Language learning: Genetic? Too chaotic for computers?

Gussenhoven’s article relates the functions of the organs of speech to an introduction to the International Phonetic Alphabet (IPA). It sheds additional detail on some of our class activities, discussing pitch, or fundamental frequency, and how laryngealized voice, called “creaky voice,” is produced by “tight vocal folds,” and is used intentionally in some languages and dialects. The article describes how organ structures alter sounds to produce distinct effects in speech. By the end, a reader understands how each IPA sound relates to physiological actions in the body.

Kenstowicz’s piece examines phonetic rules in different languages, exploring how it is that we “don’t know” the explicit rules of language, “but we can tell when someone doesn’t follow them.” It builds on Gussenhoven’s article by assuming a strong foundation in IPA. Kenstowicz discusses allophones, for example, how there are eight acceptable allophones for the letter “t” in English. Starting in section 2.6, the author discusses how simply a single level of representation would cause significant linguistic information, for example, vowel length, to be lost, adding yet more layers of complexity to approaching the representation of language rules.

Gussenhoven’s article provided a foundation for understanding Kenstowicz’s piece, and Kenstowicz’s piece drew me to many big picture questions about language. Firstly, the issue about second level representation made me think about how complicated simulating grammars must be. If we could teach computers to keep track of context, they would be much better than humans at interpreting these rules. That being said, the degree of chaos present in the natural world makes a complete abstract representation of language rules an incredibly difficult task. Consider accents, dialects, languages where stress greatly changes meaning, languages where tone denotes meaning, all of which this article mentioned, and it quickly becomes clear to me that it is probably impossible for a computer to learn all of these rules in a “generative” way. On the other hand, manually “telling” the computer rules we have know will always be incomplete.

Kenstowicz’s conclusion brings up questions about language acquisition and genetics, positing there may be unidentified genetic components to the acquisition of language rules, explaining the thorough acquisition of horribly irregular language rules by children, despite language lacking an “invariant core” of rules. I’m interested in the cutoffs for language instinct development and how language patterns learned early can be uncovered later in life, or be used to shed light on what aspects of language are learned and retained young. I’ve had the experiences of childhood exposure, followed by lack of exposure, followed by reexposure to a language, as well as first exposure to a language in adolescence, as well as continuous study from a young age of a language. Naturally, I’ve also had the experience of learning one native language since birth. In my unscientific personal observations, each experience results in different outcomes as far as retaining/learning native pronunciation and in learning or retaining that “native speaker instinct” for what just sounds right or doesn’t.