CSCI 544 — Applied Natural Language Processing

Research Project

• Proposals are due February 22, 2017, at 23:59 Pacific Time (11:59 PM)
• Code demonstrations/presentations will take place on April 10.
• Final reports are due April 28, 2017, at 23:59 Pacific Time (11:59 PM)

Overview

The research project is an in-depth activity that will be carried out in teams of four. The project can be on any aspect of natural language processing. You will formulate a research question, identify resources and tools to address the question, implement and evaluate a system that uses these resources and tools, demonstrate the system, and write up a report.

• The research must involve a language other than English.
• The project must be a new effort, conducted specifically for this class. You may build on and extend previous research, but this project needs to add to that research, not just reuse old material.
• You must submit your code by putting it on a publicly available repository (such as GitHub or Bitbucket) and making the link available, at least for the duration of the class.
• Your code may use external, publicly available tools and resources, but only ones that you are able to provide when you submit your code.

Procedure

• Form teams of four students; you may use the Piazza forum to team up together based on interests such as language and topic. Each team should have strengths in all of the following areas: theory, data, coding, and writing.
• Once you have formed your team, add your information on the team spreadsheet, and create a thread on Piazza to discuss your project ideas with the instructor and the TAs. In the discussion, identify the language and problem you’re working on, the data and tools you will use, and the effort you will put in. You should aim to receive feedback before writing up your project proposal.
• Submit your project proposal by the deadline.
• Implement and evaluate your project.
• Demonstrate/present your code (details will be arranged later), and submit your code through a public repository.
• Submit your final report by the deadline.

Proposal structure

The proposal describes your plan for the research project, and will serve as the skeleton for the final report. As a plan it is subject to change and does not represent a firm commitment, but it should show that you’ve thought through the relevant aspects of your research. The proposal should be a document of about 500 words, written in English in good academic style. Proposals that substantially exceed this length (above 600 words) will be penalized. The structure of the document should be as follows.

• Title for the project.
• Names, USC IDs, and USC emails of the team members.
• Introduction. Motivate a specific problem that you will consider, which involves the use of real-world natural language data. Describe the problem you are trying to solve, why it is interesting or challenging, what existing work has been done, and how your contribution relates to that.
• Method.
  • Materials. Identify the source data that you use, such as a specific corpus that you can get access to or collect yourself. Describe the data in some detail, including the source, the amount of data, what kinds of annotation it has or needs.
  • Procedure. Describe what methods you will use to process your data, algorithms, features, and tools, and what annotations you will make (if needed).
  • Evaluation. Describe how you will evaluate your system’s performance, and your annotation procedure (if needed). What measures will you use? What baseline system will you compare to?
• References cited.
• Division of labor between the teammates.
• Word count for the document.

The proposal should be written after you have received some feedback about the general direction of your project. You will receive written feedback about your proposal, which should help you with writing the final report; however, feedback on the proposal might take some time, so don’t delay collecting your data and implementing your system while waiting for comments on your proposal. For feedback on specific issues that arise with the project, use Piazza.

Code demonstrations

Code demonstrations/presentations will take place on April 10 in PHE 516. Each team will have a 5-minute slot to talk about their work and demonstrate how their code works. No presentation slides (there’s no time for that). The code demo is a progress check; teams should have some working code to show, but it is not expected to be a final version.

Final report

The final report describes the research you have done, reporting on the method and results, relating the research to other work in the field, and offering conclusions and directions for future work. The report should be about 2000 words long, not counting the references; reports that substantially exceed this length will be penalized. The structure is similar to the proposal, but with more detail, and two additional sections following the method section.

• Title for the project.
• Names, USC IDs, and USC emails of the team members.
• Introduction. Motivate the problem that you have worked on, which involves the use of real-world natural language data. There is no need to motivate Natural Language Processing in general, but rather your specific application. Describe the problem, why it is interesting or challenging, what existing work has been done, and how your contribution relates to that.
• Method.
  • Materials. Identify the source data that you use, such as a specific corpus that you accessed or collected yourself. Describe the data in some detail, including the source, the amount of data, what kinds of annotation it has, and what annotations needed to be added (if any).
  • Procedure. Describe the experimental procedure, that is what methods you use to process the data. This may include algorithms, features, and tools, and any annotations you made. Well-known methods (such as Naive Bayes or Conditional Random Fields) do not need to be explained, but you do need to explain how you use them, for example the features you choose. If you created a tool (such as an annotation or visualization tool), describe it in some detail.
  • Evaluation. Describe your method for evaluating the system’s performance, and for evaluating your annotation procedure (if any). Include a description of the specific measures will you took, and the baseline to which you compare your system.
• Results. Report how your system performs, and how it compares to the baseline or to other comparable work. Discuss what it gets right, what it gets wrong, and why.
• Discussion. Discuss conclusions that can be drawn from the research, implications of your findings, the overall contribution to the general NLP community, and directions for future research.
• References cited. You may choose your preferred style for in-text citations (for example, numerical or author-year) and for the reference listing, but please keep it consistent across the document. The reference listing should contain all the information required for accessing the reference – author(s), year, title, and publication information (such as the conference, journal, volume etc.).
• Division of labor between the teammates.
• Word count for the document, excluding references.

The six main content sections (introduction, materials, procedure, evaluation, results, and discussion) carry equal weight. Therefore, they should be of similar lengths – this means reserving about 300–350 words for each section. This is only a general guideline, as you may find that some sections require more text than others. However, if you find you have more to say than fits within the length requirement, then you’ll need to concentrate on the more important aspects of your project.

When giving examples of text in languages other than English, please use the following multi-line format, to make the examples readable to English speakers. Below is an example for how to present a sentence in Hindi.

The second line is not needed if the language natively uses a version of the Latin script.

Grading

The grade for the assignment will be broken down as follows.

• 10% Discussion of the project with the instructor and TAs.
• 10% On-time submission of a coherent proposal.
• 10% Difficulty, creativity and originality.
• 20% Code demonstration and code quality.
• 50% Final report.

The research project counts for 30% of the overall course grade.