Abstract
The thesis that the vocabulary of English reflects a highly structured system of concepts is essential to the structural theory of metaphor. I follow those who argue that conventional or common linguistic knowledge (stored in “semantic memory”) is encoded in a network of associated concepts (what I call “the Network”). Section 2 describes the Network. I discuss several large-scale structures in the Network. Section 3 deals with the taxonomic hierarchy of types (in which sortal concepts like [dog] and [mammal] are grouped by subtype-supertype relations). Section 4 discusses the mereological hierarchy of types (in which sortal concepts like [wing] and [bird] are grouped by part-whole relations). Section 5 deals with the taxonomic hierarchy of processes (in which action concepts like [see] and [perceive] are grouped by the is-a-way-of relation — seeing is a way of perceiving). I deal with contrastive structures in section 6. Section 7 talks about symmetries in conceptual structures. Section 8 deals with inference rules (IF-THEN) rules and lexical entailment. Section 9 deals with clusters of densely interrelated concepts (fields). All these structures are crucial for metaphors: the taxonomic hierarchies of types and processes are useful for analogies; contrastive structures and the mereological (i.e. part-whole) hierarchy of types are also useful for working with analogies; inference rules are used to interpret metaphors; conceptual fields are used in every aspect of metaphor processing.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Notes
Two considerations motivate my use of conceptual networks to encode lexical knowledge. First, such networks render lexical relations explicitly; explicit rendering of such relations is needed in order to isolate conceptual fields, to induce analogies, and to generate metaphors. Second, such networks permit rapid and efficient search using spreading activation. I need efficient search mechanisms in order to construct analogies. NETMET encodes its propositional inputs in an internal conceptual network.
The insight that types can be organized into taxonomic hierarchies has been expressed metaphysically in genus-species hierarchies such as the Tree of Porphyry and computationally in object-oriented programming (Khoshafian & Abnous, 1990). Type hierarchies are sometimes construed as set-inclusion hierarchies. For instance, Kintsch construes the “noun system” as a set-inclusion hierarchy (Kintsch, 1972: 266–69). I assume that there is only one taxonomic hierarchy of types in the Network; it has the unique beginner [object]. Believeing that words like “object” or “entity” have little semantic content, Miller (1990: 257) prefers to organize his noun system in WordNet into several distinct taxonomic hierarchies, each with its own beginner. WordNet contains 25 such hierarchies. I put [entity] at the top, and put Miller’s 25 beginners underneath.
The fact that [T] is a supertype of a set of types [{S1, . . . Sn}] is encoded conceptually as [SUPERTYPE(T, {S1, . . . Sn})]. For each [S]] in [{S1, ... Sn}], the conceptual network contains an expression of the form [SUPERTYPE(T, Si)]. If [T is a supertype of S], then [S is a subtype of T]. For each expression of the form [SUPERTYPE(T, Si)], there is a converse expression of the form [SUBTYPE(Si, T)]. The supertype relation between concepts mirrors the hypernymy relation among words (Lyons, 1977: vol. 1, 9.4; Miller, 1990: 245–252); the subtype relation mirrors the hyponymy relation. Types in the taxonomic hierarchy can have many supertypes, so that the hierarchy is a directed acyclic graph. For instance: both [pet] and [mammal] are supertypes of [dog].
Way (1991, 1995) argues that type hierarchies constrain the selection of features and structures shared by source and target; she argues that metaphors add new concepts to type hierarchies. She is surely correct on both points. For instance: the Platonic MEMORY IS A WAX TABLET analogy adds the concept [heat] to the target; the LIGHT IS A WAVE analogy added the [medium] concept (which was later deleted). Thagard (1992) has discussed such modifications to our conceptual hierarchies.
The transitivity of the part and whole relations leads to the distinction between direct and indirect containment. A type W directly contains another type P if and only if W contains P. For example, [head] directly contains [eyes]. A type W indirectly contains P if and only if either (1) W directly contains P or (2) there is some Z such that W indirectly contains Z and Z directly contains P. For example, [head] directly contains [eyes], and [eyes] directly contain [pupils], so [head] indirectly contains [pupils]. The distinction between direct containment and indirect containment is the basis for the distinction between the parts of W and all parts of W. PartsOf(W) is all concepts denoting parts directly contained by W. AllPartsOf(W) is all concepts denoting parts indirectly contained by W.
