Posted by A couple of years ago my family and I were on holiday in Belgium when I developed a fascination for Flemish. Most people where we were staying spoke French, so it was easy enough to chat and ask for things, but this Flemish business had me absolutely captivated, the etymology of the language and the sound of it.
When I got back home, everyone else was on Duolingo, the language app that is so popular nowadays, so, I looked up Flemish. They didn’t have Flemish but they did have Dutch.
I did ‘A’ Level History but the period I did was 1648-1815 so I didn’t learn that Belgium used to be part of Holland and gained independence in 1830. We were there last year on Independence Day and enjoyed the fireworks. I did skirt about the issues of the similarities in the language Flemish/Dutch etc., for a while as language is complex. It is identity, culture and a fascinating ponder in a whole other blog, but the Belgians are not precious about it. They say yes Flemish is Dutch.
That sort of thing might work in Holland
The day after Belgium Independence Day, I said to the man on the Bruges boat tours: Vanaf vandag ben ik de baas hier which means: From now on, I am the boss here. He looked at me unamused and said in English: That that sort of thing might work in Holland, but definitely not here. That night back at the campsite, the campsite owners, who appreciated my forays into learning Dutch, roared with laughter as I told them what he had said and typing this now thinking of them roar with laughter is so much fun and makes me laugh.
And so here I am, 525 days later faithfully learning Dutch daily, endlessly charmed and fascinated by this language that I have absolutely no need to learn. All the Dutch people I have met speak excellent English or as in the case of a Flemish couple with whom I got chatting in the Horta Museum in Brussels, excellent French. And even though I’ll probably never become fluent, I am amazed how far I have gotten with five minutes a day on an app, without ever writing anything down.
Last October, we went to Germany so our youngest could practice German but alas, the town was on the border between Germany and Holland so everyone spoke a big Dutch-German mash-up, which I found endlessly fun and entertaining, but is not so good for German GCSE practice.
My latest favourite word is fisdrank (literally fizzy drink), though I’ve just googled that and it is a brand name. According to Duolingo the American-English translation is soda but for me I would say fizzy drink, or can of pop. However, my absolutely favourite word so far is winkelwagon (supermarket trolley). Every time I learn a new word which charms me, everyone has to hear about it and I go on and on about the words and I won’t even get into the conjugation of can you, can I, can they, and how that made me laugh.
Ik spreek Geordie-Dutch
I love the way my husband in Dutch is not only my man, in the same way that my neighbour is my neighbour-husband or neighbour-man or neighbour-wife or neighbour-woman (buurman and buurvrowe) but also my spouse echtenoot, which to me sounds like an astronaut.
And, I love the way that hailing as I do from the north east of England, when ik gaan naar de tuin which is I go into the yard, it sounds completely like Geordie-English and I speak it exactly like that.
There is something endlessly fascinating about learning Dutch and I absolutely love it. Even so, I cannot believe that I have arrived on Day 525 without missing a day, which is nuts. I thought that I would do it for few weeks and then swap to something else or give up but no, here I am still going strong. I can manage quite a few words, and I can actually read some of the Horta Museum email in Dutch, though the other day I got confused and was thinking: This is brilliant I am so good until I realised I was reading the French bit not the Dutch. Honestly!
Sometimes, I make mistakes because I haven’t translated from English-English to American-English and I lose my love hearts as I don’t say the words they are expecting and I have to wait for new love hearts. I did buy the slimmest Dutch grammar book I could find in Waterstones because as I was still unsure about when I should be conjugating certain past participles as they seem to have a few like sometimes you put I am or I have and then ge– in front of a verb itself to say I have run or I ran… or at times it seems literally to be Ik heb/I have [rest of sentence] aan/at the [verb] and I am not sure quite why.
The grammar book was really boring and quite involved, so I’ve put it on a shelf until I need it. However, a google about tells me that the main criticism people have of Duolingo is its difficulty in explaining grammar rules. But, for me, grammar rules are boring and if it had been explaining them to me, I would have given up by now.
