On the meaning predictability of novel context-free converted naming units *.
In the field of predictability, the effort of both morphologists and psycholinguists has been primarily concentrated on primary compounds. Little attention has been paid to the meaning predictability of converted naming units. This article discusses some aspects of a theory of predictability which is related to an onomasiological theory of word formation, and points out the specific features of meaning predictability of converted naming units. Theoretical considerations are applied to a sample of non-established converted naming units within experimental research. A method of calculating the predictability rate is proposed. The predictability of various naming units may be compared by means of an objectified predictability rate. The factors affecting the predictability of novel context-free naming units are identified and discussed.
This article discusses the predictability of meaning(s) of novel naming units (1) under the conditions of a context-free interpretation. Predictability is conceived of here as a relative notion, that is to say, it is defined in relation to all the possible meanings of a naming unit, as the degree of probability that a particular meaning of a naming unit, encountered for the first time by a language user, will be picked up in preference to any other possible meanings of that naming unit.
The research into the predictability of isolated novel naming units seems to have been restricted to topics connected with the predictability of primary compounds. Products of other word formation processes have been almost generally excluded from this kind of examination. This holds true of both morphologists and psycholinguists. Their achievements indirectly or directly related to the topic of the present article may be summarized as follows:
1. Observations of generative morphology made about the semantic interpretation of compounds and the related classifications, such as those proposed by Lees (1960), Levi (1978), Li (1971), and Finin (1980), are too general to contribute to the predictability of meanings of novel naming units and do not exhaust all the possible thematic relations. Therefore, they cannot be used as the exclusive basis for the meaning-prediction process. At best, they can be understood as one step in the meaning-prediction process which must be completed with a subtler analysis, possibly in the sense of Murphy's (1988) conceptual "elaboration" of the basic interpretation.
2. More relevant are works explaining how new isolated compounds are processed and/or interpreted. This group includes, among others, Zimmer (1971, 1972), Gleitman and Gleitman (1970), and Downing (1977), who emphasize that meanings reflecting habitual, permanent, and/or fixed relations are more predictable than meanings based on accidental, temporary, context-bound relations.
3. A crucial role for the interpretation process is played by the semantic representation (a series of papers by Coolen et al. [1991, 1993; see also van Jaarsveld et al. 1994]) and by the conceptual level of coinage analysis, based on a respondent's world knowledge, experiences, and pragmatic factors in general (Finin 1980; van Lint 1982; Cohen and Murphy 1984; Meyer 1993). Importantly, however, the interpretation process also makes use of linguistic knowledge (e.g. the knowledge of the availability of thematic relations [Gagne and Shoben 1997; Gagne 2001] in combination with nonlinguistic knowledge [Clark and Clark 1979]).
4. The combination of two concepts constituting a new concept, that is to say, the semantic features of the motivating constituents, must be compatible (van Lint's  "necessary association of features," Zimmer's  "appropriately classificatory relationship," and various variants of the "schema [feature] model," based on the slot-filling principle, such as Allen's  "variable R" principle, Finin's  model, the "concept specialization model" of Cohen and Murphy , and the "selective modification model" of Smith et al. ). The fundamental idea underlying these approaches is the relative salience of particular meaning aspects. In other words, not all semantic features of the motivating words (features of concepts) are equally significant for a coinage (Hampton 1983, 1987). Clark and Clark (1979) call them "predominant features." This is closely related to the idea that the interpretation of novel words is based mainly on prototypical features of objects (Meyer 1993).
5. The interpretation of combined concepts is significantly influenced by thematic relations connecting the combined concepts, in particular, the thematic relations stored with the modifier (Gagne and Shoben 1997; Gagne 2001).
6. The interpretation process can, in some cases, rely on analogy as proposed in various forms of the analogy-based model (Derwing and Skousen 1989; Skousen 1989; Ryder 1994; and also Gagne 2001).
In spite of the growing attention paid to the meaning predictability of novel naming units, some important issues are still left unnoticed, while others are answered in a disputable way. These issues include:
(i) While noun + noun compounds can have several potential meanings representing different relations between the compound constituents (combined concept) this fact is mostly disregarded, and it is only one of them which is usually taken into consideration. The existing models propose various approaches to the interpretation of "conceptual combinations," but do not attempt to identify one or several readings of a combined concept that has (have) the highest chances to be picked up by a language user when encountered for the first time outside context. The reasons for discussing only one particular meaning are mostly absent.
