Gerard Kempen's Website

Theo Vosse & Gerard Kempen started the Unification Space parser project in 1987 at the NICI (now part of the Donders Centre of Radboud University) in Nijmegen, and continued until 2009, with interludes of several years. They published three versions of the parser: in 1989, in 2000, and in 2009. The 1989 version was developed in Nijmegen, the 2000 version at the Cognitive Psychology Unit of Leiden University, and the 2009 version at the Donders Centre and the Max Planck Institute in Nijmegen. Each publication was accompanied by a demo movie showing parser performance for a small set of English sample sentences that are often discussed in the literature.

The demo of the 1989 version was an 8-minute videotape recorded with a VCR off the screen of a Mac computer in June 1989. Here is a digitized version of that videotape (8 minutes). The yellow text at the top of the screenshots is "Unification Space"; the graph at the bottom-right includes the term "Excitation" in blue (low excitation or red (high excitation). The "blue" demos are supposed to simulate parsing by normal language users; the "red" demos attempt to simulate parsing by agrammatic (aphasic) language users. The videotape and the digitized file have not been edited, except for two places that are indicated explicitly. For further explanations, see the 1989 paper.

The demos for the 2000 version, which were prepared already in 1997-98, took the form of Quicktime movies on Mac computers. Here, a subset of them that has survived (all in English), is combined into a 4-minute movie, interleaved with some explanatory printed texts (4 minutes). Notice the color coding for the decaying activation levels of syntactic nodes (gradually changing from red to blue) and for the variable strength of attachment/unification links (varying over time between weak/yellow to strong/green). In order to identify the shape of the emerging tree (or tree fragments) one should follow the darkest green lines and the nodes connected by them. The top-left corner is supposed to be a "peg" on which the tree-under-construction is hanging.

The 2009 demo (5 minutes) is a small collection of runs of English sentences embodying syntactic structures often discussed in the psycholinguistic parsing literature. Every window shows the layers of neural-network nodes that constitute the Unification Space, as described in the paper. Each black or blue square or dot in a layer represents a (morpho)syntactic object (node, feature, linear order position), and the lines connecting them represent the flow of feedforward activation (dark green), feedback activation (yellow), or inhibition (red). The activation levels of functional and phrasal nodes (in the Unification layer) is coded in terms of the font size of the node label. Only connections whose activation or inhibition level exceeds a threshold value, are shown. The lower half of the window shows (part of) the unfolding activation and inhibition processes; the upper half shows the syntactic tree (or tree fragments) extracted from the the unfolding activation pattern.

Human Grammatical Coding: "Unification Space" parser demo movies

The demonstration movies can be accessed here (YouTube).