August 04, 2012
How We Make Sense Of Sentences
Got some questions to answer.
After a plane crash, where should the survivors be buried?
If you are considering where the most appropriate burial place should be, you are not alone. Scientists have found that around half the people asked this question, answer it as if they were being asked about the victims not the survivors.
Similarly, when asked "Can a man marry his widow's sister?" most people answer "yes" - effectively answering that it would indeed be possible for a dead man to marry his bereaved wife’s sister.
It is too much work to scan carefully for errors in all the sentences we read and hear all day. Our sentence interpretation circuitry probably does some sort of compare of the sentence against competing meanings and uses some words to influence the meaning assigned to other words. Our minds arrive at interpretations that make definitions assigned to individual words fit into the context of the words around them. So the widow's sister becomes interpreted into something like the dead wife's sister since widow and widower involve someone dying and the man is assumed to be still alive since a question about his intentions is being asked.
EEG scans provide evidence that suggests our brains aren't even slightly noticing errors in sentences.
What makes researchers particularly interested in people’s failure to notice words that actually don’t make sense, so called semantic illusions, is that these illusions challenge traditional models of language processing which assume that we build understanding of a sentence by deeply analysing the meaning of each word in turn.
Instead semantic illusions provide a strong line of evidence that the way we process language is often shallow and incomplete.
Professor Leuthold at University of Glasgow led a study using electroencephalography (EEG) to explore what is happening in our brains when we process sentences containing semantic illusions.
By analysing the patterns of brain activity when volunteers read or listened to sentences containing hard-to-detect semantic anomalies - words that fit the general context even though they do not actually make sense - the researchers found that when a volunteer was tricked by the semantic illusion, their brain had not even noticed the anomalous word.
Semantic illusion experiments I'd like to see: first test a large number of people for IQ and then test their ability to detect semantic illusions. Do smarter people detect semantic illusions at a higher rate? Do some people have very intensive sentence interpretation machinery that enables them to detect semantic illusions at a rate disproportionate for their IQ? Do any subcategories of autistics have enhance ability to detect semantic illusions?
Even if you could reallocate neurons in your brain to give yourself enhanced ability to detect semantic illusions it is not clear to me that's the best way to spend your neurons. If semantic illusions aren't causing you much misunderstanding then assigning more neurons to spatial reasoning, mathematical calculation ability, or larger working memory might make more sense. Personally, I'd ope for larger working memory if I could enhance something in my brain. I want a bigger copy and paste buffer and stacks for putting thoughts onto when I get interrupted.
I noticed the first one immediately, not the second. (But then I've always had some trouble internalizing the vocabulary people use to describe relationships to other people since I come from a very nuclear family).
This particular cognitive flaw isn't limited to language. Then again, I don't know that it's much of a flaw either. Glossing over missteps and other random stray data points to understand deeper meaning and pattern is useful.
We probably do statistical prediction. I knew both immediately because I loved logic puzzles and riddles as a kid and those are standard in the genre.
This is very interesting... In my case, when I heard the word "Survivor" all it conjured up was the general sensations and images of an accident in which people died, not even necessarily a plane crash. From there, it wasn't a jump to question where the dead should be buried. "Victims" and "Survivors" would have conjured up the same exact images, effectively representing the same symbol... The actors they referred to in the situation were of secondary importance.
The second one was much the same. "Someone died... how can we console the living?" Just who it was that died doesn't even seem to be stored in the same word or mental symbol.
In Chinese and Japanese, where they have pictogram-based writing systems, they are able to read much faster because they don't have to transform the words into speech-sounds before comprehending. In fact, where they do hear the words in their head, they actually come AFTER they comprehend the meaning, maybe even coming from comprehension of it. The words themselves seem unimportant, only their symbolic representation in the mind.
This may be an important thing to consider in developing machine-based language comprehension... Consider the recent google AI project that was able to learn how to always identify cats in any youtube video. It was never programmed to search for cats by the researchers, if formulated the "concept" of a cat on its own from a pattern that was repeated often. Maybe the same principle could be used for decoding language?
I don't really see this as a weakness. It's just a quirk of how the human mind streamlines the process of deriving meaning from language.
The human armamentaria, to which language is a relative newcomer, is good enough to survive until reproduction.