For the longest while I was wondering how Duolingo worked, and as I have been preparing a talk for the Women’s Institute Learning Hub next Wednesday, which is all about how McDonald’s withdrew its AI from the drive-thru last year after it went ‘rogue’ and started Heston Blumenthal-style putting bacon on ice-cream and giving side orders of 200 chicken Mcnuggets, I looked for and found a lovely article about it here: How Duolingo’s AI Learns What You Need to Learn.
Click on the picture below to sign up and hear all about USA drive-thru history, The Two Ronnies’ sketch about bacon ice-cream, and strange but true stories about having a Middlesbrough accent and the interesting situations I have ended up in talking to people and speech recognition software. I’ll also describe exactly how natural machine learning and natural language processing work.
But for those who can’t wait:
How Duolingo’s Birdbrain actually works
In a nutshell, Duolingo calls its AI Birdbrain which performs natural language processing for machine translation using a long short-term memory model in a deep learning neural network. So, let’s break that down:
Natural language processing is made up of speech recognition or automated speech recognition (ASR), not voice recognition which is used for identification purposes. For example, telebanking at the HSBC uses voice recognition, so when I call the number, it will ask me to repeat the sentence: My voice is my password, as part of the security checks. My unique voice for access to my unique bank account.
Basically, speech recognition is an input method for everyone accessing all parts of a system and natural language processing does the rest. In the case of Duolingo it goes off and does machine translation in my case it does it from Dutch to English and vice-versa.
This includes using an artificial neural network (ANN) which is modelled on the idea of the human brain looking at a discrete word and then that word’s place in a compositional sentence creating a Gestalt effect of the sum is greater than its parts. The ANN steps through a sentence or indeed a whole text and find patterns in it lots of times.
When setting up the ANN humans tell it what it needs to look for by labelling the input data in a supervised-learning approach. Once it comes out the other side of the ANN, it feeds back the results in a process called backpropagation, and then feeds that forward to create a language model which compiles a table (or model or hierarchy) of the text based on what it saw. For example, counting how many times certain words appear and if they are important or not. ChatGPT is, of course. the most famous of large language models (LLMs) used for generative AI and has a language model of 50,000 words (or tokens). Language models need many layers in an ANN which we call ‘deep learning’, so many people now use deep learning and neural networks interchangeably.
Birdbrain uses a long short-term memory (LTSM) which has a special memory cell of three states: input, output and forget so it can ‘decide’ based on how many times words appear. Birdbrain represents a learner’s history of Duolingo exercises into a set of 40 numbers, or a 40-dimensional vector, and every time the learner completes another exercise, Birdbrain updates the vector based on its prior state, the exercise that the learner has completed, and whether they got it right. This vector represents a learner’s ability, which the model uses to make predictions about how they will perform on future exercises, and choose say whether to give that learner fewer past-tense exercises because they do them well and more future tense practice which they do not do so well instead. This allows for more personalization in the lessons for each Duolingo learner.
It is super impressive and for me using this technology is way more fun than if I was in a class having to learn at the same pace as everyone else as I learn languages slowly, I am not great at hitting the books and will lose interest if I get left behind. I learnt French by watching a lot of TV and Italian (which I am still terrible at and can’t seem to improve) by sharing an office with an Italian bloke.
Boro speak
For me the speech recognition is the hardest part, given that I have a Middlesbrough accent and a lot of my vowels are flat but, I have found if I keep getting a sentence wrong, I can sing the sounds of the words, in that I might not get them right, but I get where they start and stop correctly in the wave form at a good enough level for Duolingo to let me through as I am not a typical American mid-west person on whom the system has been designed, and I never will be, no matter how many waveforms they store to pattern match my voice on. I mean how many people from Middlesbrough are learning Dutch on Duolingo?
And even so, I can barely get into my own bank account which is matching my voice with my voice and that’s on a good day, so no wonder if I am tired on Duolingo I can barely get through a lesson without losing all my hearts.
I still have a lot of Dutch to learn but I love it. I sometimes wonder if I will ever get to the end of the learning Duolingo journey and how good will I be, though I am in no rush. I don’t want the fun to end, but if it does, then I can hotfoot it back to Belgium or Holland to practice, all the while saying:
Hallo! Ik spreek een beetje Nederland.
In the meantime, I can’t wait to tell everyone on the WI Learning Hub all the fun words I have learnt. I hope to see you there:
