Word Worries

When I talk to friends and colleagues working in education, they tend to do a lot of groaning about AI content.  The reason, of course, is that because so much of the school system involves writing (and writing is the main output of AI generators), the issue of how to prevent plagiarism (and presumably maintain the importance of “real” learning) is occupying a lot of time and attention.

Clearly, there is no benefit to students from turning in assignments made by artificial intelligence other than making them better at asking for the right things from Chat Programs.  Chat AIs constantly evolve based on user input, and every time someone gets found out for using one and then complains to the AI, it gets better at covering its tracks, so to speak.  In my opinion, there isn’t really a “solution” to the problem.  In the same way that math classes now basically recognize that calculators are useful tools in performing many mathematical tasks, eventually academic courses will transition to teaching students how to effectively use AI to help organize, shape, and inform writing rather than generate the content.

All of that is something of a bigger, future-oriented conversation.  What I want to talk about here is the small observation several friends have passed on, which is that unless the AI is given the right parameters it has a very difficult time with the proper use of idioms.  In fact, this is often how papers written by Chat bots are discovered.

An “idiom” is a word or phrase that is used in a specific way by particular groups of people, often in a way that is different from the literal meaning of the words involved.  The word comes from the Greek “idios,” which refers to something that “pertains to oneself.”  The word “idiosyncratic” has to do with personal language usage.  The word “idiot” originally referred to individuals who were unable to communicate outside themselves very easily.  The term “idiopathic” refers to diseases that appear to arise from out of themselves.

In any case, an idiom is basically the same as an “expression, saying, or figure of speech” used in a defined area of by a defined group.  In that sense, it’s similar to “colloquialism,” which is a definition that is limited to a certain area.  

In the United States as a whole, there are plenty of examples of region-defining idioms.  One of the primary examples that is always raised is the way that carbonated, sweetened drinks are referred to.  Idaho is a state that is actually on something of a fault line as far as the issue goes.  The word “pop” is most commonly used in the Midwest, while the word “soda” is more commonly used in the northeast and the west coast.  Both of those areas are responsible for migration to the mountain west—some from the northeast and some from the Midwest.  As a result, both expressions are used here, though by groups with different origins.  My family, which immigrated from the Midwest in the mid-1800s has always used “pop,” while my best childhood friend’s family, which came from New Jersey, always said “soda.”  What we can all agree on is that it’s weird to use the word “Coke” to refer to soda pop, regardless of the specific brand, as they do in the south.  Saying “soda pop” is also weird…

Anyway, the issue of idioms in some ways gets to an issue in language that can be a little hard to wrap your head around, which is the idea of what is “correct.”  In short, the two broad schools of thought on the subject are usually referred to as “prescriptive” and “descriptive.”  The former holds that there are rules, and they should be followed in a systematic way.  One rule, for instance, is that a sentence cannot end with a preposition, because Latin phrases don’t allow for it (the same thing applies to infinitives, which should not be “split” according to prescriptive grammarians).  Winston Churchill famously explained that “ending a sentence with a preposition is something up with which I will not put” in order to emphasize the point.

And that’s the limitation of prescription—it may be “more correct,” but it also may be less effective, as the focus can become the expression rather than the content.  “Descriptive grammar” on the other hand, holds that the way people actually speak is more meaningful than how people “should” speak.  It’s a way of acknowledging that people are going to say what they say, and no amount of force or martinet-like goading is likely to change that.

AI, of course, runs into exactly the problem of trying to split the difference.  In order to sound authoritative, AI tends toward prescriptive expressions.  As a result, it often sounds artificial, which it is.  On the other hand, when trying to sound “real” AI has to try to use idioms accurately, not just within the sentence grammar, but within the location the communication occurs.  

An AI could be in every way convincing here in Southeast Idaho, but if it called a drinking fountain a “bubbler” there’s no way anyone would believe it was a third generation farmer from Downey.  Presumably, AI advances will eventually allow the software to overcome that barrier, but for now we should celebrate the idioms that make us who, and where, we are.  As long as everyone agrees on the difference between a “ditch” and a “barrow pit,” a “highway” and a “freeway,” and which week “last Tuesday” refers to, there shouldn’t be any problem, right?