Professor Victor Raskin
Raskin at Purdue
Raskin at hakia.com
It is a formal and comprehensive linguistic theory of meaning in natural language. As such, it bears significantly on philosophy of language, mathematical logic, and cognitive science. It is a rapidly growing area of intense academic research and of active practically implemented applications.
Ontological Semantics offers an advanced methodology and technology for natural language processing, the only one of its kind, so far, to access the full meaning of the text it handles. As such, it is also a set of well-developed and constantly improving resources, including a language-independent ontology of thousands of interrelated concepts; an ontology-based English lexicon of over 100,000 word senses, and counting (plus, the lexicons for several other languages under construction); and an ontological parser which "translates" every sentence of the text into its text meaning representation, approximating the complete understanding of the sentence by the native speaker.
An important part of the approach is the well-developed and constantly improving acquisition toolbox which ensures the homogeneity of the ontological concepts and lexical entries by different acquirers of limited training and the proven portability and extensibility of the resources to new domains. Constantly developing, Ontological Semantics is actively researching meaning-based inferencing and reasoning as well as the closely related perennial issues of linguistic semantics, viz., ambiguity, focus, and ellipsis.
Different Usages of the Term Semantics
There are three major disciplines using the label "semantics," at least occasionally, namely, the assignment of symbol interpretation to formal languages in mathematical logic, the philosophy of language in philosophy, and the study of meaning in natural language in linguistics.
Logical Semantics
Let us introduce a very simple formal language L. It will use three variables, x, y, and z; one one-place relation, m; two two-place relations, M and P; and the standard logical operation of negation, ¬. We define the syntax of L by the following rules of term formation:
mvariable, e.g., mx;
variableMvariable, e.g., xMy;
¬term, e.g., ¬xPy.
If we now interpret the variables as people, m as male, M as married, and P as parent, the examples above will mean, respectively:
x is male;
x is married to y; and
x is not a parent of y.
L now has semantics. Of sorts.
Philosophical Semantics
This discipline is interested in the truth values of the propositions expressed by sentences, e.g., the sentence Connecticut is a USA state expresses a proposition that is true (T) while the sentence Canada is a USA state expresses a proposition that is false (F). It shares with logical semantics an interest in truth preservation and formulates such rules as T & T = T, i.e., if two propositions are true then their conjunction is also true.
Linguistic Semantics
This discipline, of which hakia.com's ontological semantics is the most advanced school, studies the meaning of sentences and texts as they are understood intuitively by native speakers. Because native speakers have internalized large lexicons, based presumably on a large ontology, as well as the rules combining word meanings into sentence meanings, making inferences, etc., ontological semantics has committed a major effort to the acquisition of such resources and discovery of such rules. This approach is called representational. Most linguistic semanticists do not have the know-how or resources to practice this approach and, intimidated by the computer scientists and engineers dominating computational/NLP semantics, attempt to take a short cut into the non-representational approach by replacing the resources and the rules with logical or statistical methods to linguistic meaning. The currently still dominant "formal semantics," a combination of logical and philosophical semantics from above, replaces the meaning of the sentence as the goal of its study with the truth value, thus severely limiting the scope of linguistic meaning to what can be easily logicized. Thus, they can handle the meaning of every in every chair by applying the universal quantifier
("all") to it but are incapable
of accounting for the meaning of chair. Similarly, they
easily represent John loves Mary with love (John, Mary) and
John hates Mary with hate (John, Mary) but cannot access the
meanings of love or hate. What the formal
semanticists cannot account for, which is most of linguistic
semantics, they define out of semantics and out of the scope of
formal methods.Unfortunately, the formal semanticists have robbed a whole generation of young linguists of the descriptive and analytical skills for understanding and studying meaning, thus crippling the semantic labor force for applications. The statisticians have had the same crippling effect by propagating the tagging methods of, for instance, disambiguation, in which a group of native speakers chooses one of the two possible meanings of pre-selected ambiguous words on part of a corpus (the training corpus), and then a bunch of statistical measures try to emulate this result on the rest of the corpus (the test corpus). Even with such doctored and limited data, very different from real life, the accuracy of the methods falls far short of human usersā acceptance level. Neither the logical nor the statistical methods ask or answer any questions about the nature of meaning nor contribute to its understanding: they have no resources and object to rules. The latter also mislead as to the capability and reliability of machine learning.
