It was during an activity in section that I realized there existed a cognitive process that enabled the efficient understanding of phonemes. The activity was rather simple. translate a long passage in IPA to plain English. For context, I think it's worth mentioning that I have had zero exposure to linguistics before, and that I was somewhat daunted by how long it would take me to complete the assignment, since it was my first time dealing with IPA hands-on. Yet to my surprise, only the first couple of words were challenging, the rest took only marginal effort.
What I realized there was that as I knew the basis of the sentence, the first couple of words, my mind had a probabilistic guess of what the following words could be, so it became much easier to find my guess written in IPA and calling it out, than to construct new words from the phonemes alone. Now, the implications of this capacity are remarkable, because it means that not only can we learn semantics from sound alone, but we can also process sound faster because we already have an understanding of semantics.
It is extremely fascinating to me that in that sense, language understanding becomes a virtuous circle. Understanding itself drives understanding. The fact that we always have a "guess" lined up for what word is likely to come next allows us to be more efficient in understanding the message efficiently. It is even interesting how learning can take place when both our guess is right, and when our guess is wrong. Take the example of Disney's Frozen, where a song in the movie has lyrics that sing,
"[Hans:] I mean it's crazy...
[Anna:] What?
[Hans:] We finish each other's-
[Anna:] Sandwiches!
[Hans:] That's what I was gonna say!"
Where clearly the song plays with the listener's guess of what word will come next, and even when they surprise the audience, they still allude to the listener's guess with the final, "That's what I was gonna say".
That probabilistic guess is a cognitive tool that enables us to understand and learn much faster than if we had to decrypt every word individually outside of any context. And building up from Alden's post, it may help in allowing the nonlinearity of language to be processable.
Moreover, this subconscious process allows us to reduce the enunciation of very common words, as postulated by Gahl, Jao, and Johnson in a paper titled, "Why reduce? Phonological neighborhood density and phonetic reduction in spontaneous speech" (2011). A 2008 paper by Bell, Breinier, Gregory, Girand and our very own Dan Jurafsky also confirms this process. This means that not only do we have this subconscious tool, but we have already adapted to it in common, everyday speech by giving ourselves the luxury of optimizing for time in the enunciation of popular, predictable words.
With all of this in mind, my question is the following: how does probabilistic predictability contribute to speech, apart from aiding us in the decoding of phonemes and allowing us to reduce pronunciation? What else does this imply?
I think this post highlights the important fact that words are rarely said in isolation of one another. Keeping this in mind helps us to both transcribe sentences from IPA to english but it also happens in our everyday speech. When we speak, the word that we will say next often informs our pronunciation of the first word. Sometimes we shorten or drop certain sounds all together in anticipation of making the sounds for the word that comes next. Words and sounds don't happen in isolation. This is why after decoding some words in the sentence you were more easily able to decode the rest and it is why when we speak the other words and sounds in context can influence the sound of a single word.
ReplyDeleteBravo, Cristian! I could not agree more with this analysis and it resonates with me in particular. In middle school I was known as that kid who "knew every song." In reality, I would just guess the lyrics fairly accurately, and was not embarrassed at the prospect of getting a lyric wrong.
ReplyDeleteThe probabilistic approach to language prediction seems accurate, and likely extends beyond rhymes. My experience suggests that many visual-gestural signals also help us predict languages. For example, I can predict when my dad will say, "I don't know" when I see his heightened brow and he rounds his lips. This is purely a function of my past exposure to his gestures and subsequent speech, and my ability to recall that information.
I very much enjoyed your blog post Cristian. To answer your question, I think that probabilistic predictability is very useful when it comes to learning languages. If I were to say the word 'peach', then my brain would highlight possible connections that have come up in the past (with the most common ones perhaps surfacing the most). I can see this being very useful as a language learner when recalling words is most difficult.
ReplyDeleteTo go on a tangent: considering the role probability has in statistical language models, I find it particularly fascinating how language models assign probabilities to a sequence of words occurring in a sentence together. These techniques are widely used in speech recognition and machine translation, and very much mirror the way our minds process speech - in a deeply probabilistic way. What is most interesting is that computers have much trouble estimating probabilities in sequences of words that the model has never seen before. The human brain, however, does not have this prerequisite. Why do you think? Do our brains have extra semantic knowledge that allows to make connections across words?
Great job on clarifying something that may not always be noticed, but plays a huge part in our communication. I agree with Quentin in thinking that probabilistic predictability is important to learning languages. I think most of us learn our first language at home, and having the knowledge over time to predict certain things when communicating at home makes the learning process faster and smoother.
ReplyDeleteGoing off of what Aaron said about being able to predict his dad's "I don't know", I think this demonstrates the "home" important of this skill. I would extend this from the home to anyone that is communicated with frequently, most likely friends and family, and those at work/school. These associations and predictions are tailored to the individuals we communicate with, in addition to the general speaking predictions (such as in the lyrics to Frozen's "Love Is an Open Door").