It will be demonstrated that the interpretation of context-free novel coinages in general cannot disregard the mutual INTERPRETATION-CONDITIONING RELATIONS BETWEEN THE MULTIPLICITY OF POSSIBLE READINGS, and that the STRENGTH OF ANY READING DEPENDS ON THE NUMBER AND THE STRENGTH OF THE OTHER COMPETING READINGS.
(ii) The various theories referred to above account for the interpretation of compound words as either relation-based or slot-filling-based (property mapping), or the former relation in some cases and the latter relation in some other cases. Thus, for example, Wisniewski (1996: 428) distinguishes two different strategies for two different readings of box clock, a property mapping for 'square clock' and a relation-linking for 'clock contained in the box'. On the other hand, there are also slot-filling models, such as those proposed by Finin (1980) and Murphy (1988) that combine slot-filling with the representation of relations between the modifier and the head.
Stekauer (2005) assumes that the meaning-predictability process can be put on a common footing for all novel coinages, and that the basic structure that underlies the act of naming as well as the meaning prediction process per se includes PROTOTYPICAL FEATURES of the motivating constituents (the concepts of which are combined into a complex concept).
(iii) None of the models take into consideration the interconnection between the genetic aspect of word FORMATION process and the MEANING INTERPRETATION of its product. In general, the interpretation-oriented theories disregard some factors that should be reflected in any meaning predictability-oriented theory, such as the word formation factor, the relation between the morphological and the semantic structures, the underlying morphological type and word formation type, and the competition between the various possible readings. It will be demonstrated that the INTERRELATION BETWEEN THE WORD-FORMATION AND THE WORD-INTERPRETATION PROCESSES can be beneficial to a theory of meaning predictability.
Primary compounds, which represent the main object of research in this field, represent only one, albeit significant, part of novel naming units, and therefore, it is surprising that hardly any attention has been paid to the intricate issues of the meaning predictability of other types of naming units, including isolated converted naming units. Still, there is at least one important exception to this general orientation, notably Clark and Clark's "When nouns surface as verbs" (1979). Although their article deals with "contextuals," that is, new noun-to-verb conversions which--as it follows from the label--depend heavily for their interpretation on the specific context in which they occur, it presents highly valuable ideas some of which are of general nature, because they
(a) also apply to coinages resulting from other word formation processes (as admitted by Clark and Clark in their concluding remarks); and
(b) are relevant to both context-free and context-dependent interpretation (and, therefore, also to the predictability) of meaning(s). Their innovative denominal verb convention is a case in point (Clark and Clark 1979: 787):
(1) In using an innovative denominal verb sincerely, the speaker means to denote
a. the kind of situation
b. that he has good reason to believe
c. that on this occasion the interpreter can readily compute
e. on the basis of their mutual knowledge
f. in such a way that the parent noun denotes one role in the situation, and the remaining surface arguments of the denominal verb denote other roles in the situation.
Disregarding points (a) and (f), and taking into consideration points (b) through (e), the convention can be translated, for our purpose, in such a way that only those meanings are predictable which are formed by a speaker in view of their meaning predictability, that is, with regard to the SHARED LINGUISTIC AND NONLINGUISTIC KNOWLEDGE of a coiner and an interpreter, and that only those semantic components of the motivating words are of relevance which can be inferred by an interpreter without intensive guesswork.
In addition, Clark and Clark use the term "salience" to refer to those features that are "conspicuously unique, given our mutual knowledge" (Clark and Clark 1979: 787). Thus, the sentence The boy porched the newspaper, including the "contextual" to porch, is interpretable because the interpreter can identify the salient feature of 'porch' (a shelter adjacent to the main door into a house). In addition, (s)he is able to relate it to his/her extralinguistic knowledge of how the newspapers are delivered to subscribers in the U.S.A. From this it follows that the Clarks aptly assign a significant role to "world knowledge," including the more or less generally shared "generic knowledge" (what people know about space and time, the basic physical laws, natural kinds, artifacts and their functions, etc.) and "particular knowledge" differing from individual to individual.