I suppose that the mereological hierarchy has a unique beginner, the top concept [everything] or [reality]. Each thing is part of reality. But there are logical problems with that, so I don’t press it. Although it might appear that a part can have only one whole, it can have many. According to Kintsch (1972: 271): “one would want to have joint and limb both as parts of leg and arm”. So, like the taxonomic hierarchy, the mereological hierarchy is a directed acyclic graph. The [contains] or [is-a-whole-of] relation between concepts mirrors the holonymy relation among nouns (Lyons, 1977: vol. 1, 9.8; Miller, 1990: 255); the conceptual [is-a-part-of] relation mirrors the lexical meronymy relation (Lyons, 1977: vol. 1, 9.8; Miller, 1990: 255–257).
The [is-a-way-of] relation is the troponymy relation (Fellbaum, 1990).
Each concept in any contrast set must belong to the same lexical category (e.g. they must all be noun-concepts, verb-concepts, adjective-concepts).
Fellbaum (1990) distinguishes four kinds of lexical entailments among verbs: (1) troponymy; (2) presupposition; (3) backwards presupposition; (4) causality. I add a fifth type of lexical entailment: (5) converseness (Lyons, 1977: vol. 1, pp. 279–280).
The degree to which words have something to do with one another can also be quantified by applying clustering techniques to large text corpora (Brown et al., 1992). For every occurrence of a word in a very large text corpora (e.g. every page the New York Times has every printed), Brown looked at a “window” of 1001 words centered on that occurrence (i.e. the 500 previous and 500 next words). Brown looked at the probability that any other word in the corpora occurred within that window, and used these probabilities to determine the semantic stickiness of pairs of words. Brown’s semantic stickiness is a formalization of the notion of “having a lot to do with one another”. Semantic stickiness is used to form topical clusters of words: conceptual fields.
NETMET allows clusters to be defined both explicitly and to be created on the fly by clustering algorithms. Since NETMET’s clustering algorithms are weak, I usually explicitly divide the input to NETMET into distinct fields.
Topic-based conceptual fields (whether scripts or schemata) may be further classified into functional fields and structural fields. Functional fields are organized on the basis of dense interactions; structural fields on the basis of part-whole relations. The functional field of the [atom] contains information about what the atom does. The atom interacts with other atoms to form molecules. Using the functional field of the [atom] as a target description results in the analogies AN ATOM IS A HUNGRY ANIMAL and AN ATOM IS A LUSTFUL ANIMAL. (Think of flourine’s desire for an electron; flourine is lustful or hungry). The structural field of the [atom] contains the atom along with all its parts (down to some level of detail). Using the structural field of the atom as a target description results in analogies like AN ATOM IS A SOLAR-SYSTEM and AN ATOM IS AN ONION.
In Wordsworth’s poem “On the Extinction of the Venetian Republic,” the source description is WOMAN and the target description is VENICE; the metaphor is VENICE IS A WOMAN. Kittay shows that the source field WOMAN is a chronologically organized structure with eight stages: [{ birth, childhood, maidenhood, marriage, childbearing, old-age, death, mourning}]. The target field is also historically ordered. A history is “a set of narrative events, or a narrative description of events, arranged as a chronological ordering.” (Kittay, p. 260). According to Kittay, the chronological structure of these fields plays an important role in the metaphor because “Wordsworth imposes the stages of the life-cycle on the chronology of events in the history of Venice.” (Kittay, p. 260)
In the SOCRATES IS A MIDWIFE metaphor from Plato’s Theaetetus, the target is INTELLECTUAL CREATIVITY, and the source field is HUMAN REPRODUCTION. Kittay has shown how the source HUMAN REPRODUCTION is a chronologically organized structure with seven stages: [{matchmaking, conception, pregnancy, labor, delivery, post-partum, child-rearing}]. The target of INTELLECTUAL CREATIVITY is a chronologically organized structure, beginning with the conception of an idea, moving through the long and difficult process of its expression, and ending with the Socratic analysis of the expressed idea. As in the VENICE IS A WOMAN metaphor, the stages of HUMAN REPRODUCTION are precisely correlated with the stages of INTELLECTUAL CREATIVITY.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2001 Springer Science+Business Media Dordrecht
About this chapter
Cite this chapter
Steinhart, E.C. (2001). Conceptual Structures. In: The Logic of Metaphor. Synthese Library, vol 299. Springer, Dordrecht. https://doi.org/10.1007/978-94-015-9654-1_3
Download citation
DOI: https://doi.org/10.1007/978-94-015-9654-1_3
Publisher Name: Springer, Dordrecht
Print ISBN: 978-90-481-5712-9
Online ISBN: 978-94-015-9654-1
eBook Packages: Springer Book Archive