analytic/synthetic dichotomy (thus nailing shut the coffin of logical positivism).


Which brings to mind an example provided by A. J. Ayer. He used the statement "Loadstones attract iron" as an example, and suggested, quite reasonably, that this statement must have begun life as an empirical generalization, but ended up as an analytic statement, reflected in the fact that we just wouldn't count anything as a loadstone if it didn't attract iron.


In an example I discussed in my dissertation, a semantic key to the debate between pro- and anti-abortion advocates is whether or not a fertilized human egg is a human being. For anti-abortionists (at least those of a religious persuasion), "A fertilized human egg is a human being" is what Kripke would classify as an analytic a posteriori statement -- analytic because no empirical evidence would be counted by them as evidence against the statement, and a posteriori because it is about "matter of fact".


Reverting to Ayer's example, the semantic drift of the pair "loadstones" and "attract iron" was driven to an increasingly strong disposition to disallow counter-examples. Certainly this semantic drift ended up with the disposition to disallow counter-examples first in the scientists of that day, then eventually in the wider community of speakers. The drift, for this pair of expressions, was driven by the increasing scarcity of statements purportedly expressing counter-examples. Each person's semantic net evolved over time from a state in which counter-example statements would be at least considered, to a state in which they were no longer considered.


For each such person, over time, patterns of linguistic usage strengthened the connection between the two terms until the statements expressing those connections became "true by definition". Across increasingly wider linguistic communities, the statistical aggregation force led to lexicographic revisions of the relevant dictionaries.


But this drift across the dispositional strength spectrum is not itself free. "Loadstones" and "attract iron" are each connected with a large number of other potentially co-occurring expressions, some of which would pull (via the dynamism of neural connectivity) against that particular drift of that particular pair of expressions. And so I reach a pale wash of neural metaphor over Quine's conceptual holism.


I'd like to know if you can tie this informal description of the proto-theory I have been working on for four decades to academic work you are familiar with.


Regardless, thanks for already responding to my original message to you.






Leo, you asked me, earlier, to clarify some of what I said above, explaining in more current terminology what my interests are, and where they are situated in the corpus of current work. I am busy reading your Geeraerts references right now, and hope, with that, to be able to respond in a week or so. 


I have a lot of catching-up to do!




On Saturday, September 19, 2015 12:05 PM, Thomas Johnston <tmj44p@xxxxxxx> wrote:


OK Leo. Here's my re-post, to the forum, of what I've posted earlier to you. 


For others, the topic I'm concerned with in these postings is lexical semantics -- the semantics of sub-sentential expressions. Also, I have begun reprising some of my ancient notes on this topic, and they may be difficult to follow because over three decades of work in lexical semantics have taken place since I wrote them. And so there may be terminology I use that needs clarification. Also, my current research in this field is at a rudimentary level, and so there may be developments that, once I am aware of them, will point out the error of my ways.


Contributions from others in this forum with interests in lexical semantics are most welcome.


And so, what I wrote earlier was this:





Thank you for the overview of distributional semantics which, in fact, I was unaware of. Also for the references you provide. 


The first thing that comes to mind is that the lexical use patterns which these statistical techniques will certainly reveal / have revealed need a theory, an explanation. Recently, I have gone back to fairly extensive unpublished material which I wrote in the 70's and 80's. Combining it with my current research and notes, I have a corpus of work which I provisionally think might form the basis of such a theory. 


Currently, I'm trying to figure out how much of Gardenfors' new book, The Geometry of Meaning, has already anticipated my work on such a theory. Unless you have already concluded (as I nearly have) that his conceptual spaces won't carry all the weight he puts on them, you might find this book of his interesting.


I learned my compositionality lessons from Jerry Fodor's extended development and defense of the Language of Thought. But what still fascinates me is the psychological phenomenon of a child's progression from pointing and naming to his construction of elementary sentences. It seems to me to be one of the miracles/mysteries of human intellectual achievement. I look forward to an ANN account of compositionality, from which point of view Fodor's symbol-based LOT will be seen as an abstract description of what neural networks do, and which will begin the process of removing the mystery from this miracle.


But the semantic forces which account for the statistical patterns discussed in your references are still what interest me the most. I can get into the literature starting from those references, of course; but if there is anything else you know of that is specifically concerned with a theory of lexical meaning (and lexical meaning change), please let me know.


Thanks once again.






(There is one more catch-up post I will publish to the forum, and then we're all starting from the same page.)




On Friday, September 18, 2015 7:50 PM, "Obrst, Leo J." <lobrst@xxxxxxxxx> wrote:




[I originally posted this just to you, but we have agreed to share our exchange and invite others into the discussion.]


As you undoubtedly know, recently there’s has been the emergence of so-called “distributional semantics” these days in computational linguistics/NLP. This is based on corpus linguistics, i.e., large-scale statistical, but light-weight “knowledge-based” or formal linguistic/semantic methods.


Distributional semantics:  words as “meaning” the contexts/collocations they can occur in, what I consider kind of Witgenstein 2 (Investigations, not Tractatus) in nature.


