Linguistics Graduate Student
University of California Berkeley
skatseff@berkeley.edu
| |
Research Interests: language acquisition, phonological categorization, language change MAJOR UPDATE IN PROGRESS.I have to admit that my bookshelf is pretty varied. My chosen profession is linguistics, which means I think it's neat that if you hold any language up to a magnifying glass, you'll find that it shares some characteristics in common with every other language on the planet. Linguists believe that these commonalities spring from the fact that languages are made by people, and people all use the same internal wires to converse about the weather and talk back to their parents. I want to know how these wires get connected in the first place. Just as infants learn the most important features of bottles and blankets without any prior knowledge of green, square, or soft, they also need to learn the important features of the language they hear, be it Portuguese, Pomo, or Tibetan. This process is hard because language comes associated with sounds, sights, and (trickiest of all) meanings. An American automated speech recognizer has been taught the sounds of English before it gets to work, so its job is mainly to ask, "does this bit of speech look most like a p, f, or s?" An American infant, on the other hand, has no idea how many sounds English has, let alone what they are. There are two ways to figure this out. One is by direct observation: you make a camera that can see inside an infant's head and watch firing patterns as they change in the first two years of life. Not only is this cruel, but it's difficult to do. Instead, I'm listening to kids' speech sounds as they are learning to talk and playing their voices back to them to figure out what they think makes an s sound like an s. Child sounds turn out to be surprisingly difficult to categorize, and differ quite a bit from their adult equivalents. more... This problem is really another way of asking what we remember about speech sounds. All of the distinctive components of a given sound must be stored in memory, and most of the things we remember about a sound are potentially useful in identifying it. I've just begun work on a novel way to answer that question, too. Read more about the plan here. I'm also a sucker for any problem involving similarity and classification in linguistics. This goes for language relatedness, speech recognition, and cognitive modeling. |