Monthly Archives: October 2014

Modeling Lexical Production (Guest Speaker Chris Hammerly, 10/7)

This Wednesday UMD post-baccalaureate researcher Chris Hammerly visited us to talk about lexical production – specifically, gendered words, how we make nouns, and what a word is. The slides are up here

Notes (thank you Brian Morris and Emma Jin):

Modeling Lexical Production:
– trying to represent reality
– use computational and experimental methods

– pertaining to the core elements used to build utterances (not words)

– the act of constructing linguistic output in real time

What’s a word?
– no spaces between words (when speaking)

Analytic (separated)                <——>            Synthetic (all together)

Chinese, Hawaiian                                                      Turkish, Ojibwe

– different languages have different notions of words
– have fun trying to define “word” universally

Words vs. Morphemes (to be elaborated upon next week by one Mr. Brian Morris from the Linguistics Department of Montgomery Blair University…)
– morpheme: smallest unit of meaning
– word: one or more morphemes

1. all words are pre-specified for syntactic category
2. morphological processing happens solely in the lexicon, not in syntax

Failures of Lexicalism
– “curl,” “curly,” and “to curl” are completely separate in the lexicon (because syntactic category is pre-specified)
– words are privileged: some languages do a lot of work in their lexicon, while others do next to none
– gender?

Fix this
– throw out nouns, verbs, and adjectives
– two types of things instead:
– L-morphemes – open class, conceptual bundles
– F-morphemes – closed class, feature bundles
– this way, one root (√CURL) gives multiple words in different categories (curl, curly, to curl)
– morphological processing is in the syntax, making all languages equal
– words and phrases are built through the same process rather than different ones

Late Insertion
– separation between morphosyntactic, phonological, and semantic expression
– one-to-one-to-one relationship between morphemes, vocabulary items, and encyclopedia entries

French has 2 genders
– la souris = √SOURIS + n [+ fem]
– la chienne = √CHIEN + n [+ fem]
– le chien = √CHIEN + n [- fem]

How do people remember the gender of words?
– memorization?
– no, shut up, lexicalism. That’s stupid
– also, French language speakers will for the most part give the same gender to a made up word, without any communication between them
– two avenues for remembering:
– conceptual: male=masculine, female=feminine (but this approach doesn’t make sense for inanimate objects)
– phonological: phonological patterns in the word/sound
– e.g. -tion, -sion, and some other endings are generally feminine, and -on is masculine otherwise
– that particular example holds for 98.2% of nominals ending in -on

Goal: formalize conceptual and phonological “licensing” of gender
– learnable (by babies)
– cognitively viable

– kids use past experience in language to make judgments about a new word’s gender
– Bayesian probability: look this up elsewhere if you’re interested, or take Statistics (take Statistics anyway), the short version is that the probability of something is based on both past evidence and the reliability of that evidence (generally, more trials = more reliable)

Applying Bayes to French
– kids make connections between phonological and conceptual patterns as they learn
– recurring evidence for certain connections over others (e.g. the association of chien with masculine, chienne with feminine, not vice versa) strengthens the connections

– no simple definition of “word”
– lexicon broken into free roots and categorizing heads (noun, verb, etc.)
– acquisition: kids are extremely skeptical of what they hear around them (with Chomsky-esque thoughts of “are they using language properly”) and also experts in Bayesian probability
– joking aside, this is probably somewhat innate and true
– use things learned in acquisition in production