Sunday, May 22, 2016

A quick and dirty medical dictionary for your Mac

I work on a lot of health IT projects at work and, as such, I face a constant stream of red squiggles provided by the spellcheck utilities of almost every major editor.  Over the years, I've built up quite a few 'User Dictionaries' by adding terms to each spellchecker one-by-one.

Today, I decided to look into whether my Mac offers a more centralized user dictionary to see if I might pre-populate it with all these medical terms instead.  Answer: Yup. And its just a simple text file with one term per line.

Now this is where the little-known SPECIALIST Lexicon from the US National Library of Medicine comes in.  The lexicon is a freely available "syntactic lexicon of biomedical and general English" that includes terms in the following format:
{base=cotton roll gingivitis
spelling_variant=cotton-roll gingivitis
entry=E0000001
        cat=noun
        variants=glreg
        variants=uncount
}
{base=2060 virus
entry=E0000010
        cat=noun
        variants=reg
}
...
There may be a version with one term per line, but I figured it would be easier just to use some unix-fu to get it in the right format.  So, if you're following along, simply open a Terminal and do the following:

1. Navigate to the latest Lexicon release and download the Lexicon (text) by right-clicking the link and choosing the appropriate save option (don't left-click/navigate to the link unless you want your browser to hang for a bit...the file is fairly large).

2. Make a LocalDictionary backup (optional)
If you already have a LocalDictionary, you may want to back it up first in case something goes wrong:
cp ~/Library/Spelling/LocalDictionary ~/Library/Spelling/LocalDictionary.backup
3. Navigate to where you downloaded the LEXICON file and add the terms from it
Perhaps a little dirty, but I don't think this LEXICON format has changed in like 10 years, so here are a couple lines to add the terms to your Local Dictionary in the proper format:
grep "{base=" LEXICON | cut -c 7- >> ~/Library/Spelling/LocalDictionary
grep "spelling_variant=" LEXICON | cut -c 14- >> ~/Library/Spelling/LocalDictionary
4. Sort the terms
According to that tutsplus article I linked earlier, the Mac spellcheck file needs to be sorted alphabetically.
sort -o ~/Library/Spelling/LocalDictionary ~/Library/Spelling/LocalDictionary 
If you made a LocalDictionary backup, I won't tell you to go delete it yet because I've only just made the change and so I'm not sure whether having an 8 MB LocalDictionary will have any downsides, but so far it seems to be working.

Friday, January 22, 2016

Training your digital assistant

Despite reading articles like http://www.wired.com/2015/09/voice-interface-ios/, every time I use a digital assistant application to interact with my device, I'm reminded of how limited they really are. Still useful for a certain set of commands (notes, reminders, factual questions, etc.), but not even close to the type of "conversational computing" imagined in videos like http://blog.ibmjstart.net/2016/01/07/why-being-able-to-talk-to-your-computer-may-change-your-life/

So what I'm asking for is much simpler; just let me easily program a set of voice commands for my own device. Or, more precisely, let me teach my digital assistant how to interpret my words over time. When you search google about training a digital assistant like Siri or Cortana, all the information seems to be about training the voice recognition.  But thats only a small part of the problem.  Why isn't it easier to create simple verbal commands to trigger a specific action on your device?

For example, if there is an app I use a lot, I should be able to set a verbal launch command (without buying a 3rd party app launcher).  I should be able to record "macros" of events on the phone and then assign a verbal command to replay that exact set of events (e.g. open an app, click a button in the app, swipe to the second tab/screen of it).

Basically, I think the voice recognition should evolve with me to understand the types of things I'm likely to ask for. If it misinterprets what I'm asking for, I should be able to give it a simple piece of negative feedback so that it can try a different interpretation in the future. Or better yet, it could learn to measure its success based on my implicit feedback (if I redo a very similar command, I'm probably looking for a different result).

I really do hope that WIRED article is right and that smarter digital assistants are right around the corner, but I wouldn't hold my breath. If they do arrive, we'll just have to solve that silly little problem of getting people more comfortable with talking to their devices.

Training the next generation to dictate

Recently, I've shared some thoughts on the current user experience of digital assistants like Siri and Cortana.  A couple years ago, I bemoaned the lack of a more direct voice-to-system channel in our modern consumer technology:
http://parenthetical-thoughts.blogspot.com/2013/11/walkie-talkie-ultimate-phone-experience.html

One of the reasons I'm so stuck on this is that I feel like we finally have the technology to make conversational computing a normal mode of interaction with our devices.  I've been surprised to see that so many people are carrying smart phones and yet very few "normals" are using voice input.

I doubt that many would disagree that mobile devices are not good for typing. And yet, much to my surprise, younger generations have turned to text messaging as a primary form of communication, defacing the english language to make it slightly less painful to type their messages on their tiny little devices.

Sure, there are many contexts in which a voice input channel doesn't make sense, but I would argue that much of the time it does. So, if dictation is a better, more efficient, input mechanism for the majority of users, in the majority of cases, why isn't it more common? One explanation for this is that dictation is hard.

When I'm driving and try to dictate an email, text message, or note on my phone, its always full of stutters and breaks. To do it properly, I need to think out my entire sentence before I verbalize it to my device. While typing, I find it much easier to finish the thought in the midst of entry. Its funny how different it feels to talk to your device than it does to talk to a person...even if that person is not in the same physical location.

My theory is that this is learned, societal behavior. Further, I suspect that if I started dictating at a very young age, that dictation process would feel much more natural. Has anyone done an experiment like that? I'm seriously considering teaching my daughters to get into the habit of talking to a device just to see...maybe some kind of diary app?