What we'll do
Please come to meet up with fellow NLP enthusiasts to chat and to see a couple of interesting talks! Food and drinks will be provided by IBM, while the venue has been provided by the University of Melbourne.
Doors open at 6pm for talks starting at 6:30.
Talk 1: Ekaterina Vylomova
Languages differ in the way they express meanings. Some languages, like Chinese, use word order to designate the relations between words within a sentence while others pay more attention to the forms of the words but allow freer order. This talk focuses on modeling of morphology, the internal structure of words. Many languages present high morphological complexity (for instance, Archi, a Caucasian language, might have up to thousands of forms for a particular verb), and in NLP-related tasks, it becomes crucial to construct a good model that maps the form of a word to its meaning and vice versa. In many cases, this mapping is compositional, i.e. the meaning of the word as a whole is predictable from the meanings of its constituents (morphemes), and we aim to explore the mapping.
We will first discuss how well do contemporary neural models capture various word-related information. We will then proceed to automatic generation of inflected forms and derived words and illustrate advantages of neural models in high-resource languages as well as discuss various ways of improving the models for low-resource ones. By doing evaluation of the models in many languages we will demonstrate the importance of such a cross-linguistic comparison. Finally, we will state a new task of contextual morphological (re-)inflection and show its utility for debiasing the corpora used for machine translation and language modelling (for instance, in terms of gender).
Ekaterina Vylomova has just completed a PhD in Computer Science at the University of Melbourne, where she was co-supervised by Timothy Baldwin and Trevor Cohn. She is also a part of the team organizing the SIGMORPHON shared task on morphological reinflection. Her research was supported by a Google PhD Fellowship. Prior to that, she has been a visiting scholar at Montclair State University funded by a Fulbright Fellowship. She also received MSc in Computer Science at Bauman Moscow State Technical University (Russia) as well additional MSc in Machine Learning and Data Mining at Moscow Institute of Physics and Technology. In addition to that, Ekaterina participated in a number of summer schools and workshops such s CoEDL’17, JSALT’15, NASSLLI’12, RuSSIR’09/12.
Her research is focused on compositionality modelling for morphology, models for derivational morphology, neural machine translation in morphologically rich languages as well as modeling of low-resource languages. She is also interested in linguistic typology and cognitive aspects of language processing.
Talk 2: Michael Zhang
Publication information in a researcher’s academic homepage provides insights about the researcher’s expertise, research interests, and collaboration networks. We aim to extract all the publication strings from a given academic homepage. This is a challenging task because the publication strings in different academic homepages may be located at different positions with different structures. To capture the positional and structural diversity, we propose an end-to-end hierarchical model named PubSE based on Bi-LSTM-CRF. We further propose an alternating training method for training the model. Experiments on real data show that PubSE outperforms the state-of-the-art models by up to 11.8% in F1-score.
Michael Yiqing Zhang works as a Data Scientist at PredictiveHire and Hudson. He holds a Bachelor of Engineering degree from Tsinghua University (majoring in Computer Science and Technology) and an MPhil degree from the School of CIS at the University of Melbourne. His personal webpage is https://yiqingzhang.github.io/