Chart of the Field
The chart below attempts to capture the main groups (in the columns) making semantic claims and to describe them in terms of a selection of relevant features (in the rows).
Table 1: Meaning/Semantics
[1] Nirenburg's and Raskin's branches of ontological semantics have emerged from the common platform developed over 20 years of joint or coordinated research culminating at the Computing Research Laboratory, New Mexico State University, in 1994-2000. Since then, Nirenburg has continued at his newly formed (2002-) Institute for Language and Information Technology the University of Maryland Baltimore County, while Raskin has continued at his Natural Language Processing Lab (1986-) and expanding to the Center for Education and Research in Information Assuarance and Security (1999-) at Purdue University. The difference between the two branches is captured very crudely in the chart above and is not intended as a judgment or criticism of either.
It may be useful here to provide a similar chart for the ontological involvement by various groups:
Table 2: Ontology
[1] Note that there is no separate column for searches in Table 2. Most searches adopt the government/military or genome-type positions or and use flat and partial taxonomies in the name of ontologies, most of them with just the Is-A property or one or more, such as part-of.
Presenting Ontological Semantics Outside
While ontological semantics allies itself with computational semanticists in their common emphasis on meaning and with ontologists in their common commitment to a language-independent conceptual basis for meaning, it differs from others in its uncompromising commitment to a comprehensive, practical, systematic, principled, theory-based implementation. It is essential to realize that ontological semantics is not "just doing it differently" but that it is unique in doing what needs to be done for a full and deep meaning representation and text understanding, rather than what can be done easily, and what nobody else seems to be in a position to do. The advantages of ontological semantics lie in its:
The standard smart response by the opponents of ontological and generally representational semantics to what we are doing is that it may well be everybody's dream but it is not achievable while what they are doing is practical and "now." The best and most prominent practitioner of each and every new buzzword-based approach to avoid meaning, Ed Hovy, expressed it best, with his virtuoso eloquence, at the 2nd Text and Meaning Interpretation Workshop at the 2004 ACL in Barcelona, when he responded to Nirenburg's objection to the approach du jour by saying that ontological semantics "aspires to fly while [they want to crawl]." The best response to that is a demonstration of the resources and results of ontological semantics on the OntoSem lab pages, a demonstration which will constandly change and improve with the technology.
Alternatively, ontological semantics can compare its opponents' position to that of the old Jew in the first pre-Holocaust joke below and their choice of methods to the drunk's methodology in the second.
Joke 1. An old Jew from a small East European village is taken to a big city zoo for the first time in his life. He looks at the giraffe and says, "There cannot be such a long neck!"
Joke 2. A drunk is looking for something under the street lamp. A passer-by, willing to help, asks him what it is that he lost. "My keys," answers the drunk. "You dropped them right here?" "No, over there." "So why are you looking for them here?" "Because it is dark over there."
Interactive Demo
hakia labs
Related Reading
Ontological Semantics
(Pre-publication working draft.
S. Nirenburg and V. Raskin, MIT Press, 2004)
Preface
Part I:
1- Introduction
2- Prolegomena
3- Study of Meaning
4- Lexical Semantics
5- Formal Ontology
Part II:
6- Meaning Representation
7- Knowledge Resources
8- Basic Processing
9- Resource Acquisition
Bibliography
Book on Amazon.com
Ontological Support for Domains (NSPW 2001)
[NCOR]: A rich medical ontology effort for human use
Ontologies for USG-1
Ontologies for USG-2
Semantic Forensics