Semantics

Semantics is the study of meaning, in particular the meanings of linguistic expressions, but because of the various approaches to studying semantics, the term itself is somewhat ambiguous. Depending on the research background of someone using the term, at the very least, it may take on different meanings. I find it useful to recognize at least three main senses of the term semantics:

  1. Formal Semantics. In logic, semantics refers to the truth-conditional propositional meaning of linguistic expressions. Truth conditions are the conditions that would have to obtain in the world (or in a set-theoretic model of the world) in order for the expression to be "true".
  2. Information Retrieval Semantics. In Information Retrieval, the chief task is one of accepting query, i.e. a representation of what a user wants to find (usually one or two words), and retrieving from a store of information documents that are deemed to be relevant in some way. Information Retrieval specialists talk about semantics in terms of the relatedness of two documents to one another, or the relation of a query to a document.
  3. Linguistic Semantics. In linguistics, semantics concerns the meanings of unrestricted linguistic expressions, the different meanings they take in different situational and social contexts, their historical development, their relation to other meaningful expressions, processes of inference, etc. In this sense, meaning is regarded as a highly complex object of study that is not readily reduced in one direction or another.

Without going into the philosophically-fraught semantics of "true", the logical notion is intuitive and straightforward on the surface. However, there are some subtle issues with the notion that make it difficult to work with in certain circumstances. For example, many linguistic expressions are intensional in nature, in that they represent non-existent states of affairs, or situations obtaining in someone's belief but not in the actual shared world we experience. Anchoring these expressions to some notion of logical meaning requires extra complications in the set-theoretic model theory that is used. There is not necessarily general agreement on the best way to do this.

Another characteristic of approaches to semantics from formal logic is that they often treat meaning as a literal object, i.e. some kind of bona fide thing unto itself. This is largely due to the Platonistic commitments of many people working in the field of formal logic, but such commitments can become awkward in different situations, for example when semantic change is considered, or when entirely new meanings arise that were not in existence before. While these things have their standard interpretations in a Platonistic framework, those interpretations are not necessarily helpful to those wishing to understand such phenomena more deeply.

Information Retrieval semantics has similar issues. On the one hand, a more or less vague notion of relatedness is often useful to work with when considering semantics, especially at the document level. Being able to handle relatedness in terms of degree allows one to use thresholding or ordering (ranking) strategies to present information to users. However, some extremely important relations from a linguistic perspective, such as negation, are difficult to represent in such a system, and Information Retrieval applications often make no attempt to represent such relations at all. Hence, relatedness (or similarly, association) can only offer a partial picture of linguistic semantics.

Semantics, therefore, needs a richer aparatus than is provided by IR for the representation of linguistic meanings. At the same time, it should not necessarily be encumbered by Platonistic philosophical commitments, and it should provide a rich vocabulary for describing meanings, realtions of meanings to linguistic expressions, and relations of meanings to other meanings and entire systems of meaning. Linguisitc semantics is the field which takes this approach, and there are many approaches (formal, cognitive, social, historical, etc.) within the field, but all linguistic approaches share a recognition of a wide range of phenomena and the vocabulary describing them. Some of these concepts are especially useful to technological contexts and the notion of emergent semantics.

  • Classification and taxonomy
  • Hypernymy and hyponymy
  • Metaphor and metonymy
  • Predication
  • Implicature and inference

Tagging and "Folksonomies"

Tagging is a method of organization that has now become widespread on web-based applications. When an application supports tagging, it basically supports open-ended user-applied metadata to be attached to resources. The metadata are presented in various summaries, and search interfaces are implemented on top of them.

On a technical level, tagging is very easy for system designers and administrators to implement as it uses only technlogies that are likely to be already in place for other purposes. Enthusiasts appreciate tagging for its ease of use and its transparent extensibility. Instead of being confined to rigidly-defined taxonomies, users may interactively create new organization strategies. The term folksonomy, shortened from "folk taxonomy" has become a widespread way of referring to such organization schemes.

Emergence

Contained within the notion of folksonomy is the notion of emergence, where emergent behavior is a manifestation of complexity that inheres in a set of actors all using simple rules while interacting with one another. The analogy with tagging is that actors following simple rules ("apply the tags that are meaningful to you") interacting in a collaborative environment (such as a social bookmarking service, wiki or photo website) results in a global organizational scheme that is meaningful to all members of the group. Hence semantic emergence is conceived as the self-organization of classificatory knowledge in a collaborative information environment.

In my research on tagging, I use statistical and linguistic methods to observe coherence in structural patterns among collections of tagged resources in user communities, in order to try to recognize semantic emergence. Our results sho that even for classificatory knowledge, this picture is somewhat simplistic. Classifications often make use of negation, predication, metaphor, implicature and other linguistically complex strategies, and this is even more true when informal, open-ended and individualistic classifications are used. Hence the linguistic semantics of tags is more complex than the mere application of labels. Moreover, on the social level, there is not necessarily a strong relationship between social behaviors and the semantic dimensions of tagging, as would be expected under notions of semantic emergence.

In order to understand the processes of semantic emergence better, we need a perspective that permits us to observe social processes through which meaning is contested and negotiated. Sometimes people's meanings for terms will change under the pressure of contact, other times people resist and different conflicting meanings for the same terms will reside among different groups of people. At times such terms or meanings can serve as ideological markers for a person's social position or orientation toward competing groups or the larger social environment. I am actively pursuing research on this topic at present.


Categories: Semantics, Emergence, Dynamics