That's it. No more is needed.
Our brain handles ignores the vast majority of sensory inputs because it is incapable of handling everything.
My thoughts? - The survivors should be dumped naked in the ocean - when they die - because that's the most efficient and complete form of useful recycling here on earth.
That being said - I have often noticed that I - and others with exceptionally High IQ - like many readers here - are a lot less prone to being tricked by such semantic illusions.
I have also noticed that those that are more susceptible to such illusions are often unable to recognize their own cognitive dissonance on various issues.
(for example my mother-in-law believes gays protesting chick-fil-a are "hateful" even though years ago she boycotted the Levi corporation when they simply offered a more liberal set of work benefits to “domestic partners”)
Imho - then - clearly there is a positive correlation between the ability to think critically and to be able to avoid being tripped up by these particular language tricks.
The reason is simple. Speech is dynamically created and has a very significant error content.
Parsing speech takes this into consideration and does not grind to a halt on encountering errors - it just inores them.
I think this may be a feature, not a bug. Instead of focusing on each word, we look towards the general meaning of the sentence. We get in trouble when presented with these "trick questions." When we encounter real life conversations and don't hear every word we can easily, and in most cases correctly, guess what the speaker meant.
The present king of France is bald - True or False ?
Two students were arguing one day about the grammar in their compositions.
The issue was that Sue, where Jane had had had had had had had had had had had their teacher's approval.
The argument was over grammar, not punctuation. There is however a perfectly correct and acceptable way to punctuate the previous sentence.
Those are actually questions I use as part of a first day of school activity with my 9th grade students. We discuss how our biases, vocabulary knowledge, etc. would affect our understanding of the question, and therefore our answers. I use it to remind students that it's important to read carefully.
I wonder if the same person would process the trick sentence differently when spoken or read. I am pretty sure I catch verbal tricks more easily than written ones, because I tend to read holistically rather than word by word. But I tend to hear exactly what words are spoken.
I suggest you read Aristotle on predication.
It might help you avoid falling for faux-science. Maybe not.
Each one is better than the next.
This fills a much-needed gap.
I thought about it for a long time, and decided that learning language was learning cliches, and how to disassemble and join them. Logic doesn't rule over convention. It's the convention that you learn. Logic is an advanced course in literalmindedness.
I wonder if the same thing is at work when you take words, mix up the letters except for the first and last and people can still comprehend sentences made up of these words.
I'm also curious to see the results using subjects where English isn't their first language. You could test people with various degrees of fluency and see if they miss the semantic illusions and at what level it starts.
Whenever we perceive anything, whether it's visual, auditory, tactile, prioreceptive, whatever- we don't go through a procedural method of semantic or syntactic analysis. What we do is pattern matching. We compare our perceptions to existing patterns stored in memory and from that, we construct meaning by linking together things we already know. That's why we can read garbled text, see faces in the moon, and make sense out of sentences that, if critically analyzed, are revealed as nonsense.
Disclaimer: I have a BS and MA in cognitive psychology, and my thesis was in psycholinguistics. So take this with a grain of salt: ;-)
"The reason is simple. Speech is dynamically created and has a very significant error content.
Parsing speech takes this into consideration and does not grind to a halt on encountering errors - it just inores them."
Was the "inores" a deliberate attempt at irony, or to demonstrate the notion perhaps?
Did they run the same test on non-English speakers? It may be an artifact of grasping English with all its strange rules that forces us to ignore errors and ambiguity. I'd like to see the results of tests run on Japanese, Korean or Chinese readers, to see if the same kinds of mistakes get through.
One of the most famous semantic illusions is a scam:
"Heads I win, tails you lose"
One day, early in December French TV had a prankster
coming up to people in the street and asking:
"This year, New Year's Eve falls on Friday the 13th;
will that be a problem for your going out?"
Every person fell in the trap,
stating proudly that HE (or SHE) was not superstitious,
and that they would still be going out to party…
Fortunately, languages are highly redundant (and people rarely say anything other than pro forma jibber jabber), so the gist of a communication is usually intelligible enough with minimal attention.
Absolute semantic understanding is rarely required for survival, since other tools such as shouting and the use of attention-getting keywords are available in extremis, such as when attempting communicating with teenagers.