In this post, I would like to explore some of the nuances of language and communication, especially as they arise in everyday speaking. Particular focus is on context, tone, and non-verbal cues that add additional complexity to an already complex situation. I close with some thoughts about how we can ever teach a computer to navigate through this highly complex landscape in hopes of someday having software mimic this behavior.
Learning language was the focus of the last post, where I argued that every human shares in a capacity to distinguish and classify object properties, and that capacity is what allows humans to learn to speak. In the situations examined in that earlier post, the teacher of the new language (be it the traveler’s foreign friend or the newborn’s friends and family) genuinely wanted the student to learn the new language. That is a special context that is rarely available in common social interactions.
Most communication events take place casually, where both communicants expect roughly the same level of maturity and expertise of the other. The nuances then arise from either an assumption made by one party that is only partially, if at all, shared by the other, or from a conscious effort by one party to layer on additional meaning or adapt the plain meaning to something else.
Into the first category fall all the little innocent miscommunications that fill our day-to-day lives. Into the second fall things like half truths, jokes, double entendres, left-handed compliments, sarcasms, hints, lies, and cons. In other words, this category contains all the spicier and more interesting aspects of language. We’ll deal with each category in turn.
First consider the innocent miscommunication. It is best exemplified by an anecdote. Last Friday I was meeting a friend for lunch. He had arranged to pick me up, and most of our communication, after the initial phone call, was done by texting. At 11:42 am he issued the following text:
On my way in 5 min.
I promptly got up and went to wait for him. At 11:56 am, I called him asking where he was, to which he responded “I’ll be there in about 10 minutes”. Needless to say, I was confused. When he arrived, we talked it out. What he meant by ‘on my way’ was that he was leaving in 5 minutes whereas I interpreted it to mean that he would be arriving at my location 5 minutes after the text was sent. Clearly the text meant different things to each of us
Sender: ‘On my way in 5 min’ = ‘I am leaving my location in 5 minutes’
Receiver: ‘On my way in 5 min’ = ‘I will arrive at your location in 5 minutes’
A simple foul-up really, but one that resulted from context and assumption.
Next, consider the much more interesting realm of the intentional addition of meaning to add information or flavor, or to deceive. I don’t have the space or the inclination to cover all aspects, but I will touch upon 3 of the above-mentioned items: hints, left-handed compliments, and sarcasm.
The hint category is, of course, a staple in murder mysteries, and finds frequent expression in the works of Agatha Christie. She actually goes into this point with the words from one her most famous sleuths, Miss Marple. In the ‘Thumbmark of Saint Peter’, one of Miss Marple’s nieces calls for help when her entire village begins to ostracize her, believing her responsible for the sudden death of her husband. In the process of solving the mystery for her niece, Miss Marple resolves to find meaning to the seemingly feverish last words of the dying man. As a preamble to her explanation, she starts by discussing context. To quote:
“Has it ever occurred to you,” the old lady went on, “how much we go by what is called, I believe, the context? There is a place on Dartmoor called Grey Wethers. If you were talking to a farmer there and mentioned Grey Wethers, he would probably conclude that you were speaking of these stone circles, yet it is possible that you might be speaking of the atmosphere; and in the same way, if you were meaning the stone circles, an outsider, hearing a fragment of the conversation, might think you meant the weather. So when we repeat a conversation, we don’t, as a rule, repeat the actual words; we put in some other words that seem to us to mean exactly the same thing.”
This “as a rule” substitution is part and parcel of how we think and express and describe events in the world around us, and it is entirely contextual. It’s behind every hint ever given for a riddle or a problem. The plain meaning of the words in the hint are augmented by the context into which the words belong. It isn’t clear how to describe it, you either get the hint or you don’t.
Next is the idea of a left-handed compliment. When I was in high school, one guy stands out as the king of left-handed compliments. Like most high-schoolers, originality and variety were not high on his list. His standard stock in trade was to walk up to someone in the hall and say, in a fairly high voice “That shirt looks good…on you!” To pull this highly dazzling witticism off, he would exaggerate the pause and then say the last part “on you!” with a clownish emphasis reminiscent of Steve Martin’s “well excuse me!”
Of course, he was versatile enough to change “shirt” to “hair cut” or “pants” or whatever part of the victim’s appearance was most deserving of derision.
Finally consider the related category of sarcasm. Often a standard joke when dealing with smart yet social-awkward robots (I’m thinking Red Dwarf here), there is something altogether funny and yet nearly indescribable in the situation where a statement meant sarcastically is interpreted seriously. I really can’t improve on the following clip from the Big Bang theory – so enjoy
Anyway, I hope that after that short tour of the contextual modes of language it is clear that there is far more to meaning than what is found on the written page or in the dry monotone speaking. Artificial intelligence practitioners have yet to capture even the basic capacity that allows a newborn to learn its own mother tongue, let alone allow a person (Sheldon not withstanding) to be able to understand hints and humor and sarcasm.
And as far as the dangers of the AI demon escaping the pentagram, well, let me say a couple of things. First, if only AI were that capable, my tech support experiences would be a whole lot better. Second, until that day comes, I’ll keep worrying about those human scam artists, snake-oil salesman, and con artists who are a lot less book-learned and a whole lot smarter than the average PhD in computer science.