Speech Comprehension: Decoding meaning from speech
Dec 29, 2015
Disambiguating Homophones
• All meanings accessed at first, in all parts of speech• (Swinney, 1979)
– Heard: “Rumor had it that, for years, the government building had been plagued with problems. The man was not surprised when he found several spiders, roaches, and other bugs [1] in the [2] corner of the room.”
– Seen: ant or spy or control sew– Task: Lexical decision– Facilitated for both meanings at position [1], but only for the
appropriate meaning at position [2]
• Context disambiguates after lexical retrieval
A few pieces of trivia:
• The average college-educated adult has a speaking vocabulary of 75,000-100,000 words (Oldfield, 1963)
• If I guessed what word you said, and all words in a language were equally probable, my chances of guessing right would be between 0.00001 and 0.000013, but not all words are equally frequent.
Sentence parsing and syntactic ambiguity (a few definitions)
• Sentence parsing: assigning words to appropriate linguistic categories in order to determine the syntactic structure (figuring out who did what to whom)
• Syntactic ambiguity: more than one interpretation given the potential grammatical functions of the individual words
2 kinds of syntactic ambiguity
• Local ambiguity: the sentence is ambiguous to a point– The bus driven past the school stopped
• Standing ambiguity: either reading/parsing of the sentence is acceptable– The old books and magazines were on the beach– I saw the man with the binoculars
The Garden Path Model
• Perform one syntactic analysis, and if it doesn’t work, go back and start again
• 2 principles:– late closure
• Because Jay always jogs a mile …
– minimal attachment (simplest syntactic structure)
– Because Jay always jogs a mile, this seems like a short distance to him
– NOT: Because Jay always jogs, a mile seems like a short distance to him
The Constraint Satisfaction Model
• More than one syntactic analysis of a sentence may be generated, but one is dominant
• If we discover we’ve made a parsing error, we activate an alternative interpretation
Pauses in speech
• Syntax, not CO2, dictates when we pause!
– (constituent boundaries)
• Most pauses come before words of low probability– Gives speaker time to retrieve the word – Warns listener that something unexpected is
coming
• 40-50% of speaking time occupied with pauses
Verbatim Recall?
• Memory for actual surface structure fades quickly
• Memory for propositional content much stronger– The gentleman picked the cat up.– The gentleman picked up the cat.– The gentleman picked the bat up.
• Verbatim recall influenced by the nature of the message– personal criticism recalled fairly accurately
What do we remember?
• Sachs (1967) – subjects listened to paragraph-length stories that contained a critical test sentence, “He sent a letter about it to Galileo, the great Italian scientist”
• 4 conditions:– identical sentence– active/passive change: “A letter about it was sent to Galileo,
the great Italian scientist”– formal change: “He sent Galileo, the great Italian scientist, a
letter about it”– semantic change: “Galileo, the great Italian scientist, sent
him a letter about it”
• Task: Determine Same/different?
Sachs (1967) - Results
• 100% accuracy with no intervening material for all conditions
• 80 syllables later:– Performed at chance on identical sentences– Detected active/passive changes and formal changes with
60-70% accuracy– Detected semantic changes with 85% accuracy
The Moral: we store meaning more accurately than we store structure!
Context effects
• Shadowing (Marslen-Wilson, 1975)– Subjects rarely lag behind– With lags, word recognition within 200 ms of a
word’s onset (in context)
• Gating (Grosjean, 1980)– Played sentences, and then the first __ms of the
final word– What’s the final word?– Gates at 50, 100, 150 ms … until word is
recognized• 175-200 ms with context• 333 ms average out of context
Clausal Processing: breaking up language into bite-sized chunks
• “I was going to take a train to New York, but I decided it would be too heavy.”
• We interpret the first clause before we hear the second!
How do we know?
• Reading – 2 techniques:– Self-paced reading
• Read one word at a time. Press a button to advance
– Eye-tracking• Shine an infrared light on retina, and this shows where
the eye is moving
Self-paced reading: chunk effectsStine (1990)
• “The Chinese, who used to produce kites, used them in order to carry ropes across the rivers”– Long pause on 2nd “used”– Longer pauses on “carry” and “ropes”
• Slow down at the beginning of a new clause to integrate new information with the information from the previous clause(s)