The cat ate the keat
April 18, 2011 6:10 PM   Subscribe

What term describes this property of human language?

Ok, maybe this is Linguistics 101, but I can't think of a term for it.

Consider this example: If you change one pixel in a digital picture, you've made a very small change in the overall image: the change in the message is proportional to the change in the medium. If you change one note in a song, you've made a very small change in the song. But if you change one letter in a string of text, you've potentially totally changed (or destroyed) its meaning ("The cat ate the meat" / "The cat ate the seat" / *"The cat ate the keat"). The change in the medium is small, but leads to a large change in the message. What term describes that property of text?

I realize that human readers can gloss over typos, and reconstruct intended meaning when errors are present. And I realize that certain letters within text "matter" more than others ("The cat ate tha meat" is still readable). But it still seems like there's something qualitatively different between certain classes of information (e.g. human-generated text; computer binaries; DNA sequences) and other types (like images, sounds, gestures, etc.), related to their sensitivity to change. Intuitively, that difference appears categorical to me, not merely a matter of degree. But maybe I'm wrong.

There must be a term for that, right? Maybe something from information theory or signal processing?
posted by molybdenum to Writing & Language (17 answers total) 3 users marked this as a favorite
I would suggest that if a word took 5 minutes to listen to, changing one letter wouldn't be as big of a deal. Changing seat to keat is a 25% change. Changing 1 note in a 5 minute song is a very, very small percentage.
posted by Nonsteroidal Anti-Inflammatory Drug at 6:13 PM on April 18, 2011 [1 favorite]

(But you may find something interesting in Gestalt Psychology)
posted by Nonsteroidal Anti-Inflammatory Drug at 6:14 PM on April 18, 2011

posted by elektrotechnicus at 6:18 PM on April 18, 2011

I agree that the scale makes your examples problematic. Anyhow, you can quite often change one random letter in a long semtence without necessarily changing its meaning or readability. (In fact, I just did it in that last sentence.)
posted by hermitosis at 6:18 PM on April 18, 2011 [4 favorites]

I think a useful concept for thinking about this is information entropy, but (natural) language is usually a good example of how an information system can prevent a catastrophic loss.

However, I think there's a flaw in your question as the "total language" we experience (beyond mere sounds) is far more like a picture or a gesture (indeed, a lot more like gesture in some respects) than it is like DNA or computer code.
posted by Jehan at 6:22 PM on April 18, 2011

Edit: I note that the question is more about text than language, so maybe my answer is not so useful.
posted by Jehan at 6:24 PM on April 18, 2011

Also, tiny changes in images can indeed have a great impact. When I found this photo, it had a stain on it that unfortunately landed right on the woman's teeth. It was just a handful of pixels, but it completely altered the way you'd perceive her facial expression (very unflatteringly).

I corrected it in Photoshop, because I wanted to be able to enjoy that confident smile. I did the same for this photo, which had a tiny white speck right over one of the eyes. To a human viewer, this slight change makes a vast difference.
posted by hermitosis at 6:26 PM on April 18, 2011 [2 favorites]

I think the first part of your question is talking about minimal pairs.

In your examples, what matters is that the letter you change is a meaningful sound distinction in English - m vs. s results in meat and seat, two totally different things.

In your example the vs. tha, the sound difference does not create a meaningful semantic difference, so we substitute in the closest thing we think makes sense. There, you might be looking for top-down processing, where we read an entire sentence or 'chunks' of text, not just letter by letter in our brains. When we see "the cat ate.." our attention to grammar rules in English indicates that we'll probably see an article next, so we gloss over the typo.
posted by nakedmolerats at 6:35 PM on April 18, 2011 [1 favorite]

Maybe start with the smallest possible unit of speech/writing that has semantic meaning - or then non-semantic units, phonemes (for speech) and graphenes (for written text). Changing a phoneme, or graphene can result in a change or loss of semantic meanink.

Would you class DNA and binary text as having semantic meaning? Or is the semantic meaning less accessible in these forms, such that altering a unit of DNA or a bit does not change the meaning, as these texts are not "read" in the same way as written text, or we are not trained to easily recognise these changes as representative of something. A change in a unit of DNA or a bit still changes what the entire string represents. Changing one pixel of an image does little to to overall image, but as hermitosis states, can still have an effect on part of the image. Changing one graphene of say, War & Peace, does little to the overall meaning of the whole text, but a lot to the sentence that it occurs in. Readers of a text can easily gloss over errors/typos and still interpret the meaning of the text - there is some sort of built in error redundancy system.

I'm sorry, I'm not really being helpful. Just mulling it all over in my head.
posted by robotot at 6:37 PM on April 18, 2011

This is sort of perpendicular to your question, but the examples you provide (seat/meat/keat) are called minimal pairs in the study of phonology. They are good for provide that two sounds in a language are different phonemes. Seat/meat is relatively boring because /s/ and /m/ are very different acoustically, but when sounds are similar, you can use minimal pairs to find out how many phonemes (meaning-relevant sound categories) there are. You can find all sorts of examples in English where changing between d and t (very close in phonological terms) changes the word:


On the other hand, you'll never find a minimal pair in English where the difference between aspirated and unaspirated /d/ changes the word - those two phones are part of the same English phoneme /d/. In another language, they could (and sometimes do) belong to different phonemes. Note that in your "tha" example, minimal pairs don't mean that *any* phonemic change could and should be perceived as a different word - if we don't have a word, we assimilate it to something similar. But if you were to invent a new delicious pastry and name it a Tha, this would be comprehensible and not too confusing to English speakers.

