Algorithms of Predictive Text
This week's task was to predictive text to create a thread in microblog format.
How hard could that be, let's find out...
Whew!
For some reason, I was expecting all kinds of fun and having interesting words come up to choose from. Instead, it was quite frustrating and slow, which is the opposite of what predictive text is supposed to be.
I originally approached the task without a real end goal in mind of what I wanted to say. So, my first inclination was to think that it didn’t work because I hadn’t planned for the end result. Then I tried with an idea of what I wanted to say; still, no success as I wasn’t given the words that I needed to express my ideas. I even tried a method where I would pick the biggest word offered as my path, and use an emoticon whenever it was offered. The result was even more unintelligible.
As I watch the video I can see that most of the words prompts I received were either pronouns or connecting words. There were other prompts that mainly related to time of day, day of week, or meals. The missing prompts that I needed were weighty specific words that I could use to express my own opinion.
The slow process was mostly due to my pausing to choose between three unsuitable/unwanted words. At times there was quite a bit of backspacing; my thought being that I could back up and pick a better path that leads to better words. In a lot of ways, it reminded me of the ‘seek your own adventure’ game idea and of the Twine Task from Week 5. However, it turned into a very boring ‘adventure’ as inevitably I seemed to end up getting prompts that only allowed me to string pronouns and conjunctions together, the further I went more my choices seemed to spiral down to pronouns/conjunctions.
​
This week’s material about algorithms and artificial intelligence had me thinking that perhaps it was the application I was using. I only really use my iPhone text app for friends and family, using minimal words and basic ideas; dinners, coffee, quick hellos, etc.
Perhaps the app’s AI hadn’t had the chance to learn some weighty words for the task. I use WhatsApp with my students, surely, I use some good big words there? As you can see on the left the results were marginally better, I got to use some words describing teamwork, meetings and the idea ‘best’.
In my last attempt, I tried my work email app, which should have some fancy big words in it. The result was that I was able to go a bit longer but mostly dealt with meetings, work, schedules, and teamwork. This makes sense as up until recently I was in an administrative position dealing with exactly the words that were coming up.
Work Email
In reality, the end results of my attempts went barely beyond gibberish with no real thoughts expressed, especially not my own unique thoughts
-
I couldn’t really think of any textual products that read like my statements. Certainly not academic articles, novels, or even magazines.
-
The way I express myself is vastly different from these generated texts. They seem very generic, non-specific, and rambling.
-
I can’t see my voice at all in the statement. I was unable to express any opinions or strong opinions that I had, which is quite unlike me.
My concern with predictive text would center around the non-specificity and generic quality of the statements created. Predictive text is supposed to make our communication easier and faster. I think the original intent was short quick messages
Currently, we have a vast amount of words available to us. Looking at just the English language; How Many Words are in the English Language? ; there are over 470,000 entries. The predictive texting certainly doesn’t seem to have as many words available to us.
In a way, it also reminds me of the Golden Record Task; curating down the essence of humanity into 10 representational songs. In that task, I was the algorithm, with my rules of what to include and not include. Here the algorithm is set to the most commonly used words, with the original intention of sending brief messages back and forth.
O’Neil (2017) pointed out the consequences of pernicious feedback loops in crime prevention software. My fear is that as we use predictive texting more and more, are we also dealing with feedback loops. The more we use certain words, the more they appear and our vocabulary diminishes.
So, as we seem to move further towards phones, apps, ad ease of use are we limiting our imagination and creativity. Will our use of varied language decrease and our communications become more and more generic?
Oh Oh… am I also part of the problem?
Can we be part of the solution?
I don’t text very much. Perhaps I’m not spending enough of my time on my text resulting in the AI not learning the words I need to express myself. If we are headed to a future where microblogs and texting will be our mode of communication I better start training my phone.
How Predictive Keyboards Work (and How You Can Train Yours Better)
​
​
O’Neil, C. (2017, April 6). Justice in the age of big data. Retrieved June 18, 2019, from ideas.ted.com website:
Thoughts of the Future
One of my thoughts and hopes for the future is developing a way that predictive text could become richer and more personal. Allowing users to express their voice and opinions. Perhaps the time has come for us to have the ability to obtain or download …. chosen ‘predictive text vocabulary'.
​
I could download a vocabulary pack for teacher, nurse, feedback, or even ‘MET student’ to give me the words I need to express myself.
Then I could teach it my ‘voice’
I did get an interesting glimpse into the future recently. I use a program called Grammarly to help mark student papers. I’ve been noticing lately that Grammarly has been ‘judging’ the quality of my emails and texts. As you can see below, whatever I was working on didn’t have much joyfulness, was only slightly friendly, but at least I was optimistic.
Interestingly it is a Beta version that asks for feedback on whether it ‘got it right’.
​
The AI is trying to learn…
...and I’m about to teach it that I am very friendly, joyful, and optimistic !!!
