Chapter 2: Properties of Spoken Language Production

Speakers access word representations based on meaning first, then assemble sounds.
Support from speech errors and experiments analyzing the time course of word production.
- Semantically related distractor words slow object naming when presented simultaneously or up to 400 ms before the object.
- When presented after the object, they have no impact or result in faster latencies.
Phonologically related distractor words sometimes have no effect when they precede object presentation.
Phonologically related distractor words facilitate naming when presented simultaneously or after the objects.
Phonologically related distractor words facilitate naming when presented as early as 300 ms before objects.
Speakers produce words by processing meaning-level properties first, then sound-level properties.
Two stages in word production:
- Word-level
- Sound-level representations
Meaning-level representations make lexical-level representations available, which then provide access to individual sound/segment representations.
It is not possible to go from meaning to sounds except through a lexical representation.

Intention for producing a word leads to the activation of a family of words sharing some aspect of the intended meaning.
Two theoretical assumptions:
- Decompositional views
- Non-decompositional views

Primitives of semantic representations are smaller than the words they support.
Played an important role in the development of connectionist models of word processing.

Representational bases of words and their meanings bear a one-to-one relationship.
Atomic meaning representations are called “lexical concepts”.
WEAVER++ model adopted non-decompositional views.

When a word is intended, other words with similar meanings in the given context become available and compete for selection.
These words share:
- Grammatical class
- Taxonomic category
- Level of specificity with the intended word.
Speakers take longer to label objects, actions, or colors in the presence of semantically related distractors relative to unrelated distractors.
Example: “Dog” and “Bone” do not compete for selection because they are related in meaning but their meanings are not similar.
Competition is restricted to words that express similar meanings rather than simply related meanings.
"Level of activation" of words affects the speed and/or likelihood that a speaker will select another word. The more activated one word is, the more it prevents or decreases the activation of other words.
Semantically related distractors interfere more than unrelated ones because they are activated more strongly and so more formidably compete for selection.

Unrelated distractor words do not receive the 2nd source of activation.
Example: Naming a picture of a “lion”.
- The lexical representation of “tiger” would be activated by:
1. The distractor word “tiger” itself.
2. The semantic representation of LION.
- The lexical representation of “table” would only be activated by the distractor word “table” itself.
It takes more time for speakers to generate the names of objects that have multiple context-appropriate names than those that have a single dominant name.

Being semantically related is important for selection, but the competition is limited to words of the same grammatical class (nouns substitute for nouns, verbs for verbs, etc.).
This is also valid for distractor words.
In substituting phonologically similar words, grammatical category is maintained.
Constraints on maintaining the grammatical class of an intended word appear stronger than the constraints on expressing the intended meaning.
The selection mechanism is blind to category mismatches (a noun can only fill a place, which is suitable for a noun).

The concreteness or imageability of a word is a factor.
- Semantically rich or “heavy” verbs rely on semantic cues more than on syntactic cues.
- Semantically impoverished “light” verbs rely more on syntactic cues.
Words with concrete, imageable meanings have richer semantic representations that guide word selection more efficiently than words with abstract meanings.

Highly imageable words are also more prone to substitution by semantically related words.
Sentence context clearly affects the speed of word selection. More predictable words cause fewer pauses during production.

The properties, which make particular word meanings “imageable” or “concrete” also provide these meanings with additional features and thus richer meanings.
These additional features should increase the activation levels of consistent word representations.

These might take a similar stance, but rather than proposing that “imageable” and “concrete” words meanings have richer meanings, they might propose that those meanings participate in more richly interconnected meaning networks, which might selectively promote the activation of imageable or concrete word-meanings.