I’m sitting in the car, waiting to begin another long road trip, and in keeping with recent practice, this will be another chance to test my hands-free learning tools. But in light of my current ALG experiment, there will have to be some changes to the plan.
◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇
First up, I won’t be using the English translations, and I’m going to limit my use of several other features, too.
The objective in ALG is to continually expose your ear to natural speech. When I’m listening to Norwegian spoken quickly, I often wish I could pause the audio and give myself time to process what’s being said.
Frankie does this naturally, with its “one sentence at a time” language study philosophy, but it does not yet handle audio content very well. That’s why I created the quickstudy tool. It’s a fast and dirty engineering mockup that lets me test different possible behaviors and decide what works best before adding the features to Frankie.
Quickstudy includes a slow speech mode that can play the sentences at a slower cadence, emphasizing the spaces between words. This makes it easier to decode the speech into words, but in the context of ALG, that seems a step too far. So I won’t be using the Slow Norsk button during this test.
Similarly, allowing myself to repeat a sentence if I didn’t catch it also misses the point of ALG, so I’ll try to resist using that button too.
Ideally I should be pressing the Next button as fast as possible, pushing my ear to keep up with the natural pace of the speaker, but I’m just not there yet. So I’ll cut myself some slack at the start, and then increase my pace, until I hopefully reach the point where the audio is playing continuously, without any pauses at all.
The other change I’ve made to quickstudy is that I can now run it on a bunch of lesson bundles at once.
During my last test, I was frustrated by having to pull the car over and restart the tool every time I completed a lesson pack and wanted to move to the next one. With these new changes, I can now run it on the entire directory of lesson bundles and then switch between the lessons with a simple button press. (The 8bit Micro Bluetooth game controller is proving to be an outstanding solution for my hands-free screen-free use-case.)
But time’s up for this update. My wife is ready to go so it’s time to get this test lab rolling. See you soon.
Update #1: The first leg of the trip wasn’t very productive. Of the 11 lesson bundles I’ve got with me (selected episodes of the Lær Norsk Nå podcast) 10 were too quiet to hear clearly over the road noise. I don’t have access to my audio processing lab, but I think I can write a script on my phone to reprocess the bundles and normalize the clips at a higher gain. If it works I can run a better test on the way home.
Note-to-self: Add clip normalization step to the bundle production tool chain.
Update #2: The return trip was better, in that the audio was now loud enough to hear over the road noise, but that only revealed the second problem: timing.
Most of my clip bundles were assembled using the forced alignment algorithm I outlined here, but this is the first time I’ve tried to listen to them all the way through. They were indeed sentence-long clips, but the start and end timings are not accurate enough. Many clips start or end half a word too early or too late, which I found far more distracting than I should have. Instead of focusing on the sentences, I found myself increasingly irritated by the clumsy edit and focusing on that instead. In the end, I abandoned the experiment, to preserve the tranquility of the drive home.
Note-to-self: Add a manual boundary-tweaking step to the automatic alignment process. Using aegisub, I can quickly scan through the timings and move the clip boundaries to silent moments visually, by referencing the waveform display. It’s the same tool I used for creating the clips manually, but by starting with computed boundary times, it should be much faster.
Language is an extremely flexible tool. You can assemble words into messages that express an unlimited number of thoughts. Anything from “Where food?” to “Energy equals mass times the speed of light squared.”
But in everyday speech, we don’t typically use words to communicate. We actually assemble most of our day to day utterances using larger chunks of language. What are those chunks? And why do I think they can be used to supercharge your language studies?
Let’s take a look…
While working on the ear training features for the FrankenTongues app, I stumbled across a reference to the Automatic Language Growth (ALG) model of language learning, and the moment I read it, I had to stop everything I was doing to investigate.
Because it resonates loudly with my own views on how we learn languages.
With my recent experiments in ALG-style language study, I needed to shift gears in my Norwegian reading practice. Instead of reading adult-oriented books and using translation tools to help me over the rough parts, I’m embracing kids books and white-knuckling my way through them without a safety net - no dictionaries, no translations, just context and guesswork.
Today I finished my first one, and thought I’d share some notes about this new approach.