Clark and Clark maintain that denominal verbs mostly rely on generic knowledge about concrete objects. However, this claim seems to be too strong. In general, novel naming units may come into existence in speech communities of various size. A small group of friends is a sufficient speech community to justify the coining of a new naming unit which may fulfill its function within this microcommunity by relying heavily on their group-bound, and therefore, more or less individualized, knowledge and/or experience. To reintroduce my example from Stekauer (1996: 125), if, in such a small group, Peter is notorious for wasting a lot of money by gambling, then the meaning of conversion to Peter in a sentence like I Peter'd all my salary last week is predictable thanks to the mutual knowledge of that microcommunity. By implication, it may be concluded that the interpretation of novel naming units relies on the mutual world knowledge of a speaker and an interpreter, this ranging from particular knowledge of a minor speech community to the generic knowledge of the major speech community.
Another important observation made by Clark and Clark, highly important to any theory of meaning predictability, is the role played by PREDOMINANT FEATURES defined as features which "are more central to the characterization of the category than others" (Clark and Clark 1979: 789). Here, in fact, the authors give support to the theory of prototypes: "A predominant feature of a category is one that tends to hold for most of its members--especially its typical members--but not for members of neighboring categories" (Clark and Clark 1979: 790). For example, a red brick is more central to (or typical of) bricks than gold bricks, wooden bricks, glass bricks, bricks of cheese, etc. Importantly, a category may have more than one predominant feature. The notion of predominant features and their role in predicting the meaning of novel naming units is integrated in our theory described below in the form of "prototypical semes" and their place within a hierarchy of seme levels.
Beard (1995) made important observations concerning the meaning predictability of converted words which, in his model of "lexememorpheme base morphology," are treated as "transpositions." Let us illustrate his approach with the noun [right arrow] verb "transposition." The most common semantic functions in this type of "transposed" words (2) are BECOME(XY) and CAUSE(XYZ). These meanings are bound to the grammatical properties of a verb: "If the output of a verbal transposition is marked [-Transitive], assign it the predicate structure [BECOME(XY)] ..." (Beard 1995: 181) and "[i]f the output of a verbal derivation is marked [+Transitive], assign it the predicate structure [CAUSE([Xy.sub.i][BECOME([Y.sub.i]Z)])] ..." (Beard 1995: 182). Not all "transposed" verbs correspond with this common scheme. A semantic shift may result in more specific meanings that are more difficult to predict:
(2) a. He brushed his coat with his hand.
b. I paddled the canoe with a copy of the New York Times.
c. He combed his hair with his fingers.
These verbs are based on nouns that are lexically instruments. Beard assumes the following:
... since the definition of a noun contains its natural function, all features denoting anything other than the natural function may be ignored when the noun is used in verbal contexts. The predicted meaning of the verbal derivation, then, is the natural function of the noun. Thus, (to) hammer will mean 'pound', (to) brush will mean 'wipe (with the intent of removing)', and so on. Indeed, since the meaning of the derivation is the natural function of the base, no grammatical functional derivation can be involved. (3) (Beard 1995: 183)
And therefore, "[t]he specific output of the derivation is most accurately predicted by the input, that is, the semantic representation of the base" (Beard 1995: 185). As it follows from the phrase "natural function," Beard apparently refers to prototypical features of objects as important reference points for meaning-predictability establishing, as demonstrated below, the most predictable readings.
2. General theoretical background
The theoretical framework for the discussion of the meaning predictability of converted naming units is the onomasiological theory of conversion as CONCEPTUAL RECATEGORIZATION outlined in Stekauer (1996). In Stekauer (1998), it is integrated in a comprehensive model of word formation covering all word formation processes. The model distinguishes the following levels:
(3) extralinguistic level
The point of departure for an onomasiological theory of conversion is the postulate of the four most general conceptual categories of SUBSTANCE, ACTION (with subcategories ACTION PROPER, PROCESS, STATE), QUALITY, and CIRCUMSTANCE. Each naming unit results from an intellectual analysis of an extralinguistic object to be named. Within this analysis, the object is classed within one of the abovementioned conceptual categories. The individual aspects of extralinguistic reality do not, however, exist in isolation; on the contrary, they can be conceived from different points of view and subsequently linguistically expressed in various relationships. These different "angles of reflection" of extralinguistic reality can be cognitively brought into a close relation by re-evaluating the already existing logical spectrum. Thus, the most striking feature of conversion is that it always linguistically expresses a conceptual recategorization of extralinguistic reality.
What is the mechanism of these changes? Individual logical predicates are of different levels of abstraction and generalization and constitute a hierarchy. When a new, dominating, logical predicate is added to such a hierarchy or a former dominating logical predicate is removed, the hierarchy is changed and becomes dominated by a new logical predicate which determines the conceptual category of a new extralinguistic object to be named. The conceptual re-evaluation of extralinguistic reality precedes the linguistic processes proper.
