Morphology and Lexicography

I think these are a useful collection of readings. The Haspelmath gave a clear introduction to morphology, and it was helpful to see key definitions, distinctions, and methodologies before they were used (sometime implicitly) in the other two readings. The Slobin and Atkins-Levin texts gave me a better sense of what is at stake with all this talk about morphemes and helped me better understand how linguists marshal evidence from a corpus to defend their theses.

We have discussed machine learning models in this class, and so I think it’s especially appropriate to begin by bringing up a connection between the Haspelmath reading and machine learning. In an assignment for an Artificial Intelligence class I am taking, our job was to implement a machine learning algorithm to learn to “segment words,” i.e. given some string with spaces removed (“thisisastring”) the model should output the most likely candidate for the original string (“this is a string”). Sometimes, the input is ambiguous, e.g. “avideonow” could be “a video now” or “avid eon ow.” It turns out the models have trouble with compound lexemes like “firewood,” and part of the problem is that there is a real difference in meaning between the compound lexeme and its constituents. I think this is further evidence that compound lexemes are genuine lexemes and not just word-forms.

What interested me most about the Slobin reading was the appeal to psychological explanations for phrasing. That is, we have to have some way to choose between all the lexically available expressions that have the roughly correct semantics, and often our choice is determined by psychological facts including what details we judge are salient and how we expect our audience to interpret our words. The whole discussion reminded me of the perspective that words are tools of psychological expression. Generative grammarians sometimes claim that grammars are designed to be efficient. One of Slobin’s most important insights is that we exploit language in order to be expressive.

Finally, I would like to make one observations about the Atkins-Levin paper.

An important step in the authors’ argument is showing that although the shake verbs are close in meaning, they exhibit different syntax, which contradicts the naive expectation that “verb meaning determines verb behavior.” Overemphasis on semantics seems to be a common theme in our readings. For instance, the syntax readings pointed out that the folk-linguistic explanation that nouns are “people, places, or things,” is mistaken, and that the part of language is determined by the surrounding environment.