Some folks are trying to combine distributional semantics with more formal compositional semantics, the latter of which has not focused primarily on lexical semantics, but rather on the composition of words into higher forms and their semantic  interpretations, i.e., going back to Montague in the late 1960s.  E.g.,  see [1, 2]. Also, for a good overview of lexical theories, see [3]. For a new type-based approach, see [4].


However, there are potentially some useful emerging approaches in ontology research, i.e., quality (or value) spaces, and so-called semantic reference spaces/ranges, especially [5-6]. We are using this in our current clinical care healthcare ontology research (forthcoming), which can combine quantitative and qualitative quality value spaces, so that, e.g., nominal qualities (“named” qualities; think of “low/medium/high X”) can be mapped into a quantitative range, though imprecisely, given that you have some ordering on the regions. I am myself thinking of something similar to this for so-called “semantic fields”, i.e., that one can begin to think of these spaces and their points/regions as “drifting” over time.





[1] Lewis, M., & Steedman, M. 2013. Combining Distributional and Logical Semantics. Transactions of the Association for Computational Linguistics, 1, 179-192. https://tacl2013.cs.columbia.edu/ojs/index.php/tacl/article/view/93.

[2] Baroni, M., Bernardi, R., & Zamparelli, R. 2014. Frege in space: A program of compositional distributional semantics. Linguistic Issues in Language Technology, 9. http://csli-lilt.stanford.edu/ojs/index.php/LiLT/article/download/6/5.

[3] Geeraerts, Dirk. 2009. Theories of Lexical Semantics. Oxford University Press.

[4] Asher, Nicholas. 2011.  Lexical Meaning in Context: A Web of Words. Cambridge: Cambridge University Press. 2011.

[5] Probst, F. 2007. Semantic Reference Systems for Observations and Measurements. PhD dissertation, U. Muenster, Germany. http://ifgi.uni-muenster.de/~probsfl/publications/PROBST-Thesis-SemanticReferenceSystemsForObservationsAndMeasurements.pdf.

[6] Probst, F. 2008.  Observations, measurements and semantic reference spaces. Applied Ontology 3 (2008) 63-89.




From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx [mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx] On Behalf Of Thomas Johnston
Sent: Tuesday, September 15, 2015 11:42 PM
To: [ontolog-forum] <ontolog-forum@xxxxxxxxxxxxxxxx>
Subject: Re: [ontolog-forum] FW: CfP 11/16/2015: Knowledge-Based AI Track at 2016 FLAIRS




The description of knowledge-based vs "statistics"-based AI is an entry point into so much that I am interested in. Perhaps the distinction will eventually cease to be a distinction between the association of ontologies with the former but not with the latter. Perhaps the distinction will eventually become one between software systems which are given human-developed ontologies, and software systems which abstract their own ontologies from the patterns they discover by naming those patterns and then developing additional patterns by playing with set-theoretic quasi-random combinations of those named patterns.


And the way we play these set-theoretic games in our heads, I have suspected for a long time, is by means of background processes in which Venn-diagram-like representations of those labelled patterns are what is actually manipulated, at close to the neural level. (This is a bit like Gardenfors, but also quite a bit unlike him.)


I suspect this is far too brief, and also far too off the beaten academic path, to be more than vaguely suggestive, if even that. The interests that it hints at are (i) my interest in lexical semantics, vs. what I see as a one-sided concentration on the semantics of statements on the part of philosophers especially; and (ii) my interest in diachronic semantics, i.e. in how the semantic web that is embodied in the neurochemical web of our brains evolves over time, as it obviously does.


I'm tempted to just erase this. But I feel among friends here, and so I've decided that it doesn't matter if this sounds foolish. And if there are others here to whom this is not completely non-sensical, I'd enjoy hearing from you, and especially hearing about the "core bibliographies" relevant to this stuff that you are currently working with.







On Tuesday, September 15, 2015 12:27 AM, "Obrst, Leo J." <lobrst@xxxxxxxxx> wrote:





-----Original Message-----

From: Christian Hempelmann [mailto:C.Hempelmann@xxxxxxxxx]


FLAIRS Knowledge-Based AI Track


Special Track at FLAIRS-29, Key Largo, Florida USA


In cooperation with the Association for the Advancement of Artificial



Key Largo, Florida, USA

May 16 - 18, 2016


Paper submission deadline: November 16, 2015.

Notifications: January 18, 2016.


Camera-ready version due: February 22, 2016.


All accepted papers will be published as FLAIRS proceedings by the AAAI.


Call for Papers


What is Knowledge-Based AI?

After an early dominance in AI, especially in NLP, approaches based on

engineered knowledge-resources such as rule bases and ontologies modeling

(part of) the world, the statistical winter set in in the 1980s. Fueled by

increased computing power, statistics-based replacements for modeling the

world, including machine-learning and neural networks, lead to early

successes before they hit their ceiling and resulted in algorithmic arms

races. To get AI out of these trenches and back into mobile warfare,

knowledge-based methods have not only been being paid  lipservice to in

the many "semantic" revolutions, but actual applications have been built,

often with complementary methodologies that paired statistical and

knowledge-based solutions.

