“Unsupervised learning" is a common phrase in Artificial Intelligence (AI), which means that an underlying pattern can be learned directly from the data without human intervention. Dr. Yann LeCun, an AI guru at Facebook, now calls it "self-supervised learning" and pointed out that "most human and animal learning is unsupervised learning" . In this article I will share a story of such learning process when I was a post-doctoral researcher at the Massachusetts Institute of Technology (MIT) in the United States.
The setting of this embarrassing story was the welcome party hosted by the Linguistics department at MIT in 2016. I am a linguist, who studies languages in scientific perspectives, but I was left speechless during that party. It was a very cold day with drizzling rain, but the bad weather did not knock me down. On my way to the party, I was very thrilled about having a chance to meet several big-name scholars. I prepared a lot of ice-breakers for the party, but all of them were research-related topics. When I entered the party, I ran into those big names with a big smile, started off by introducing myself, and then talked non-stop about my studies. I could tell that they gradually lost interest in my talks, followed by their turning to banana bread and cold-brew coffee. Beyond my expectations, in this kind of social events they talked more about weather, vacations, political issues, etc. than academic topics. I had a hard time joining their conversations. My tongue became slurred and my eyes out of focus. I felt like a stray dog wagging her tail to please a new owner who showed little interest in her. I felt ashamed; I am English-majored, holding a Ph.D., and have learned English for more than 20 years, yet I couldn’t chat in English with them fluently. I finally realized that the barrier was the scope of my knowledge. I, as representing the majority of students in Taiwan (at least for those who are about my age), have learned and spoken English in classroom settings for a long time but made too little use of English in daily life. Moreover, TV news in Taiwan reports much more about trivial local news than significant international news. We care less about Brexit but more about an increase in gas price by one cent.
As a linguist, I know that chitchatting is the key to learning a new language. When we chat in a foreign language, our brain searches for proper lexicons for the intended meanings. This search takes some time, leading to speech disfluency. However, every search reinforces the neural connections between lexicons and meanings; the next time you retrieve the same lexicon, it will be a bit faster. The same process applies to other aspects of language usage such as grammatical structure and prosodic information.
Self-monitoring your own speaking is the next important step. One of the mostly accepted theories of language learning is Frank H. Guenther's biological neural network model of speech production, called DIVA model. The model explains that when an error occurs in speech production, the brain can detect it through both auditory perception and somatosensory system in articulators (i.e., tongue, palate, jaw, etc.), and it will measure how far away this wrong production is from the expected correct utterance; the brain will then store this difference and correct it for the next output. This process is self-supervised and the learning effect will retain longer (thus the name "self-supervised learning") than learning from someone else correcting your errors. But how to do it for English learning exactly? You might ask. The short answer is to "speak more." Take memorizing vocabulary as an example. When trying to memorize a new word, I would listen to a good exemplar (you can listen to the pronunciation examples in an on-line dictionary) and then imitate it by repetitively reading the word "out loud" 1000 times. Don't be scared by the number. The duration of a three-syllable word is usually 300 msec; 1000 repetitions only take roughly five minutes. Again, the same trick applies to other aspects, such as memorizing new phrases and sentence patterns. When you prepare yourself with a good size of vocabulary and inventories of phrases and sentences, you can try to do some chitchats and play hard in English-speaking social events.