To answer your actual question, one way of describing this property of language is that it has phonemic structure in addition to phonetic/acoustic content: the sound of your voice has allll sorts of information carried in the acoustic signal, but there is a unit, the phoneme, which is in some sense the smallest size chunk that affects the meaning of a word. (In languages with tone and vowel length contrasts, you also get phoneticians talking about "tonemes" and similar chunks.) In contrast with pictures, there isn't really a single unit, or at least, a pixel isn't it. We perceive changes in structure/meaning when a "part" changes - like hermitosis mentioned - but those parts have to do with the structure of the face or whatever else is pictured, not parts the size of a pixel.

On preview - ah, I see the minimal pairs brigade has already arrived. Carry on.
posted by heyforfour at 6:40 PM on April 18, 2011 [1 favorite]

provide = proving, and someone already gave you the minimal pairs link I botched :)

To substitute, here's a link to more discussion on Hindi aspirated/unaspirated consonants and how to produce them.
posted by heyforfour at 6:42 PM on April 18, 2011

Text is a highly compressed encoded form of language. If you heard someone speaking and mispronouncing a word in a way similar to your examples, you'd very likely still understand what was being said because of all the additional information you'd be receiving: inflection, facial expression, etc. Text, however, requires a lot more effort on your part to decode and interpret, and if part of the text is damaged, it can become very difficult to interpret and correct the error, or even perceive it. By the same token, you may simply skip over it and think you read the correct letter/word, like you may have in hermitosis' semtence above.

To stretch your analogy further, it's like a minor scratch on a vinyl record; there's a pop or a click, and perhaps some of the music is obscured, but the music itself is not changed. Change the wrong few bits in an MP3 file, and there'll be a loud unpleasant noise, and file may even become unplayable.

Caveat: I'm blathering off the top of my head. If it's a term you want, the minimal pairs thing looks pretty good.
posted by Devoidoid at 7:17 PM on April 18, 2011

Thanks everyone for the replies! A few responses...

It's true that humans have really good "error correction" systems that cause us to gloss over typos. But I would argue that whether you notice the typo or not, the meaning of a sentence has been changed (or destroyed) if you alter just one grapheme: we're just really good at (subconsciously) fixing that damage. If that thought experiment doesn't work for you, imagine a computerized text processor doing the same thing: it lacks our correction machinery, so would get hung up on any change at all (which is why computer programming can be so tedious).

The woman's teeth example is an interesting point. Some pixels "matter" more than others, don't they? Maybe the intermediary concept I'm missing is "meaning". Presumably, a black pixel on a model's teeth matters more than the same pixel in the background somewhere because the collection of pixels representing the woman has more "meaning" to us than the background does.

Maybe the difference is that text (and computer software, etc.) is 100% meaning, whereas pictures and music are somewhere in between 0% and 100%...?

Maybe this is related to entropy, as Electrotechnikus suggests, though the connection isn't springing to mind...
posted by molybdenum at 7:29 PM on April 18, 2011

I'd suggest that it is just a difference of scale, not of type.

"The cat ate the meat" has 16 letters. Changing one letter changes about 7% of the sentence, which is a fairly large chunk! Changing 7% of a picture could have a drastic effect (in a group shot, that could give someone a different face for example). Changing 7% of the code in a piece of software would be disastrous.

I think you should compare changing a single pixel in a picture to a single letter in a book. It's a rare case that changing one letter would alter the plot of the entire work, even if it did change a sentence.
posted by auto-correct at 7:42 PM on April 18, 2011

I don't think I have an answer for you, but in order to further the analysis of the problem I would draw your attention to a particular category: pairs of words that both a) are morphologically similar and b) are synonyms. It's easiest for me to come up with examples if we choose to ignore any strict adoption of either American or British english.

So take color/colour for example. The words look very similar, they mean the same thing, and given our assumption above, these are both legitimate words, so their interpretation is not a matter of error correction. I would say that such an example is closely analogous to a pair of photographs that are the same except for some small change that is irrelevant to our interpretation of the photograph.

I could then argue that, because of examples like color/colour, language is not absolutely different from, say, pictures, in that regard. However, language seems to have the interesting feature that such examples are very rare whereas in images they are very common.
posted by Anything at 11:47 PM on April 18, 2011

As elektrotechnicus says it's about the entropy. Specifically, about how compressible data is.

To explore such issues further you may be interested in Coding Theory.
posted by anateus at 2:45 AM on April 19, 2011

I'm reading The Information by James Gleick and in the first chapter he was discussing african drumming messages and how they used redundancy to eliminate ambiguity. So in african drum language your sentence, the cat ate the meat might be expressed as the little furry feline creature who sleeps all day has chowed down and eaten up the meat the tasty flesh of the cow who used to give us milk that white liquid which we sometimes put cocoa powder in.

I think you would love this book.
posted by mearls at 7:31 AM on April 19, 2011

« Older Cheapest TV?   |   I'm about to take my GED test, and it's long... Newer »
This thread is closed to new comments.