Homework Coding Assignment

Due October 18 at 11:59 P.M.
1 point late penalty for each minute late

The objective of this homework is to give you some practice with actual language data in preparation of selecting a project topic for the class. I have annotated a small conversation transcript to indicate a set of pronouns that we want to know more about. Your assignment is to add information about the meaning of the pronouns and think about what information would be needed in a computational model of context to provide those meanings for the pronouns.

In this assignment, your data will be a portion of a transcript of two partners working together on a visual world task, accompanied by the video record of the task. The pronouns of interest in the transcript and their form have already been indicated by me.

For each noun phrase, you need to add annotation for:

  1. The referent or meaning of the pronoun. If the referent is a physical item in the task world, please use the map of the game world for item names.
  2. The mutual knowledge category of the referent. Do the two participants have the referent in their shared knowledge due to linguistic mention, physical copresence, etc.
You can use the video of the session to keep track of what the dialog participants know at each point in the dialog.
  1. A Movie of the game interaction from pogo-cat's perspective. Pogo-cat's utterances are labeled 'leader' in the transcript.
  2. Movie of the game interaction from penguin's perspective. Penguin's utterances are labeled 'follower' in the transcript.

Installing and using mmax

The MMAX annotation tool can run on any platform where java is installed.
  1. Download this file and untar it for the necessary class files.
  2. Make a different directory for your working files and download this set of annotation files there. These files must all be kept together in one directory.
  3. Copy or modify one of the 'startmmax' scripts from the main mmax directory to point to the 'quake.anno' file in your annotation directory.
  4. Run the start_script. This will bring up a java text pane with the transcript.
  5. To annotate this file, click on each highlighted pronoun and fill out the fields in the annotation window.

What to submit

Email me your markables.xml output from the tool. Think about the following questions, which we will discuss in class:
  1. Which referents would a text-only context model (such as BFP) be able to provide for pronoun resolution.
  2. What are all the 'triggers' that seem to make referents salient enough to be mentioned as pronouns.
  3. What additional attributes would you like to analyze on these pronouns.
You can discuss the assignment with other students in the class, but I would like each student to turn in his/her own work.
donna byron
Last modified: Tue Nov 9 10:11:29 EST 2004