November 27, Morning Sessions (9:00 - 12:30)

T1: Deep Learning in Lexical Analysis and Parsing (Room 504a)

Wanxiang Che and Yue Zhang

Lexical analysis and parsing tasks, modeling deeper properties of the words and their relationships to each other, typically involve word segmentation, part-of-speech tagging and parsing. A typical characteristic of such tasks is that the outputs have structured. All of them can fall into three types of structured prediction problems: sequence segmentation, sequence labeling and parsing.

In this tutorial, we will introduce two state-of-the-art methods to solve these structured prediction problems: graph-based and transition-based methods. While, traditional graph-based and transition-based methods depend on “feature engineering” work, which costs lots of human labor and may misses many useful features. Deep learning just right can overcome the above “feature engineering” problem. We will further introduction those deep learning models which have been successfully used for both graph-based and transition-based structured prediction. 


T2: Multilingual Vector Representations of Words, Sentences, and Documents (Room 504b)

Gerard de Melo

Neural vector representations are now ubiquitous in all subfields of natural language processing. This tutorial focuses on multilingual and cross-lingual vector representations, covering representations of words, but also of entire sentences and documents. For word vectors, one can distinguish several broad classes of algorithms. The first strategy is to train multiple separate vector spaces using standard methods and then align them cross-lingually. The second strategy is to rely on parallel corpora and directly optimize a cross-lingual objective that considers aligned sentences. A third strategy is to draw on additional sources of supervision such as lexical resources or Wikipedia.

The tutorial will then turn to vector representations of entire sentences. After briefly reviewing some of the well-established methods for producing sentence vectors, we will discuss techniques targeting multilingual and cross-lingual sentence vectors. The final part of the tutorial will cover document vectors. While some papers treat sentences and documents as interchangeable, we shall see that there are significant differences between the two and discuss methods for better document-level embeddings in multiple languages. Overall, this tutorial will provide not only details of methods to produce cross-lingual representations, but also deliver practical advice of how to exploit such resources effortlessly in downstream applications.


T3: Open-Domain Neural Dialogue Systems (Room 504c)

Yun-Nung (Vivian) Chen and Jianfeng Gao

Until recently, the goal of developing open-domain dialogue systems that not only emulate human conversation but fulfill complex tasks, such as travel planning, seemed elusive. However, we start to observe promising results in the last few years as the large amount of conversation data is available for training and the breakthroughs in deep learning and reinforcement learning are applied to dialogue. In this tutorial, we start with a brief introduction to the history of dialogue research. Then, we describe in detail the deep learning and reinforcement learning technologies that have been developed for two types of dialogue systems. First is a task-oriented dialogue system that can help users accomplish tasks, ranging from meeting scheduling to vacation planning. Second is a social bot that can converse seamlessly and appropriately with humans. In the final part of the tutorial, we review attempts to developing open-domain neural dialogue systems by combining the strengths of task-oriented dialogue systems and social bots.



November 27, Afternoon Sessions (13:30 - 17:00)

T4: Neural Machine Translation: Basics, Practical Aspects and Recent Trends (Room 504a)

Fabien Cromieres, Toshiaki Nakazawa and Raj Dabre

In just a few years, Neural Machine Translation (NMT) has become the main approach to Machine Translation as well as one of the most successful application of Deep Learning to NLP. It leverages powerful machine learning techniques to train complex translation models in an end-to-end manner. Although this area of research is pretty new, the many recent developments combined with the practical difficulties of deep learning can make it difficult for a researcher lacking the background and practical experience to develop state-of-the-art models.

This tutorial is aimed at people who want to conduct NMT research but have little prior experience in this field. We will first describe in-depth the Encoder-Decoder model with Attention mechanism which has been the de facto baseline for NMT for the past 2 years. We will then consider all the practical aspects (preprocessing, training schedule, etc.) involved in obtaining state-of-the-art NMT results. Finally, we will cover the most recent advancements such as convolutional and feed-forward models. We hope that by the end of the tutorial the audience will have a working understanding of the basics, practical aspects and the recent advancements in NMT.


T5: The Ultimate NLP Research Makeup Tutorial: How to Polish your Posters, Slides and Presentations Skills (Room 504b)

Gustavo Henrique Paetzold and Lucia Specia

There is no question that our research community have, and still has been producing an insurmountable amount of interesting contributions in the most diverse areas of NLP. But for as long as interesting work has existed, we've been plagued by a great unsolved mystery: how come there is so much interesting work being published in NLP conferences, but not as many interesting and engaging posters and presentations being featured in them?

In this tutorial, we present practical and straightforward solutions to the most common problems found in posters and oral presentations. We will teach authors to change the style and content of their posters and slides in order to allow them to draw and keep the attention of other researchers, and will also introduce simple solutions to those who dread oral presentations, with a strong focus on helping non-native English speakers feel more confident.


T6: The Challenge of Composition in Distributional and Formal Semantics (Room 504c)

Ran Tian, Koji Mineshima and Pascual Martinez-Gomez

The principle of compositionality states that the meaning of a complete sentence must be explained in terms of the meanings of its subsentential parts; in other words, each syntactic operation should have a corresponding semantic operation. In recent years, it has been increasingly evident that distributional and formal semantics are complementary in addressing composition; while the distributional/vector-based approach can naturally measure semantic similarity (Mitchell and Lapata, 2010), the formal/symbolic approach has a long tradition within logic-based semantic frameworks (Montague, 1974) and can readily be connected to theorem provers or databases to perform complicated tasks. In this tutorial, we will cover recent efforts in extending word vectors to account for composition and reasoning, the various challenging phenomena observed in composition and addressed by formal semantics, and a hybrid approach that combines merits of the two.


 Important Dates

Program at a Glance

Pre-Conference Workshops and Tutorials:
November 27, 2017

Main Conferences:
November 28-30, 2017

Post-Conference Workshops and Shared Tasks:
December 1, 2017