Broadly, I am interested in computational psycholinguistics and its interfaces. I am intrigued by the innate and incredible abilities that humans have regarding linguistic processing.
In short, I like to ask the following questions: what makes humans so good at understanding challenging creative constructions like rhyme, poetry, and humor? How do models of language fail with these topics? When and where do humans and models intersect, divulge, and why?
More specific projects can be found below:
Phonology in Incremental Processing
Much previous work on incremental processing (in both reading and listening tasks) has focused on the effect of syntactic and semantic information, with more difficult structures in these two domains requiring increased processing relative to their easier counterparts. Currently, I am examining the effect of phonology in incremental processing by performing nonce word experiments. This project is ongoing.
Phenomena such as sarcasm demand listeners interpret implicit messages from explicit messages. However, the exact triggers and messages of sarcasm and its friends are often difficult to determine. Mind rhymes are a previously-unstudied, related phenomenon, where the intended rhyming target is replaced by a new word or phrase. For an example, let's turn to the 1948 song "Shaving Cream" by Benny Bell:
"I have a sad story to tell you./
It may hurt your feelings a bit. /
Last night I walked into my bathroom, /
and stepped in a big pile of _____ ."
What fills in the blank? Our knowledge of context, rhyme scheme, metrical structure, and other linguistic factors suggest that the blank should be filled by shit:
"... and stepped in a big pile of shit!"
However, the titular phrase is sung instead, subverting our expectations:
"... and stepped in a big pile of shaving cream!"
As such, mind rhymes provide an explicit case to understanding what triggers implicit messages and what those implicit measures are. This project attempts to understand what licenses these mind rhymes using experimental and information-theoretic measures. Further, I am interested in the question of how humans and models recover the intended rhyming target upon hearing the subversion and whether we can experimentally manipulate these predictions. This project is ongoing.
Phonesthemes and Humor
The relationship between a group of sounds (a word) and their meaning is often assumed to be arbitrary. Exceptions to this concept of arbitrariness are constructions such as onomatopoeia, whose sound mimics their intended semantics: "woof!", "buzz...", "mooooo."
Phonesthemes are yet another anomaly: they are groupings of sounds that have developed an underlying implied meaning. For example, the "-mp" at the ends of words like "bump", "stomp", and "dump" appears to imply some meaning of "forceful contact". What about words that start with "scr" or "pl"?
I am interested in how these phonesthemes implicate humor, as found in Surreal Speak: stonk, succ, etc. and other odd examples. This project is a halted at the moment , but is a fun collobaration with my wonderful colleague Maksymilian Dąbkowski.