Candidate topics

• Lyric translation from english to other language with meaningful sentences

• Document classification/recommendation methods that gives more control to its user.

  • For instance, we can split a single twitter timeline into multiple stream grouped by topic
  • Or we can filter out uninteresting tweets from certain timeline
  • Other features like account, retweet/share, like, conversation context could be used
    • Which means we have an opportunity to build a model upon this functionalities
  • We need to crawl data from twitter/facebook/...

• SIGMORPHON 2016 Shared Task: Morphological Reinflection

  • Well-defined problem
  • Dataset and a baseline implementation are given

• Transcription system between languages

  • Stanford Phrasal: A Phrase-Based Translation System
  • Hangulize

• Language Identification

  • Identify which language is used for a given document
  • Well-defined problem with easy-to-find dataset (Wikipedia)
  • Still has a room to improve

• Predict closed questions on Stackoverflow

---

• Wikipedia page for NLP categories

• Similar projects, might be useful for evaluation of our project scope/size

  • CS224d: Deep Learning for Natural Language Processing
  • CS 229 Machine Learning Final Projects, Autumn 2013

• Stanford's Deep Learning for Natural Language Processing course.

  • Applying deep learning to NLP becomes a quite hot trend recently, though someone need to be familiar with machine learning techniques.

• Narrative generation

• Domain specific language to simplify integration of certain NLP tasks into applications.

• Choose some NLP systems tightly coupled to a specific language and generalize it.

---

• Is is possible to apply NLP techniques to a programming language implementation?

  • Some topics could be benefited from NLP techniques
    • Parsing incomplete/incorrect code
    • Use natural language information (comment, function/class name ...) for static analysis
  • Though not sure about whether a PL can be considered as "a language other than English"

• Automatic synthesizing of a constructed language

  • Consider that given time frame is short, full construction seems to be infeasible
  • Need to focus on a specific structure like phonetic/morphological ...
  • Can be used for testing how a particular NLP technique works well in unknown language

• Identifying the geographic of a person based on his sentence formation.

• Augmenting procedural content generation using NLP techniques, somewhat related to narrative generation

  • Automatic generation of a natural language dialog with user-given constraints
  • ex. Rule-based quest line generation is not that hard, but it usually lacks of plausible dialogs

• Develop a scoring algorithm for student-written short-answer responses.

  • The problem is that we don't have data set other than English