Chapter Four
Words & meaning: From primitives to complex organization
Outline
Introduction
Words & meanings: Separate but linked domains
The study of words
- Word Primitives
- Factors Influencing Word Access and Organization
- Models of Lexical Access
- Serial Search Models
- Parallel Access Models
- Separating Words and Meaning
Meaning
- Philosophical Theories of Meaning
- Alternative Theories: Meaning is in the Public Domain
- Conceptual Primitives
- Feature Theories
- Variations of Feature Theories
- Knowledge-based Approaches
- Conceptual Organization
- Models of Semantic Representation
- A Special Problem for the Mental Lexicon: Lexical Ambiguity
- The Reciprocal and Influential Relationships of Words and Meaning
Summary
Key Concepts From the Book
I. Psycholinguistic inquiry into words (semantics) comprises the study of the relationship between words and their meanings as well as how people organize, process, and access words. The study of semantics has a history dating back to Plato and Democritus. (Page 158)
II. Words and meanings are separate but associated entities. The fact that meanings exist independently of words can be proven in three ways:
A. The translation argument: there are some meanings that can be expressed by a single word in one language, but cannot be translated into
another language.
B. Second, the imperfect mapping illustration: in any language there may be many meanings for a single word and many words to express a
single meaning.
C. Third, the elasticity demonstration: a word's meaning can change when it is found in different contexts. (Page 159-161)
III. The psycholinguistic study of words addresses two major questions:
A. In what form are words stored in our mental lexicon?
B. What factors contribute to the access or retrieval of words? (Page 161)
IV. One hypothesis about word primitives states that each word has its own separate entry in the lexicon (a lexeme). Multimorphemic words are retrieved directly: variants of words (even plurals) have their own separate entries. (Page 162)
V. An alternative view says that words that are composed of smaller parts are represented in the lexicon by their constituent morphemes. When we listen to language, we decompose the words into their morphemes. This second view places an emphasis on cognitive economy: fewer units need to be stored in the lexicon because they can be rearranged to form a large variety of words. (Page 162)
VI. One way to test whether people store words as wholes or as morphemes employs a lexical decision task that measures how long it takes for subjects to distinguish real words from nonsense words. This type of experiment finds that multimorphemic words take longer to process than single morpheme words. Likewise, in speech production tasks, the more morphologically complex the response, the longer the reaction time. (Page 163)
VII. Frequently encountered affixes and compound words may be stored whole as lexemes. However, while semantically transparent compound words such as "gingerbread" may be stored as separate morphemes, semantically opaque compound words such as "butterfly" seem to serve as their own lexemes. Semantic priming tasks have also been used to support a distinction between semantically transparent and semantically opaque compounds. Derivational endings appear to be more firmly attached to root morphemes. Inflectional morphemes are more likely to be added as we speak. (Pages 164-166)
VIII. Multiple studies have shown that we tend to respond to high-frequency words more quickly than to low-frequency ones in lexical decision tasks, naming tasks, and category verification tasks. Words also appear to be stored in different places depending on whether they are concrete or abstract. (Pages 167-168).
IX. Word association experiments have shown three things:
A. Semantically similar words appear to be stored together
B. Subjects are most likely to free associate the completion of a pair (salt triggers pepper; king triggers queen.)
C. Adults are most likely to respond with a word of the same grammatical class as the target.
These three findings reaffirm the notion that two main principles of word organization are meaning and grammatical class. (Page 168)
X. Words appear to be stored based upon their grammatical class: in speech error analysis, nouns are substituted for nouns, verbs for verbs and adjectives for adjectives. Open class (content or concept) words appear to be stored separately from closed class (function) words. (Page 169)
XI. Analysis of the tip-of-the tongue phenomenon has shown that words that sound alike also appear to be stored together. Substitutions of similar sounding words is quite common, especially of words with similar beginnings and endings, as though phonological cues are preserved as access routes within the lexicon. This is sometimes referred to as the bathtub effect. Similar sounding words may be clustered together, and attempts to retrieve one may activate its phonological neighbors. (Page 170)
XII. Models of lexical access attempt to account for the various ways in which we are able to retrieve words from the lexicon. A viable model must explain how the mind can act like a dictionary, a thesaurus, a rhyming book and a grammar book. These models are typically divided into two groups, both of which consider word recognition an automatic process, not subject to conscious examination:
A. A serial search model, such as Forster's autonomous search model, that says that we scan only one lexical entry at a time.
