AHC Seminar Series

From AHC Wiki
Jump to: navigation, search

This is a page for the seminar series of the Augmented Human Communication lab at the Nara Institute of Science and Technology. We hold presentations by excellent researchers from around the world on topics of speech processing, natural language processing, machine learning, and other related topics. Visitors from outside the lab are welcome!


_ _

2016-6-30: Alvin Grissom II -- Simultaneous Machine Translation in Distant Language Pairs

  • Speaker: Alvin Grissom II (University of Colorado, USA)
  • Title: Simultaneous Machine Translation in Distant Language Pairs
  • Date/Time: 2016-6-30 (Tue.), 15:10-16:40
  • Location: CL Lab Seminar Room, Information Science A708

In this talk, I will talk about the challenges inherent to simultaneous machine translation in distant language pairs and discuss some of our work on tackling this problem.. Recently, approaches to simultaneous machine translation—translating sentences incrementally, before they are complete—have attempted to incorporate machine learning to achieve simultaneous machine translation from verb-final languages (such as German and Japanese) to verb-medial languages (such as English and Mandarin). Due to the divergent syntax of these languages, particularly the head-finality of verb-final languages, simultaneous translation is a great challenge for both humans and machines, requiring the incorporation of predictions about the end of the sentences to achieve natural-sounding translations without waiting for them to be uttered. This problem is not trivial, yet relatively little attention has been given to it in the computational linguistics literature. We use incremental verb prediction to tackle this problem. By predicting verbs in SOV languages before they have been spoken, we can get ahead of the speaker, and by learning when to trust these predictions, we can minimize the propagation of error from incorrect predictions to the incremental translations. For the task of determining when to trust incrementally-revealed predictions and what actions to take based in part on these predictions, we turn to reinforcement learning. By combining incremental linguistic prediction with reinforcement learning, the simultaneous translation system can minimize translation errors introduced by relying on imperfect predictions, leading to less error-prone translations that still manage to be expeditious under realistic constraints on time. I will also briefly discuss some related work on understanding "Interpretese" and implementing syntax-based strategies to rewrite sentences to reduce delay.

2016-3-22: Miltos Allamanis -- Learning to Name Software Artifacts from Big Code using Machine Learning

  • Speaker: Miltos Allamanis (University of Edinburgh, UK)
  • Title: Learning to Name Software Artifacts from Big Code using Machine Learning
  • Date/Time: 2016-3-22 (Tue.), 13:30-14:30
  • Location: AHC Seminar Room, Information Science B708

Talk Abstract

The abundance of available source code raises the exciting possibility to develop novel machine learning-based software engineering tools that learn from large corpora. These probabilistic source code analyses necessitate new interdisciplinary research in machine learning and programming languages. The core challenge rests in developing methods that can handle probabilistically the multifaceted and highly structured aspects of source code.

Software developers care deeply about naming. In this talk, I discuss two machine learning models that learn to name variables, methods and classes. In the first part, I present a neural log bilinear network that learns from existing code and — using hand-crafted features — successfully suggests names for variables, methods and classes. In the second part, I present a convolutional attentional neural network that learns local and long-range features from the tokens of a code snippets — without any hard-coded features — and predicts a method name-like summary. We evaluate both models in popular real-word Java projects.

Speaker Profile

Miltiadis (Miltos) Allamanis is a PhD candidate in the School of Informatics in the University of Edinburgh, UK, funded by Microsoft Research. His research interests concern applying and creating new machine learning and natural language processing methods to create novel and smart software engineering tools. In his PhD —advised by Charles Sutton— he works on machine learning models of source code and their applications in programming languages and software engineering. Miltos has previously interned in Microsoft Research (Summer 2014 in Cambridge, UK and Spring 2015 in Redmond, WA, USA). He holds an MPhil in Advanced Computer Science from the University of Cambridge, UK and a DipEng in Electrical and Computer Engineering from the Aristotle University of Thessaloniki, Greece.