Modelling Derivational Morphology
In the 90s I was even in conversation with Harold Greenlee about putting his work online. There are numerous problems with this kind of work, though. The first is just mistakes and dubious connections. John Lee’s 2013 paper Etymological Follies: Three Recent Lexicons of the New Testament gives numerous examples. Lee is always worth listening to when it comes to lexicons!
There’s another major issue which is that expressing etymology (or even just cognate groupings) doesn’t really tell you what I actually care about which is how easy is the meaning of a lexical item to learn based on other cognate lexical items you’ve learned. I’ve previously talked about modelling the cost of learning a new form in the context of inflectional morphology but I’m also interested (as mentioned in various “New Kind of Graded Reader” presentations) in the derivational equivalent between lexemes. There’s some interesting theoretical work in this area going back to at least Jackendoff’s 1975 paper Morphological and Semantic Regularities in the Lexicon. This was picked up in Bochner’s 1993 book Simplicity in Generative Morphology which was a huge influence on me in thinking about morphology as paradigmatic relationships between words rather than morpheme-based approaches.
So for my purposes, at least, I want to model how easy it is to work out the meaning of a word from known cognates potentially given similar analogical pairs of cognates. What I’d ultimately like to develop is some sort of weighting between pairs that represents how transparent the connection in meaning is from their cognate forms.
Take for example the pair
If that pair is known, then something like
is much easier to understand. So if you understand Ἰταλία, Ἰταλικός, and Γαλατία, you can almost certainly take a stab at guessing the meaning of Γαλατικός. I care about that because a big part of my research is modelling how “easy” a passage might be for a student to read.
The analogy might be abstracted as
but it also applies to things like
which is -ος:-ικος so first/second declension doesn’t matter.
Given a new place, you could probably easily construct a plausible denominal adjective for someone from that place with -ικος. A Greek speaker unfamiliar with the philosophical school would still immediately recognize Στοϊκός as suggesting “someone from the στοά” although we might want to score the transparency of that lower that those based on geographical proper nouns.
But now consider
The meaning of the root clearly transfers to the lexical items in each pair but the relationship between the items in each pair is a little less transparent. It’s still there if you think about it but it almost certainly needs to be weighted less. κοινωνία and εἰρήνη are not physical places. The -ικος derivative is still in some sense about something coming from somewhere but rather than a person from a place, it seems to be a state coming from another state (metaphorical place).
Then you get something like
If you think really really hard about it you can see how ὀνικός (in the sense of millstone) might have come from ὄνος (donkey). But this is at best a potentially useful mnemonic for learners rather than a productive derivation. It should be weighted even lower (no pun intended). And then where might
fit in this weighting? (and to what extent do English cognates help too in cases such as this?)
I’m not yet sure how best to produce weightings for this kind of lexical relatedness. My guess is a first pass could be achieved by crowdsourcing on oxlos. Ultimately, some of the weighting could be calculated via regression based on vocabulary quizzes (although I worry about confounding factors unless the students are beginners). Even just doing the crowdsourcing would be interesting to see how much agreement there was in the “obvious relatedness” ordering of pairs like Πόντος:Ποντικός > στοά:Στοϊκός > κοινωνία:κοινωνικός > ὄνος:ὀνικός.
Finally, it occurs to me this gives a potential measure of “false friendship” amongst cognates as a mismatch between the obviousness of relatedness in form vs in meaning.
I have some old work at https://github.com/morphgnt/morphological-lexicon/tree/master/projects/derivational_morphology which I probably need to clean up at some point for all this.
As is often the case, this blog post was triggered by Jonathan Robie asking me something and me realising I’d never written up my thoughts on the topic despite having thought about it on and off for a decade :-)