home

research

publications

demo

INSOMNet Demo

The screenshot below shows the network in the middle of the second of three demos. The large window in the background is the main window that appears when "(mac/linux)-mir scripts/vmdemo.tcl" is invoked. At first, the "DEMO" button will be red like the "SOM" and "MOD" buttons in the screenshot. Click on the DEMO button to start the demo. Doing so will bring up the "Demo Remote Control Panel". Then you only have to click "NEXT" to go through the three demos. It is assumed that the user has already read or at least skimmed the article, "Incremental Nonmonotonic Sentence Interpretation through Semantic Self-Organization" submitted to Cognitive Science. The vmdemo.tgz is available to supplement the article.

The first demo reviews the main components of the INSOMNet architecture, such as the SRN at the top of the main window, the grid in the middle called the Frame Map, and the decoder, the frames of which are shown at the very bottom. The demo also introduces how frame activation levels are represented, which determine what frames are considered part of the sentence interpretation.

The second demo describes how a symbolic representation of the network activity can be ascertained based on the activation levels of the frames that are above a given threshold.

The third demo delves into how semantic self-organization allows a pointer-based system such as INSOMNet to reliably recover the proper thematic case roles of words and internal nodes in the MRS dependency graph together with their fillers.

Once the demo is finished, the user will be free to try the system out by examining the rest of the sentences in the "demoset" corpus or loading the "fulltrain" or "fulltest" corpus on which this particular instance of the network was trained. There is also a "newset" in which novel sentences made up by the user can be tried out, using the vocabulary in "tkLex" under "vmdemo/experiments/vmmrs".

    A brief explanation of each button on the window is given, followed by operations that can be performed with the mouse
  1. STEP: step through the sentence one word at a time.
  2. BACK: step back to a previous state of the sentence (does not work in demo mode, however).
  3. CLEAR: clears all activation in the network and restarts with the first sentence.
  4. ZAP: destroys all pop-up windows.
  5. RAISE: raises all pop-up windows in their original stacking order.
  6. DEMO: toggles up the demo remote control panel.
  7. INT: toggles symbolic interpretation text box.
  8. MOD: toggles the activation modulator.
  9. SOM: toggles the underlying self-organized map of semantic representations.
  10. QUIT: exits the program.
  11. Grammar entry box: replace "demoset" with another grammar dataset.
  12. Threshold: The activation level a frame must reach to be considered a part of the current interpretation of the sentence.
  13. Moving the mouse over the Frame Map shows the frame number and its current activation.
  14. Button one on a node in the frame map will decode that frame into its constituent parts.
  15. Button two on a handle field will show the node in the frame closest to the handle.

    The demo is designed to show much of this functionality.

    missing