Tips for Literature Review

Sources

• ACL Anthology
• Google Scholar
• ICAART
• Arxiv - becoming common practice for upcoming publications

The top 10 NLP conferences - a pragmatic and specific guide for evaluating NLP papers, by Rob Munro.

Accessing the article

• If you're on campus, you'll automatically get access to most journals
• If you're off campus, add this to your url after the .com: ezproxy.lib.utexas.edu and you should get prompted to enter your eid and password.
  • e.g. http://onlinelibrary.wiley.com/doi/10.1002/jrsm.1037/pdf ==> http://onlinelibrary.wiley.com.ezproxy.lib.utexas.edu/doi/10.1002/jrsm.1037/pdf
• Sometimes the proxy doesn't work. Try to access the journal directly from the UT Library Catalog
• If UT doesn't subscribe to your journal of interest, you can still request a copy through the Get a Scan service

Keeping track of it all

• Create a document-- spreadsheet is recommended
• Capture the most important info after you read each study, such as:
  • doi
  • first author
  • year of publication
  • URL to online paper
  • model(s) used
  • features used
  • evaluation dataset
  • results (e.g. f-score)
  • shortcomings
  • other relevant info you want to compare

Most people will refer to papers by their citation, e.g. (Baroni and Zamporelli, 2010), or (Levy et al., 2015). You should know these by memory, at least for anything you may cite yourself.

Keeping a well-formatted bibtex file is also highly recommended; it helps in quickly inserting citations when writing your own papers. If you keep notes for each paper in an extra bibtex comment field, then your literature reviews can practically write themselves. As an example: Peter Stone's most cited paper is a thorough literature review that he wrote by just expanding on his BibTex comments!

Writing it down

• Start writing as soon as you can
• Choose a tool that lets you version and collaborate easily, such as:
  • Overleaf
    • PROs: easy to collaborate simultaneously, has a built-in LaTex compiler, has ACL templates
    • CONs: it's public (unless you pay), and you need to remember to save a version manually
  • Bit Bucket
    • PROs: it's private, better versioning, have your code and doc in the same place
    • CONs: manual merge required if collaborating on the same section of a document, doesn't have built-in LaTex compiler?
  • Github
    • PROs: the common standard in distributing/collaborating code
    • CONs: only five private repositories (with .edu email); no latex compiler; git has a high learning curve (but definitely worth the investment)
  • Dropbox
    • PROs: pretty much everyone has one. Very easily to collaborate; used by Ray and his group
    • CONs: no conflict resolution, so you need to send emails to each other for "locking X.tex"