Thus, for example, databank represents a SUBSTANCE. When conceptually recategorized, it becomes an ACTION; experiment expresses a PROCESS--after recategorization it refers to an ACTION PROPER; limit is a CIRCUMSTANCE--after recategorization it is an ACTION; feature is a QUALITY--its recategorization yields an ACTION; insert is an ACTION--when recategorized it becomes a SUBSTANCE; stand belongs to a STATE--when recategorized it becomes a SUBSTANCE; etc.
Within the recategorization process, the semantic structure of the converting naming unit (which maps the supralinguistic conceptual level onto the linguistic level, i.e., the meaning facet of linguistic sign) is taken over by the new converted naming unit. At the onomasiological level, the two naming units are interrelated by a logico-semantic category, for example agent, instrument, result of action, manner of action, location, pattern, etc. At the onomatological level, the morphematic expression of the converting naming unit is taken over by the converted naming unit, and finally, at the phonological level, possible stress modifications take place.
Let us illustrate the point (cf. Figure 1).
[FIGURE 1 OMITTED]
The naming unit milk belongs to the conceptual category of SUBSTANCE. It has its typical hierarchy of logical predicates (from the most general to the most specific one). When the hierarchy within the logical spectrum is changed, the recategorization from SUBSTANCE to ACTION takes place. Thus, the central position within the hierarchy of logical predicates in one of the converted meanings of milk ('to obtain milk from a female mammal') is assumed by a predicate focusing on the actional aspect of the extralinguistic object (see the scheme above). The changed hierarchy within the logical spectrum is then reflected in the hierarchy of semes within the semantic structure of the converted naming unit.
Following this brief account we may proceed to a theory of predictability of novel converted naming units.
3. Predictability of converted naming units
3.1. General principles
It may be postulated that the process of PREDICTING the meaning of a novel, context-free naming unit takes the direction opposite to that represented in Figure 1. The (total or partial) identity of morphematic realization is the first factor determining the prediction process, because it directs an interpreter's focus of attention to the prototypical semes of the meaning of the converting naming unit. Since there are usually several prototypical semes, the meaning prediction process takes the form of a "radiation" of the meanings offered by the converting naming unit. In other words, the meaning prediction process is closely related to the identification of the established meanings that seem to be the best candidates for the conceptual recategorization that underlies conversion, and--what is equally important--they determine the direction of recategorization.
Consequently, the prediction of the meaning(s) of converted naming units appears to be a THREE-STEP PROCESS: the first step, the recognition of identical morphological forms of the converting and the converted naming units, determines the second step of the prediction process; in particular, a language user relates the conceptual categories of the motivating and the motivated units and predicts one or more possible, general semantic relations between them; in other words, (s)he identifies a word formation type. Within the last step, these general relations are further specified.
Thus, for example, the general word formation type [SUBSTANCE.sub.Manner/Pattern]ACTION, identified by my respondents (see below) as the best candidate for the conversion of [boy.sub.N] [right arrow] [boy.sub.V], primarily on the basis of the prototypical semes [+male] [-adult] and [characteristic behavior] is, in the third step, specified (apart from other low-significant proposals of both native and non-native respondents) as 'to act or behave the boys do (immature)' and 'to (try to) look/behave like a boy (clothes, haircut, walk)--of girls'. The two readings rest on the prototypical semes [-adult] and [+male], respectively.
To take another example, there are two central word formation types for [cableway.sub.V], following from the prototypical and specification semes [means of transportation] and [constructable]: [SUBSTANCE.sub.Instrument]ACTION and [SUBSTANCE.sub.Result]ACTION. In the next prediction step, they are specified as 'to travel by cableway' and 'to construct a cableway', respectively.
It should be noted that the bottom-to-top direction of the prediction process is not fully adhered to, because the COGNITIVE PROCESSES (conceptual level), based on a speaker's knowledge and experiences, are interwoven with the prediction process at the onomasiological level.
