Parsing
From GusWiki
Contents |
The general idea of parsing is making sense of your signals (inferring how they were generated). As such, it is closely related to Gestalt perception.
Parsing can be applied to several modalities:
natural language
NLP
images
See Scha et al for Data-Oriented Parsing
smells and tastes
In natural language, once the speech is realized, the parse trees (which, in some sense, exist in the brain of the speaker) are flattened into a 1D line, a temporal order of words (or phonemes or sounds, depending on the level of analysis). In vision, real-world scenes get flattened into a 2D image. In both cases, many ambiguities can arise from the loss of information that comes with this flattening.
The process of interpretation, which is essentially the process of reconstructing the original structure, isn't always simple, because of complex dependencies and ambiguities. -:Psycholinguists are very fond of the tricky cases (such as -:garden path sentences), and like to do surprisal experiments that involve -:priming subjects with certain structures.
We can also apply the parsing metaphor to other senses: "parsing" really just means making sense of your signals. Now, let's consider some less tangible senses: recipes have hierarchical structure: tasters will immediately perceive "bechamel sauce", even though it's made of more basic components such as "butter" and "flour" (a fact which they might not be aware of, even if these tastes are familiar).
But when parsing smells or homogenous fluids (such as well-blended drinks), we have even less structure! We don't even have a temporal order of the ingredients. Nevertheless, it is in theory possible to reliably detect commonly-occurring chunks in such media... although the lack of structure makes a lot of room for ambiguity. (For example, if two of our components can take sugar, it's impossible to distinguish the case in which the first contained 10g of sugar and the second none, from the case in which the first had 7g and the second 3g; however, plausibility judgements can still be made: "who's ever heard of honey without sugar??").
(Data-Oriented Parsing should fit very well with Bayesian CogSci. I'm wondering why I haven't heard anything about it.)
Let's create an empirical science of psychogastronomics!
(Thanks to mconst for stimulating discussions)
