STANFORD CS 224S/LINGUIST 281   -     Winter 2009
Final Project Instructions
Due: Monday March 16 at noon PST

I recommend the final project be done in groups. Groups may not have more than 3 members. The final project has two components, a poster and a paper. The poster session will be held in the Gates Lobby on Thursday, March 12, 3:15--4:30. Poster boards and easels will be available in Nikkie Salgado's office, Room 187 in Gates.

The poster

  • A poster is a "personal" oral presentation of your results. We will supply a posterboard and easel, and you can either print out a single big poster (our poster boards are 18 x 29.5 inches) or tack up separate sheets of paper.
  • Q: What should go on a poster?
  • A: A poster should tell a story; someone should be able to read it without you saying much
  • A: But a poster should not have too much text; just enough text around each of your cute graphs or screenshots for the reader to understand what the point of the graph was
  • A: A poster can also be a backdrop for a demo, if you would like to demo your system on a laptop!

    If you want to do a single large poster sheet (instead of several pieces of paper) and you have a Stanford ID, you can find relatively cheap wide-format printing at the Technology Services Desk on the 2nd floor of Meyer Library (hours, location); they charge a flat rate of $45/print for up to 4' x 6'. (Compare to $7/square foot at Kinko's). Slightly more information on this service is available here.

    The paper

    The paper should be 4 pages long, in "INTERSPEECH" submission format. INTERSPEECH format is described here:

    http://www.interspeech2007.org/papers/?f=authors.html

    INTERSPEECH format is very short (4 pages isn't much), so if you want to include other information (details about your data, or other interesting facts, or a really beautiful figure that just doesn't fit) you may add a short appendix of a few pages more. But I don't recommend it. Being able to get your idea and result across briefly and clearly is crucial (in life, not just in this class).

    Please send the paper as a PDF file to both me and to cs224s-win0809-ta@lists.stanford.edu by noon PST on Monday, March 16, 2007.

    What to put in a final project paper:

    Research papers:

    These are papers where you attempted some new research idea. This doesn't have to be publishable research; as we've discussed, it's totally great to do a replication of a result you read about. Such papers should contain clear sections describing:

  • The problem you are addressing
  • Your "hypothesis" or proposed solution (and if you are implementing someone else's solution, where you got the idea from)
  • Alternative solutions, or at least a "baseline" that you are comparing your solution to.
  • Your methdology that you used to test your hypothesis.
  • Your evaluation, i.e., the experimental results
  • A discussion of what your results imply for your hypothesis/problem.

    Implementation papers

    These are papers where you train a recognizer on some new data (building a digit recognizer in language X or a TTS system in your voice), or code up some version of someone else's algorithm just to learn the details of the algorithm. Here your want clear sections describing

  • The recognition task that you are building or the algorithm you are implementing.
  • Your methodology (What you did, how you did it).
  • The evaluation, i.e. the experimental results: did your implementation work?
  • A discussion of what you learned.

    Some previous year final projects

  • Winter 2006
  • Winter 2007

    Some publishable project ideas for this quarter

  • Phone classification: replicate machine learning algorithms (max margin HMM, MCE, MMIE, etc) and apply to the task of phone classification or recognition.
  • Automatic detection of disfluencies in various languages
  • Using prosody and other acoustic features to detect intensively emotional words in product reviews
  • Using prosody and other acoustic features to detect emotion/conversational style from speed dates
  • Automatic measurement of phonetic variation across languages; for example we know that English dialects are more variable than North America, but could we measure this? What about Continental versus South American Spanish?
  • Cross-linguistic analysis of word production: are frequent or predictable words shorter in Chinese?
  • Are words near disfluencies hyperarticulated? If so in what way? True for all disfluencies? How about for tonal languages like Mandarin