Language and Representation

Language and Representation

In “The Paradox of the Universal Triangle,” Gary Lupan detailed a series of experiments he performed in order to examine the relationship between language (specifically, categorical labels) and abstract concepts. The triangle paradox is an age old paradox about the problem of abstraction and generalization: given that we’ve only ever had experience with specific entities, how do we abstract from those specific instances to come up with a general category? Contrary to the traditional neuroscience view that we first come up with a general concept, then associate a linguistic label to it, Lupan demonstrated that these so-called general concepts are less abstract than we think. By performing experiments that contrasted people’s production, recognition, judgement, and inference of shapes based on whether they were cued as “three sided polygons” or “triangles,” Lupan realized that the label “triangle” was most often associated with a ‘prototypical’ triangle (equilateral, horizontal base), and not purely abstract. In conclusion, Lupan argued that abstract concepts are more like flexible prototypes, and furthermore, that language and its ability to make these signs for abstract concepts, is key to their creation.

In contrast to Lupan’s detailed explication of a sociolinguistic concept, John Russell Rickford’s article, “Unequal Partnership: Sociolinguistics and the African American Speech Community” addressed sociolinguistics from a wider view by discussion its relationship to the communities that it draws from. In contrast to the huge role AAVE (African American Vernacular English) played in the development of sociolinguistics, Russell argued that the field of sociolinguistics has not reciprocated this help adequately by nurturing African American linguists, educating underprivileged communities, or ‘giving back’ to these communities. One connection between these two readings that stood out to me was Russell’s description of the “Bridge” project which found that African American school children have better recall and reading comprehension for stories presented in AAVE as well as in SE. Previously, Lupan had argued that language isn’t about applying labels to concepts, but about cuing meaning. I thought that maybe reading SE only was a bit like getting cued with “three sided polygon” only: understandable, but not as fast or as intuitive. Perhaps cross-dialect course readers were more effective than only SE readers because it enabled faster recognition of these prototypes.

Coincidentally, last week in SymSys lecture we also talked about the difficulty of getting machines to abstract from specific examples to general concepts. Dr. Lassiter mentioned that although the current approach (feeding a lot of data into a machine and identifying these patterns) worked, it did wasn’t as effective or as fast compared to human children. I wonder if implanting this prototypical structure (and rules for its manipulation) in computers would lead to more human-like results than current machine learning practices? To make this approach more scalable, I’m also curious about how we as humans learn and create these prototypes initially as well.