Mise en place: Assignment 4

##Overview This assignment will be Part 1 of your final project. The goal is to think through motivation, related work, and overall approach, and collect the necessary data and (if applicable) starter code. Doing this part well should leave you with just code and analysis left before the final due date.

Keep in mind that the goal of the final project is to operationalize some aspect of what we have been discussing in class throughout the semester. You should use the project as a way to think through the challenges involved in representing language computationally, use what we have dicussed to develop a well-motivated hypothesis, test that hypothesis, and report results. It is not expected that you get state-of-the-art results on your chosen task. What is important is that you show you understand the issues in play and can talk intelligently about them.

##ToDo

1. Choose a topic. If you have already spoken to me about a custom project–great! If not, choose a one of the tasks from this years SemEval Shared Tasks.
2. Form a hypothesis. Decide on a concrete aspect of your chosen problem that you want to explore. Start with a general hypothesis which you can connect to the literature. Remember, a good hypothesis should be testable, but not provable/disprovable by a single experiment. E.g. “Knowledge of syntactic structure is important for understanding emotions in language” is a good hypothesis for this assignment. “LSTMs are good at the 2018 EmoContext task” is not.
3. Determine your evaluation. Almost certainly your hypothesis contains a claim about some feature/approach being “good at X”, “important for X”, “better at X”. You need to have a concrete definition of what “good” means for your task. If you are using a SemEval task, an evaluation metric has been set for you, and your hypothesis should be something that can be tested in terms the given metric. If you are not using SemEval, you should have some quantitative measure through which you can test your hypothesis.
4. Get your data. If you are using a SemEval task, the data should already be there, but you might need to sign up/register for the task in order to access it. If you are not using SemEval, you should determine what data you will use and get access to it ASAP if you don’t have it already.
5. Write it up. Your deliverable will be a ~4 page writeup in ACL format describing your proposed experiment. Your should writeup follow the outline given in this stencil writeup. (Please copy the project, don’t edit this overleaf directly.)

Handing In

Submit your writeup as a single pdf. Run cs2952d_handin hw4 to submit every file within the current directory.