B. A parallel access or direct access model that says perceptual input can activate a word directly. Multiple lexical entries are activated in parallel.
(Pages 170-171)
XIII. According to Forster's autonomous search model, the lexicon is like a library. A word, like a book, can be in only one place, but its location can be determined from several catalog entries. These catalogs consist of three major access files: orthographic, phonological and semantic/syntactic. Once the word is located through one of these routes, it must be checked against the input word in a post-access check. If the input word matches the word in the mental lexicon, the search is halted. If the input word does not match, the search begins again. Only one
access route can be used at a time. (Pages 171-172)
XIV. Becker's verification model is similar to Forster's. Becker accounts for priming effects by emphasizing semantic connections between associated words, positing that once a word is accessed, the system generates a list of what may come next. Glanzer and Ehrenreich's model posits two dictionaries; one large and unabridged, another small and containing only high-frequency words. (Pages 171-173)
XV. Parallel processing models of lexical access, such as Morton's logogen model, propose that words are activated to a certain threshold through all available input. According to this model, each word has a logogen of attributes that must be matched to the target word. Any logogen with enough of the same attributes as the target word is activated, and the logogen with closest match is chosen. The logogen model accounts for frequency effects by saying that high frequency words have a lower activation threshold than low frequency words. (Pages 173-175)
XVI. Morton's logogen model was the most influential of parallel word access models and served as the basis for parallel models that followed. Later versions have attempted better to account for such things as how the linguistic system responds to nonwords. (Page 175)
XVII. Connectionist models consider the process of word retrieval to be analogous to the neural structure of the brain. These models are similar to logogen models in that they accept all incoming stimuli, which is said to excite connections between nodes. Nodes are of three types: input nodes, output nodes, and hidden nodes. Connections between nodes can be either excitatory or inhibitory. When enough nodes are activated, the word is retrieved. Frequently used words have stronger connections than infrequently used words. (Page 175-177)
XVIII. According to the cohort model, which deals only with the processing of auditory language, when we hear a word, all of its phonological neighbors are activated. The correct word is chosen by eliminating words that do not match the input stimulus, either because of more incoming phonological data or because of the context of the spoken sentence. (Pages 177-179)
XIX. When we speak about the meaning of a word, we are talking about both the word's intension and the word's extension. The intension is the basic concept that the term implies: one intension of the word chair is an object to sit upon. The extension of the word chair is the category of object that could reasonably be termed chair, including armchairs, kitchen chairs, reclining chairs, etc. (Pages 181-182)
XX. The reference theory of meaning says that the meaning of a term is what it refers to in the real world. Two problems arise out of a strict adherence to this view of meaning: 1) it does not explain how two terms can have the same referent and yet have different meanings, and 2) it does not account for words that do not name things (such as function words), nor does it account for abstract terms such as "freedom" or unreal objects such as unicorns. Ideational theory accounts for these things by saying that words denote ideas rather than things. (Pages 182-184)
XXI. An alternative theory states that words have no meaning independently but are based on their connection to other words and sentences within the language. All competent speakers are assumed to use words the same way (conventionality.) (Page 184)
XXII. Just as there are several theories of what constitutes a word primitive, there are also several theories about what constitutes a conceptual primitive, or a feature. A second question is whether concepts have clear boundaries or not. A third question is whether it is sufficient to represent a category as a list of features. (Page 185)
XXIII. Most researchers who subscribe to a feature theory believe in a decompositional view of meaning. Thus, a tree is composed of attributes such as "having branches; grew from seeds," etc. Features can be perceptual, functional, microstructural or societal/conventional. (Pages 185-186)
XXIV. Two main approaches exist among proponents of the feature view. A third approach states that we know more about concepts than what features are associate with them.
