CS 519-001, Natural Language Processing, Spring 2017, Paper Review

Requirements

Your Paper Review (min. 4 pages, max. 6 pages, 12 pt font, single-spaced, figs/tables at most 0.75 pages) due on Saturday June 10 must cover at least:

non-technical background (2.5 pts):

Who: Who are the authors? Which institutions are they from? Who are the PI and who are students?
Where: Where and when was this paper published? Did this paper receive any recognition such as an award or nomination? Does this paper have impact (number of citations)?
Reproducibility: Did the authors create and release any dataset? If not, what the datasets did they use, and are they freely available online? Did the authors release their code? Are these datasets and code influential? Is there a demo? Overall, how reproducible is this paper?
core (the famous what-why-how-wow template for writing abstract/intro) (8 pts):
What: What problem are the authors trying to solve? Is this a new problem? If not, what are the notable previous approaches, and what are their limitations?
Why: Why did they choose this problem? Is it hard? Is it important?
How: What's the authors' approach to tackle this problem?
Wow: Did this approach result in great results?
further: (1.5 pts)
But: What flaws or limitations did you see in the authors' approach or results?
More: If you were to follow up on this line of work, what topics/methods will you work on?
All: What's your overall feeling of this paper? What's the single take-home message you learned from it?

Use 12pt font, single space. Use LaTeX if possible.

List of Candidate Papers

Note: if you want to propose a paper outside of this list, get the instructor's approval by the June 1st lecture. Generally, the paper should be either a hardcore NLP paper in top conferences or journals (ACL, NAACL, EMNLP, CL, TACL), or an application of NLP in other fields such as computational structural biology, computer vision, or music processing, and the paper should use techniques taught in this class (sequences, HMMs, LMs, Viterbi, syntax, CFG, CKY, EM, forward-backward, entropy/perplexity, etc.).

IMPORTANT:

each paper can only be reviewed by at most *two* students (there will be a signup sheet online), and
teammates (who collaborated on *any* HW) must review different papers.

PAPERS:

creative language generation

phonetic spelling (chinglish): How to Speak a Language without Knowing It , ACL 2014. data.
couplet generation (chinese): Generating Chinese Couplets using a Statistical MT Approach, COLING 2008, demo.
poetry generation 1 (sonnet): Generating Topical Poetry, EMNLP 2016.
poetry generation 2 (chinese): Chinese Poetry Generation with RNNs, EMNLP 2014.
side note: Shakespearean sonnets (e.g., Shall I compare thee to a summer's day) have the following rhyme pattern: abab cdcd efef gg (the last two form a couplet), while Chinese four-line poems follow this pattern: aaca or abcb, i.e., the second and fourth lines must rhyme, and the first line rhyme is optional. The American poet Robert Frost largely used the Chinese pattern, e.g., Stopping by Woods on a Snowy Evening.
language and music
music 1 (EM): Harmonic Analysis with Probablistic Graphical Models, ISMIR 2003
music 2 (CFG): Automatic Melodic Reduction using a Supervised Probabilistic Context-Free Grammar, ISMIR 2016
language and structural biology
contrafold (CFG): CONTRAfold: CONditional TRAining for RNA Secondary Structure Prediction without Energy-based Models, 2006.
language and vision
show and tell: Show and Tell: A Neural Image Caption Generator, 2014--2015
show, attend, and tell: Show, Attend, and Tell: Neural Image Caption Generation with Visual Attention, 2015--2016
parsing
parsing 1 (MST): Non-projective Dependency Parsing using Spanning Tree Algorithms, EMNLP 2005
parsing 2 (forest): Forest Reranking: Discriminative Parsing with Non-Local Features, ACL 2008
parsing 3 (span): Span-based Constituency Parsing with a Structure-Label System and Provably Optimal Dynamic Oracle, EMNLP 2016
translation
MT 1 (pharaoh): Pharaoh: A Beam Search Decoder for Phrase-Based Statistical Machine Translation Models, AMTA 2004
MT 2 (hiero): A Hierarchical Phrase-Based Model for Statistical Machine Translation, ACL 2005
semantics
language game: Learning Language Games through Interaction, ACL 2016
algebra word problem: Learning to Automatically Solve Algebra Word Problems, ACL 2014