The "modeling language" team works on the Semantic Atlas (SA) paradigm. The visualization above is one of the first drafts of semantic change modeling within the team’s ongoing work. It is based on the Automatic Contexonym Organizing Model (ACOM) and its extension that deal with co-occurrent words in text. (on ACOM: Ji, H. et.al. 2003. Lexical knowledge with contexonyms. In Proceedings of the 9th MT summit, pp. 194-201.)
On the basis of the geometrical models of the SA and ACOM, Armelle Boussidan’s PhD research focuses on the dynamic approach of meaning representation. Her work is based on corpus linguistics methods as well as diachronic analysis in the tradition of historical linguistics.
Since dynamics of change are intrinsically unforeseeable, we focus on studying the main mechanisms of semantic change on the lexical plane and work on the methods to detect and analyse it.
For now, the team works on a test corpus in order to design a flexible prototype .
We aim at making the dynamic model both a representation paradigm (model) as well as a tool for text and meaning analysis at the scientific community’s disposal.
The chosen diachronic approach is peculiar in that it focuses on very short time periods so as to extract quick and subtle mechanisms of change within large size corpora.
This approach could be compared to a dissection of meaning or to a zoom into meaning composition at a microscopic level, within a continuous paradigm which contains several levels of structure and analysis (cliques, words, clusters, maps…).
All types of semantic changes are analyzed, bringing connotation, innovation, obsolescence, metonymy and all other forms and causes of semantic change - traditionally listed by linguistic typologies - under the same umbrella. The method encompasses a great variety of phenomena and offers very fine grained analysis regardless of their typological status.
In this framework, Charlotte Franco developed programs to visualise maps on the basis of temporal sub-sections of corpora via a java applet. The applet calls a program in C that has 4 space diffusion parameters. It generates the contexonym database and the cliques’ coordinates for each period of time. The parameters are used to calculate the integration threshold of contexonyms.
The java class creates a cloud of coordinates (cliques) with its axes. We can now choose the time frame we wish to look at depending on the object of research. We then obtain series of maps that show the evolution of associated contexts in time. On the basis of these maps a dynamic device using interpolation allows for the visualization of the transition between the different stages.
The geometrical model raises additional research questions. Meanings are represented by shapes generated by correspondence factor analysis in a multidimensional space and are then structured into clusters by a hierarchy algorithm. What is the relationship between the obtained shape and the corresponding meaning? How does the transition from a shape to another mirroring the transition from a meaning to another operate? Anne-Lyse Renon offers a graphical reflection about the maps, based on the structural properties of the generated shapes, for the representation of meaning evolution.
Using the surface and outline properties of the maps’ shapes, the interpolation allows for dynamic visualization of semantic shifts through graphical shifts. It also brings out a questioning of the space-time continuum’s status in the process of the animation.
The visualization above is a representation of the semantic evolution of the French word "mondialisation" (globalization) in a 1997-2001 press corpus (the newspaper « Le Monde »). We have deliberately selected a few words brought into play ("défi" challenge, "menace" threat, "progrès" progress and "alternatif" alternative).
We can see the transition from a relatively generic meaning to more specific meanings, partly due to the creation of neologies such as « antimondialisation » and « altermondialisation » (two alternative words for antiglobalization with subtle differences) in this period of time.
We can also see the evolution of the global envelope of the word as well as the density within it. Clusters also evolve across time as they sometimes gather notions that appear as opposed later on (see "menace" and "progrès"‘s behaviour).
For more details as to this example, the reader can refer to the article (Boussidan, A., Renon, A.-L., Franco, C., Lupone, S., and Ploux, S. Vers une méthode de visualisation graphique dynamique de la diachronie des néologies. Actes du colloque international Néologie sémantique et corpus, Tübingen, Germany, to be published.)
This first dynamic representation will be followed by further work and the research team would receive any commentaries and criticisms
with interest. We think we could put to profit this new dynamic model in many ways among which metaphor detection where metaphor is considered as a triggering process to semantic change.