A. The classical view claims that for any concept there exists a list of features that are necessary and sufficient to include an object in a particular category. This view considers features the smallest units of meaning.
B. The family resemblance view emphasizes characteristic features, arguing that for most natural concepts, there are not necessary and sufficient features. A term, such as bird or game refers to objects that resemble each other in the same way that members of a family resemble each other. Another tenet of this view is that some instances of a category are more representative than others. Categories have a graded structure: the best example of a category is known as a prototype. Categories are also assumed to have fuzzy boundaries.
C. Knowledge-based views emphasize that categorization and knowledge of concepts is based on something deeper than perceptual features. This approach points out that features themselves are not well-specified, the choice and weighting of features is context- and task-dependent and we know more about the intension of the concept than a list of features suggests. (Pages 186-192)
XXV. Psychological essentialism says that people believe that things have underlying essences that make them what they are. Psychological contextualism says that items can be categorized by the context in which they are found. (Pages 193-196)
XXVI. The hierarchical network model of semantic representation proposes that individual concepts are represented by nodes that are organized in our minds like pyramids. This model emphasizes cognitive economy and states that the further the semantic distance between two concepts, the longer is takes to associate them. The second assumption has to do with category size effects: the larger the category, the more time required for search. This model works best for taxonomic categories such as animals, dwellings, etc. Critics say it fails to account for typicality effects and reverse category size effects. (Pages 197-199)
XXVII. Feature comparison models postulate that concepts are represented as lists of defining features and characteristic features. All features are assumed to be stored under all relevant concepts. Semantic verification tasks are carried out by comparing the number of overlapping features of two or more concepts. (Pages 199-200)
XXVIII. The spreading activation model holds that concepts are represented as nodes connected to related nodes. When a single concept is activated, surrounding concepts are also activated. Closely connected concepts are activated more quickly and more strongly than distantly related concepts. (Pages 200-202)
XXIX. There is no strict, one-to-one mapping between words and meaning: multiple words can supply one meaning and a word can have multiple meanings. Some words are lexically ambiguous. Many jokes hinge on a word having two meanings. Research on the processing of ambiguous words has often used phoneme monitoring tasks in which response times are slower after an ambiguous word than after a non-ambiguous word. (Page 202-203)
XXX. There are two theoretical views about the role of context in influencing the understanding of ambiguous words:
A. The selective access view holds that context biases the interpretation so that only the intended meaning is accessed.
B. The exhaustive access view claims that even with context provided, all meanings of an ambiguous word are activated. The meaning appropriate to the context is accessed first. (Page 204)
XXXI. Cross-modal priming studies show that ambiguous words prime subjects for both dominant and subordinate meanings of words. Balanced ambiguous words are those for which both meanings are equally common. Polarized ambiguous words are those for which one meaning predominates. Studies have shown that dominant meanings are accessed even when the context is strongly biased to the subordinate meaning. (Pages 205-206)
XXXII. Multiple meanings of a word may be activated in parallel, with the dominant meaning occurring first. Context may speed access to several meanings, but it does not restrict access to all interpretations. Lexical ambiguity resolution is a dynamic process with various interpretations racing against each other for access based on frequency of meaning and contextual biases. (Pages 207-208)
XXXIII. The relationship between words as communication symbols and conceptual meaning is a two-way street. Words influence the kind of meanings we can convey and meanings dictate the development of words. (Pages 208-209)
XXXIV. Researchers such as Golinkoff have postulated several key principles that word learners rely on when confronted with novel words:
A. The reference principle -- the assumption that words map onto a thing or an event
B. The object scope principle -- learners look for the whole object, rather than its parts
C. The extendibility principle -- the assumption that a word refers to a class of objects rather than a specific object. These classes can be defined in various ways, thematically, taxonomically, etc.
D. The categorical scope principle -- the extension of words to the same kinds of things
E. The novel-name/nameless category principle -- learners assume a new word refers to an unnamed object in the environment. (Page 209)
