CSCI 544 — Applied Natural Language Processing

Research Article Presentation

Due dates

• By November 2, 2020: Choose partner, time slot, and article
  • The earlier you choose, the more options available to you
• November 10–24, 2020: Article presentations (in class) and submission of presentation materials
• November 10–24, 2020: Comment on other students’ presentations (in class)

Overview

The research article presentation is an activity where students read an article of their choice and present it to the class. The article can be on any aspect of natural language processing. You will read the article, identify the central points of the research, and present that research to the class. You will also comment on research article presentations of fellow students.

• The article must be published and publicly available.
• The article may not be taken from among the required or optional readings for this course.
• The article presentation must be a new effort, conducted specifically for this class. You may not reuse an old presentation, or present an article that was read in another class.
• Your presentation will be recorded on Zoom and available for the class to see.
• You must submit your presentation materials (slides) within 24 hours after your presentation; the materials will be made available to the class on Blackboard.
• You must engage with at least two other student presentations by making substantial, live questions or comments on the presentation, either by voice or through the chat feature on Zoom.

Procedure

• Find a partner for your presentation; you may use the Piazza forum to team up together based on interests such as language and topic. Since there is an odd number of students in the class, we will approve one group of three students; please talk to the course producers if you want to be in the group of three.
• Select an original research article to present to the class; a good place to look for articles is the ACL Anthology. The article may be in any area of language processing, and on any human language. The article does not need to relate to your final paper.
• Select an available presentation time slot from the presentation spreadsheet (link will be provided on Piazza), and put your names, the article title, and article URL in the appropriate fields.
• Important: sign-up for slots and articles is on a first-come, first-serve basis. When you sign up for your slot, make sure nobody else is already signed up to present your article. If you find that your article is already taken, choose a different article.
• Present your chosen article in class at the assigned time.
• Attend at least one live presentation session (it may be the one that you present in), and make substantial questions or comments on at least two presentations in that session using either voice or chat on Zoom.
• Submit your presentation materials (slides) within 24 hours after your presentation.

Choosing a research article

• Pick an article which has an interesting theme, point, or result. Some examples of the main theme from articles read in class are:
  • Brill (1992) shows that automatically-learned tagging rules can be as effective as a statistical tagging model.
  • Gildea and Palmer (2002) explore the usefulness of full syntactic parsing to the task of semantic role labeling.
  • Mikolov, Yih and Zweig (2013) discover that training a model to predict the next word in a sentence yields a representation that can solve analogy questions.
  The theme for presentation does not have to cover the entire article.
• If you cannot identify an interesting theme, point, or result in your article, choose a different article.
• Remember that the audience would like to learn something new and interesting. Pick a theme that can be presented within the allotted time to an audience with the background covered in class.

Presenting a research article

• You have a slot of 15 minutes, including transition to the next pair. Plan on presenting for about 8–9 minutes, leaving about 5–6 minutes for questions and discussion. Practice your presentation to make sure it fits within the allotted time.
• You are encouraged to use presentation slides; remember that the conversation should revolve about the content of research in the article, not about the slides.
• The audience has not read the article: assume no prior knowledge of the topic on their part. You may assume knowledge of all the methods and techniques covered in class, but not much beyond that; if the article uses a new method or technique, explain it.
• Concentrate on presenting the main idea of the article. Present enough details and results to support the main idea, but don’t get into such detail that the main idea gets lost.
• Explain the linguistic side of what the article is trying to accomplish; you will probably also need to explain a little about the language that is the object of study (unless it is English).
• Include some critical evaluation of the article’s content. For example, did you notice any flaws in the reasoning, evidence, conclusions? Do you know of other facts that are relevant to this study? Does this relate to something else you have studied or observed? Can you think of a good follow-up? (The critique should relate to content, not writing style.)

Presentation schedule

November 10th

• 8:05 AM Improving Language Understanding by Generative Pretraining
• 8:20 AM Towards Explainable NLP: A Generative Explanation Framework for Text Classification
• 8:35 AM Attend to Medical Ontologies: Content Selection for Clinical Abstractive Summarization
• 8:50 AM BERT Rediscovers the classical NLP pipeline (2019)
• 9:05 AM Evaluating Gender Bias in Machine Translation
• 9:20 AM POS Tagging of English-Hindi Code-Mixed Social Media Content
• 9:35 AM Learned in Translation: Contextualized Word Vectors

November 12th

• 8:05 AM Tackling Error Propagation through Reinforcement Learning: A Case of Greedy Dependency Parsing
• 8:20 AM Few-Shot NLG with Pre-Trained Language Model
• 8:35 AM Bag of Tricks for Efficient Text Classification
• 8:50 AM Chinese Language Processing Based on Stroke Representation and Multidimensional Representation
• 9:05 AM Is Word Segmentation Necessary for Deep Learning of Chinese Representations?
• 9:20 AM Slang detection and Identification
• 9:35 AM Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems

November 16th

• 7:05 PM Text Summarization Using Neural Networks
• 7:20 PM Bridging the Gap between Training and Inference for Neural Machine Translation
• 7:35 PM Generating Responses with a Specific Emotion in Dialog
• 7:50 PM a fast and accurate dependency parser using neural networks
• 8:05 PM Improving Black-box Speech Recognition using Semantic Parsing
• 8:20 PM Natural Language Processing for Sentiment Analysis
• 8:35 PM Emotion-Cause Pair Extraction

November 17th

• 8:05 AM Zero-shot Word Sense Disambiguation using Sense Definition Embeddings
• 8:20 AM Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• 8:35 AM Human-Machine Dialogue as a Stochastic Game
• 8:50 AM Get To The Point: Summarization with Pointer-Generator Networks
• 9:05 AM Categorizing Comparative Sentences
• 9:20 AM A neural attention model for abstractive summarization
• 9:35 AM Formation of SQL from Natural Language Query using NLP
• 9:50 AM Universal Sentence Encoder

November 19th

• 8:05 AM Learning sentiment-specific word embedding for twitter sentiment classification
• 8:20 AM Unsupervised Learning of Video Representations using LSTMs
• 8:35 AM Semantically Conditioned LSTM-based Natural Language Generation for Spoken Dialogue Systems
• 8:50 AM Combination of Recurrent Neural Networks and Factored Language Models for Code-Switching Language Modeling
• 9:05 AM GAN-BERT: Generative Adversarial Learning for Robust Text Classification with a Bunch of Labeled Examples
• 9:20 AM REFORMER: THE EFFICIENT TRANSFORMER
• 9:35 AM Using Large Pretrained Language Models for Answering User Queries from Product Specifications

Grading

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

• 20% Organization of presentation (clarity, accuracy, thoroughness).
• 20% Delivery of presentation (clarity, accuracy, thoroughness; not speaking style, fluency, nervousness).
• 10% Critical evaluation of article.
• 10% Fitting within the allotted time.
• 20% Response to questions.
• 20% Substantial comments on other student presentations.

The research presentation counts for 5% of the overall course grade.