I caught up on Far Side cartoons from the page-a-day calendar my brother got me as a gift. It’s interesting learning how much something you enjoyed as a kid is not only no longer fun can be quite upsetting once you’ve grown up in the culture that produced them – Whee, lynching jokes! You know, for kids!
On the other hand, it’s interesting to interpret things later in ways you never would when you first saw them. Here’s a recent one that jumped out at me (pardon the hasty phone scan):
The punchline is the alien apparently understands English well enough to interpret an offer, but not realizing the “knuckle sandwich” offered is an idiom meaning a threat of violence, eagerly accepts it as an opportunity to make contact.
Why it’s interesting
As simple as this joke is, it’s actually much like the most advanced NLP systems we use today. I can guess that the very best speech recognition available from any of the enterprise vendors or big tech companies would be able to interpret a lot from the offer we infer our alien friend can guess phrasal intents like OfferAccept, or that “why yes thank you” is a synonym for “yes.”
On the other hand, this kind of crawling toward the Uncanny Valley only gets more frustrating as our progress shows how much we fail to get out of truly natural language. We can think of many examples of how our NLP especially for speech misses the forest for the trees.
Why we care
Ray Jackendoff in Patterns in the Mind uses a metaphor of a robot that understands only one language but speaks another to highlight two strange facts about language: Our knowledge of language is individual, but requires world knowledge, and using language requires reciprocity as a matter of cognitive fact.
It’s easy to imagine that robot, and how different the robot is from us makes the image stick. What’s funny to me today is that reading this old comic strip from the 1980s makes me realize as I think about it that we actually have built that robot again and again.
Our very best NLP systems don’t understand anything about language at all. As DNNs and other techniques get more complex and abstract, and processing power continues to expand, we can expect the need for VUI design to continue to expand, hand-crafting the platforms that turn what machines understand into something humans do.
Links
Patterns in the Mind on Google Books: https://books.google.com/books/about/Patterns_in_the_Mind.html?id=keMJX6BOppAC
Let’s have a conversation
Let us know what you think! Write us at: