The Dudes Do ISAAC 2012: Day 5 – Of Language and Linguists

I’ve been an SLP for almost 30 years but my first degree is  in Psychology and Linguistics. I fully intended to become a Psychologist [1] but strayed from the path and ended up in Speech Therapy. Needless to say, my fascination with our profession has always been viewed through a linguistics lens and regular readers will already have detected that. In fact, one of the luminaries of AAC, Sarah Blackstone, for many years believed I was a linguist and not an SLP, and I have been introduced as a linguist at more than one conference.

This is probably why I still like to hang out with real linguists, who are much smarter than I and from whom I continue to learn lots of new stuff. So it’s no surprise that I went along to Wednesday afternooon session entitled Natural Language processing and AAC: Current advances at the interface between technology and communication.

Natural Language Processing

NLP

 The presenters were more like a panel, bringing different perspectives on how the application of NLP could help the development of AAC [2]. NLP is a cross-over field of linguistics, artificial intelligence, and computer science that deals with analyzing, understanding and generating the languages that humans use naturally in order to interface with computers in both written and spoken contexts using natural human languages instead of computer languages. The main professional body that exemplifies the scope of NLP is the Association of Computational Linguistics, which publishes the journal, Computational Linguistics, on a quarterly basis. With the latest edition including articles with titles such as A Context-Theoretic Framework for Compositionality in Distributional Semantics, and Learning Entailment Relations by Global Graph Structure Optimization, it’s not a field that SLPs are falling over themselves to join. It’s also not a journal I read regularly but then there are so many journals out there it’s impossible to keep track.

SLPAT logo

The folks were also there to promote awareness of a new special interest group called the SIG for Speech and Language Processing in Assistive Technologies, or SLPAT for short. And yes, if you misread it as “splat,” you’re OK because its members also affectionately call it “splat” as well [3].

For those who wonder what Natural Language Processing might offer to AAC, it’s worth bearing in mind that NLP is already being used in a number of areas that we use daily. If you’ve ever used a web-based translation system to read foreign text, then you’ve made use of NLP. If you’ve ever used a speech recognition system such as Dragon NaturallySpeaking or Microsoft Sync in your car, then you’ve benefited from NLP. And if you’ve scanned a document and had a piece of software convert it to text, you’re seeing NLP in action.

In relation to AAC, a number of research initiatives are already underway. Jeff Higginbotham from the University of New York at Buffalo is working on a “just-in-time” message system that will work with AAC devices to provide web-sourced topic-based content using internet (and intranet) natural language processing techniques. Annalu Waller from the University of Dundee is working on prediction-based phonemic AAC systems  (the PhonicStick®) where NLP algorithms are used to determine which sounds are most likely to follow others. Karl Wiegard and Rupal Patel have been investigating non-syntactic word prediction to create systems that can correct user-generated utterences that have flawed syntax. And at the Health and Science University in Portland, Oregon, Melanie Fried-Oken and her colleagues are working with brain-controlled interfaces and spelling systems, the operation of which can be optimized by using special NLP-based software to improve accuracy and speed of selection.

All of these are currently still in the research phase so you needn’t be asking how much you’ll need to buy one, but it’s research like this that will ultimately lead to products, even if 90% of university projects simply end up as articles in journals or a paper that ends with the immortal lines “much more research is needed.”

And if you’re looking for questions, quite a few turned up at the session, most coming from Melanie Fried-Oken who, as a clinician, really wants to see some practical, hands-on solutions. Here are some – of several –  that interested me;

  • Can NLP help us design systems that can adapt to the actual language used by an individual with an AAC device, and maybe even reconfigure the device as a result of this?
  • Can NLP help in the tracking of the vocabulary, representation, and navigation elements of an AAC system?
  • Can NLP help design systems that identify and end-user’s language level?

These resonate because they are the very same questions my collegues and I have been asking for a couple of years now, and have been slowly working towards. In the field of AAC in general, the notion of automatic data logging is not new and has been available for some time on a number of AAC devices. The fun bit is deciding where next to go with this, and how best to leverage the current data collection methodologies. As soon as there’s something to present to the world, we’ll be happy to share!

Meanwhile, for those interested in finding out more on SIG SLPAT, or even if you want to join, you can go to their web site at www.slpat.org and read about the aims of the group [4]. There’s a special edition of the Computer Speech and Language journal out before the end of the year that will be about NLP and Assistive Technology, and the next SIG-SLPAT conference will be in 2013 in France – somewhere. There will be a call for papers later in the year so get your NLP thinking caps on and dust off that passport…

Notes
[1] Just a few weeks before I left for University, a friend of my sister was talking to my local newsagent about my moving and asked what I was going to study. Apparently she told him I was studying to be a psychopath. I sometimes wonder how un-wrong she may have been…

[2] The presenters were Kathy McCoy, University of Delaware; Annalu Waller and Alan McGregor, University of Dundee; and Melanie Fried-Oken and Brian Rourk, Oregon Health & Science University. I apologise if I missed someone.

[3] It’s pretty well impossible not to read SLPAT as “splat,” in the same way that fashion store French Connection: UK used the acronym FCUK on all their advertising, knowing full well that folks word read it otherwise! The company voluntarily stopped using the acronym in 2005, but not before stores such as Bloomingdales refused to handle FCUK branded items.

[4] I did see what would happen if I made an error and typed “splat.org” instead of “slpat.org” and found myself at a rather boring “parking site” with links to paintball activities. More fascinating was the “splat.com,” which took me to the home of the Sizzling Platter restaurant group, whose products include Little Caesars Pizza, Red Robin Gourmet Burgers, Dunkin’ Donuts, and Sizzler Steak House.

Advertisements

2 responses to “The Dudes Do ISAAC 2012: Day 5 – Of Language and Linguists

  1. Thanks for the summary. I didn’t make it to that session. I wonder if AAC is better off going to ACL with broad issues related to the kinds of interactions people using AAC have and then asking for their ideas about solutions rather than trying to adapt NLP themselves?

    • What was good to see at the session was the multi-disciplinary panel, which served to remind folks that “AAC solutions” are not, and never have been, one-sided affairs. The nature of the presentation was such that the focus of the topic was skewed toward technical AAC solutions, and more specifically to technical solutions that uses computer-based devices, which in turn might be able to use some form of artificial intelligence to make AAC devices “smarter.” On that basis, the folks working in ACL could certainly help in the nuts and bolts of developing code to enable users of AAC systems to have better verbal interactions. I’m currently working on applying the data and methodologies of Corpus Linguistics to (a) improve the design of the language systems within a device and (b) to develop databases that can used to analyze user-generated language, with the ultimate aim of using that generated language as feedback to the system as a whole, such that the system could “adapt” itself to better suit the user’s needs. This will require me (as I’m an SLP and not a software engineer) to work with folks in the ACL/NLP field to come up with good database design and clever heuristics that “understand” human language in terms of morphology, syntax, and semantics.

      The creation of the SLPAT group might help me with this, but coming from the private sector, I’ve seen a number of academic projects come to nothing because of over-analysis, professional jealousies, and failure-to-complete (the rider “more research is needed” can stop an idea ever becoming a product!) This isn’t a criticism of academia in general but just the observation that grant-based projects typically proceed at a slower pace than commercial ones – and that’s probably nothing more than a statement reflecting the different processes involved. So I may sign up and join but I’m one of those folks who feels that the bigger the group inputting a project, the slower it goes and the less likely it is to succeed; “A committee is a cul-de-sac down which ideas are lured and then quietly strangled.” (Sir Barnett Cocks, House of Commons Clerk-of-the-House, 1962-73).

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s