Computational modeling of semantics
During my stay in Colorado, I worked with profs. Kintsch and Landauer on Latent semantic Analysis (LSA), developing technologies to assist intelligence agents to process large amounts of textual information, to automatically grade essays, to assess and promote vocabulary development and to model how users interact with websites. I also worked with Simon Dennis on his syntacmatic-paradigmatic (SP) model, that tries to learn both syntax and semantics from natural corpora.
Problem solving
One paradox in the current literature of problem solving is the origin of problem spaces. If all intelligent behavior takes place in a problem space, and the creation of a problem space is intelligent behavior, then we are trapped in a recursive definition. In my work on problem solving a possible solution is presented: that problem spaces can be created in a bottom-up way automatically from experience by a mechanism that may be shared with language learning. This is a corpus based approach to representation in problem solving. I have tested this theory in different complex dynamic tasks. Applications of this proposal to real-word complex, dynamic tasks, such as flying simulators have been shown to be possible.
Judgment and decision making
I’m working with Neil Stewart, Nick Chater and Gordon Brown testing and extending their Decision by sampling theory, which accounts for key phenomena in the domains of decision under risk, trade off between attributes, and temporal discounting.
Analogy
Most current work on analogy uses propositional representations. I’m interested in how to derive representations automatically from text that can be used in high-level cognition processes such as analogy. Our current model (in preparation) can solve SAT analogy questions at around the same level as the average human. With and Walter Kintsch
Contextual Diversity
‘No single variable has been studied more in psycholinguistics and memory research than word frequency’ Balota et al (2001). However, according to the results in our lab, it seems that it is contextual diversity, not word frequency what drives several effects such as naming, lexical decision, and surprisingly enough even human frequency judgments. With James Adelman, Nick Chater and Gordon Brown.
Mental number line
The dual-representation theory (e.g., Siegler and Opher, 2003) proposes that children have both a logarithmic and linear mental number representation, and that they transition from log to linear as they grow. In my work with professor Jane Oakhill, we present an alternative theory where the mental number representation is always linear, but at early ages the line is misaligned with y=x (perfect number perception). With age, children’s slope and intercept get closer to 1 and 0 respectively.