To me this implies that we work to remember things about communicating with people we communicate with often in order to make communication more efficient for future use. Although communication can seem pretty smooth and efficient at times, we can't forget how hard it can be sometimes to communicate with someone.
This is a very interesting post. I would like to suggest my opinion on the mechanism of how we perceive “probabilistic predictability”, rather than directly answering to the question you posed.
ReplyDeleteLanguage, as I see it, is defined by a series of meta-relations of the units that we refer to when we comprehend it. That is, isolated sounds that are meaningless by themselves together form a word to acquire a meaning; isolated words congregate (according to syntax) to make a sentence and have a contextual significance; sentences again form a logical sequence to sum up to a more profound thesis.
The second level of the aforementioned steps is what I see as the basis of the predictions we are making. Just as we do not remember individual words by a set of separated sequence of sounds, we remember certain phrases or series of words rather than memorizing a single word out of context. Or at least, we search for such phrases in a given sentence if we believe that the sentence actually has a significant semantics to it. Just as we immediately think of “tive” or “tion” when we see a sequence of letters like “informa”, we think of “sandwiches!” (or “sentences”) after “finish each other’s…”.
One aspect of language related to this discussion that I am interested in is how language is used for propagandistic purpose. It seems to me that our probabilistic analyses of what words typically go together contextually can be exploited to convey some message beyond the words themselves. There is some really interesting work at the intersection of philosophy of language, political science, and communications on how propaganda works (Noam Chomsky being among those who have written on this topic). In many cases, propagandistic materials seems to take advantage of existing associations between words, or works to create new associations. Once we have these associations, only one of the associated words need be said for the other and its features to be recalled, as well. This is a different area of linguistic processing where statistical inference subconsciously plays a role in how we understand language.
ReplyDeleteI think probabilistic predictability extends past isolated sentences or paragraphs. Just think: if you are having a conversation with someone, say, about your Linguistics homework, when they comment about their roommate’s hate for broccoli without any sort of segue, their sentence might very well be properly formed, but you may not understand it at first. This lack of comprehension would not be derived from mispronunciation, but rather from a lack of context. In that sense, I think in addition to casual pronunciation, whole conversations rely on predictability of what comes next based on the context of what has been said.
ReplyDeleteThis is an interesting post but I think something else is at play here. Let's take a look at the numbers. According to the Global Language Monitor, there are 1,025,109 words in the English language. From this treasure trove of words, Dr. Louann Brizendine of UCSF has determined that women speak 13,000 words per day on average. Men, interestingly enough, speak only 7,000 words per day. Given the sheer computational power of the human brain, it becomes exceptionally easy to see that if the average person repeats, thinks and reshuffles the same set of words over and over again, day after day, year after year, it will become second nature to "finish" someone's sentence. This is probably not so much a probabilistic guess than it is a natural reflex based on years of "practicing" and hearing the same words over and over again. It's akin to Derek Jeter hitting a game winning single in his last at bat. On the surface it seemed quite extraordinary. But in the post game interview he chalked it up to good old fashioned, "experience, experience, experience." Having taken so many thousands of swings over his illustrious career, there existed no real cognitive process in that play. He just "knew" when to swing because he'd done it so many times before. I think that's how our brain works with sentences. Given that we use the same set of words every day for decades, it's not cognitive per se; it's just reflex.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteThis discussion of “probabilistic predictability” reminds me of something I’m sure we all experienced. I’d say starting maybe 10 years ago, an article started going viral and making its rounds around the internet. I’m sure we’re all familiar with it. It read something like: “According to a researcher at Cambridge University, it doesn't matter in what order the letters in a word are, the only important thing is that the first and last letter be at the right place” or rather, “Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy, it deosn't mttaer in waht oredr the ltteers in a wrod are, the olny iprmoetnt tihng is taht the frist and lsat ltteer be at the rghit pclae.”
ReplyDeleteNow, this article refers to how we read written English, but I think it’s a perfectly illustrated example of the kind of shortcuts native speakers are able to take with the language. So much of the meaning in our language is communicated through minor details, such as the first letter of a word, duration, inflection, etc. so that one almost never has to correctly pronounce every phoneme in a word to communicate effectively.
This is an interesting premise--I’ve also been wondering how we predict what words will come next in a sentence, or what sounds will come next in a word.
ReplyDeleteI think everyone knows the feeling of fumbling through a word or a sentence. Trying your best to convey the correct sounds, and eventually giving up, saying, “Oh, you know what I mean.” Somehow, the other person is likely to know what you mean. They were able to predict the rest of your word or sentence.
But what about when we’re not listening for meaning? Even if one was speaking in a language entirely unknown to another, certain sounds just seem right, while others don’t. We’re wired to understand a certain procession of noises; hearing some sounds attached to others just doesn’t fit. This makes me think of when languages are obviously made up. I know I’ve picked up books before with complex names and words like “Xryxetuz.” Such words break so many patterns, they’re hard to remember, pronounce--even difficult to read.