The scope of the track includes research, proof-of-concept and industry

applications in the area of knowledge-based AI, i.e. systems whose

functionality is informed by computational knowledge resources

(ontologies, lexicons, semantic networks and/or knowledge bases). While

the knowledge-based AI is often juxtaposed to the statistics-based AI, we

see the contrast as unnecessarily exclusionary, in that systems combining

the intuitive directness of knowledge representation with the efficiency

of statistics-based computation have distinct advantages.


What is the GOAL of the track?

To showcase recent knowledge-based theories, methodologies, and

applications in AI and to foster new approaches of this kind, also those

paired with statistical and machine-learning approaches.


Who might be interested?

The scope of the track includes research, proof-of-concept and industry

applications in the area of knowledge-based AI, i.e. systems whose

functionality is informed by computational knowledge resources

(ontologies, lexicons, semantic networks and/or knowledge bases). While

the knowledge-based AI is often juxtaposed to the statistics-based AI, we

see the contrast as unnecessarily exclusionary, in that systems combining

the intuitive directness of knowledge representation with the efficiency

of statistical approximation have distinct advantages.

What kind of studies will be of interest?

Papers and contributions are encouraged for any work relating to

Knowledge-Based AI. Topics of interest may include (but are in no way

limited to):

              € ontologies

              € spreading activation networks

              € lexicon acquisition

              € knowledge integration

              € applications in knowledge-based AI

              € hybrid probabilistic/machine-learning & knowledge-based systems

Note: We invite original papers (i.e. work not previously submitted, in

submission, or to be submitted to another conference during the reviewing



Submission Guidelines

Interested authors should format their papers according to AAAI formatting

guidelines. The papers should be original work (i.e., not submitted, in

submission, or submitted to another conference while in review). Papers

should not exceed 6 pages (4 pages for a poster) and are due by November

16, 2015. For FLAIRS-29, the 2016 conference, the reviewing is a double

blind process. Fake author names and affiliations must be used on

submitted papers to provide double-blind reviewing. Papers must be

submittcoued as PDF through the EasyChair conference system, which can be

accessed through the main conference web site

(http://www.flairs-29.info/). Note: do not use a fake name for your

EasyChair login - your EasyChair account information is hidden from

reviewers. Authors should indicate the [your track name] special track for

submissions. The proceedings of FLAIRS will be published by the AAAI.

Authors of accepted papers will be required to sign a form transferring

copyright of their contribution to AAAI. FLAIRS requires that there be at

least one full author registration per paper.

Please, check the website http://www.flairs-29.info/ for further



Conference Proceedings

Papers will be refereed and all accepted papers will appear in the

conference proceedings, which will be published by AAAI Press.


Organizing Committee

              € Christian F. Hempelmann, Texas A&M University-Commerce,

              € Gavin Matthews, NTENT.com,

              € Max Petrenko, NTENT.com,


Program Committee

              € Christian F. Hempelmann, Texas A&M University-Commerce,

              € Elena Kozerenko, Russian Academy of Sciences,

              € Gavin Matthews, NTENT.com,

              € Leo Obrst, MITRE,

              € Max Petrenko, NTENT.com,

              € Victor Raskin, Purdue University,

              € Julia M. Taylor, Purdue University,

              € Tony Veale, University College Dublin,

              € Yorick Wilks, University of Sheffield & IHMC, Florida,

              € Michael Witbrock, VP for Research, Cycorp.


Further Information

Questions regarding the Knowledge-Based AI Special Track should be

addressed to the track co-chairs:

              € Christian F. Hempelmann, Texas A&M University-Commerce,

              € Gavin Matthews, NTENT.com, gmatthews@xxxxxxxxx

              € Max Petrenko, NTENT.com, mpetrenko@xxxxxxxxx


Questions regarding Special Tracks should be addressed to Zdravko Markov,

Conference Chair: William (Bill) Eberle, Tennessee Technological

University, USA (WEberle@xxxxxxxxxx)

Program Co-Chairs: Zdravko Markov, Central Connecticut State University,

Ingrid Russell, University of Hartford, USA (irussell@xxxxxxxxxxxx)

Special Tracks Coordinator: Vasile Rus, The University of Memphis, USA


Conference Web Sites

Paper submission site: follow the link for submissions at

FLAIRS-29 conference web page: http://www.flairs-29.info/

Florida AI Research Society (FLAIRS): http://www.flairs.com


Christian F. Hempelmann, PhD | Assistant Professor of Computational


Department of Literature and Languages

Texas A&M University-Commerce

P.O. Box 3011 | Commerce, TX 75429-3011

Tel. 903.468.5291 | Fax: 903.886.5980 | www.tamuc.edu

The Texas A&M University System



Dr. Leo Obrst                The MITRE Corporation, Information Semantics

lobrst@xxxxxxxxx        Cognitive Science & Artificial Intelligence, CCG

Voice: 703-983-6770  7515 Colshire Drive, M/S H317

Fax: 703-983-1379      McLean, VA 22102-7508, USA



Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J




Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (01)

<Prev in Thread] Current Thread [Next in Thread>