Computational Natural Language Learning: +-20years +-Data +-Features +-Multimodal +-Bioplausible

Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning (CoNLL), pages 1–9, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Computational Natural Language Learning: ±20years ±Data ±Features ±Multimodal ±Bioplausible David M. W. Powers Artificial Intelligence and Cognitive Science Group School of Computer Science, Engineering and Mathematics Flinders University, Adelaide, South Australia David.Powers@flinders.edu.au Abstract This speech celebrates the 20th anniver- sary of the CoNLL conference and looks back 20 years before CoNLL and 20 years into the future in an attempt to paint a longterm roadmap of Computational Nat- ural Language Learning. The founders of CoNLL agonized hard and long over what to call our nascent field, and how to en- sure that we kept all the interdisciplinary diversity that we had in those early days, including preserving the richness of views in a field that encompassed many contro- versies. We will explore this diversity with a focus on new directions that are develop- ing; we will reflect on the changing nature of our technology including the decelera- tion of Moore’s Law and the emergence of Big Data; and we will consider the impact of and on ubiquitous technologies ranging from wearables to multimedia, from intel- ligent phones to driverless cars. 1 Introduction Machine Learning has moved out of the lab and into the field, and the explosion of language- related learning-research is a massive part of this. Companies like Google, Facebook and Ama- zon, as well as IBM, Apple and Microsoft, have emerged as huge players in our playground, and important sponsors! One of the most important changes we’ve seen over the last two decades is that we’ve fallen off the Moore’s Law curve as far as single core pro- cessors are concerned1 - for CoNLL’s first decade SPECInt performance was a around 50%, for our 1http://preshing.com/20120208/a-look- back-at-single-threaded-cpu-performance second decade it was more like 20% and accord- ing to NVidia it looks like being more like 5% for the next decade. The emphasis must now be on working smart and processing in parallel, but at the same time we are introducing much higher software and data overheads with managed code and extensive markup. 20 years ago few households had a personal computer or a mobile phone, while today few western families wouldn’t possess a range of equipment from phones to pads to MP-players to cars and the list goes on with microwaves, wash- ing machines and airconditioners, averaging over 20 processing units per household. All of these can be expected to be part of the Internet of Things in the very near future. Recent developments are not focussing on doing more with our computers, but using our computers more flexibly and univer- sally, with “the cloud” and “hands free” operation being major drivers of the technology race, with language technology “in the cloud” helping to deal with spoken or typed operations that would until recently have been regarded as office functions, with call centres being outsourced to computers, with IBM’s Watson challenging the quizmasters at their …

Content truncated to 500 words