It follows from the examples above (as will be demonstrated below in the experimental part of this article) that it is the most characteristic, PROTOTYPICAL FEATURES that generally underlie the process of recategorization, and condition the predictability of converted naming units. Thus, for example, [hammer.sub.N] can be converted to the category of verb with the meanings that are intrinsically connected with the basic functions of 'hammer' as delimited by the logical spectrum at the conceptual level of the word formation component, that is, roughly, 'a manual tool for nailing'. Obviously, it would be highly improbable and contrary to the basic principles of productive word formation if the primary conversion process was based on a very general feature(s) of 'hammer', for example, [solid]--as present in the meaning of a tool for drawing circles in sand, or for purposes similar to those of darts. (None of these possibilities can be eliminated, though: in an appropriate context the sentences My children were hammering circles in sand and Though they hammered the target for two hours they hit the bull's eye only once should be perfectly comprehensible to a native speaker.) In any case, these converted meanings are unexpected and unpredictable because they are not motivated by any of the prototypical features of 'hammer'. Instead, they are motivated by too general features that also characterize a number of other objects (that can be equally used for the respective purposes mentioned in the illustrating sentences). By implication, a naming unit standing for any [solid] object of appropriate size might be converted in the same sense ('drawing' and 'hitting', respectively).
Therefore, Stekauer (1996) distinguishes three levels of generalization of semes, with the more specific ones being the most probable candidates for motivating the process of conversion. The following semantic structure was given for the naming unit milk (based on Ondrus et al. 1980):
(4) a. Classification semes: [-proper name], [+concrete], [-countable], [-animate], [-collective], [+material]
b. Identification semes: [+liquid]
c. Specification semes: [+white color], [+sweetish taste], [+from female mammals], [+beverage], [+foodstuff]
As already suggested, since the most general, classification semes identify the referent with other similar objects in a large class of objects they, in principle, do not become motivating elements in the conversion process. Important are the identification and, especially, the specification semes. Thus, the combination of [+from female mammals] and [+liquid] motivates the converted meaning 'to obtain milk from cow', the combination of [+from female mammals] and [+foodstuff] underlies the archaic meaning 'to feed with milk' (a ewe milks her lamb); the combination of [+liquid] and [+foodstuff] motivates the meaning 'to add milk to [coffee]' (to milk the tea), etc.
3.2. Predictability of converted proper names
Stekauer et al. (1997) discusses the semiotics of proper names and their conversion. Some of the ideas are relevant to the present discussion of meaning predictability, and therefore it will be worth summing up the basic ideas from the point of view of the meaning prediction process. In principle, it is claimed that there is no difference between common names and proper names, on the one hand, and the conversion of common names and proper names, on the other. Like common names, proper names, too, can be delimited by a set of logical predicates at the conceptual level (logical spectrum). In the same way as in common names these logical predicates specify the most characteristic features of the object to be named; in this case, it is a person(ality), his/her physical and mental features, behavior, achievements, etc.
In fact, there are two important differences between proper and common names. The first one bears on the fact that while the logical spectrum of any "common object" identifies the prototypical characteristics of a CLASS OF OBJECTS through the process of generalization and prototype identification, in the case of persons as objects of the naming process, this delimitation is INDIVIDUALIZED. The individualization does not mean that the respective semes, mapping the logical spectrum at the semantic level, are what Katz and Fodor (1963) label as "distinguishers." Even at this individualized level, the center of gravity is on the general, characteristic properties, typical behavior features, etc., of a person(ality).
The second difference concerns the fact that the conceptual structure (and hence the semantic structure) of a proper name reflects, by necessity, gradual and/or sudden modifications or changes in the DEVELOPMENT of a particular person and the specific LIFE-SHAPING EVENTS. We all develop and act under the influence of our age-bound cognitive capacity and experiences and our social environment. These changes cannot but find their reflection in the conceptual and the semantic structures. In the case of common names, such changes are possible, though not necessary.
Importantly, the conceptual analysis of proper names is "objective" in the same sense as that of common names. It is independent of individual assessments and positive or negative attitudes of a language user. In this sense, the meaning of a proper name is not what I or you think of, or how I or you perceive the person(ality) named by a proper name. The meaning is determined by the person(ality)'s characteristic features and/or behavior. Consequently, it is these features (corresponding to the prototypical features of common names) that are the best candidates for the process of recategorization (conversion) and, by implication, it is these features that predetermine the meaning prediction process. And, in the same way as with common names, the meaning prediction process is hampered by the possibility of "activation" of more than one characteristic features of a converting name, which leads to the above-mentioned RADIATION OF CONVERTED MEANINGS.
Given the similarities between the conversion of common and proper names, it may be assumed that THERE ARE NO PRINCIPLED DIFFERENCES IN THE MEANING PREDICTION PROCESS either. The principle of world knowledge and experience is also vital to the interpretation of a recategorized proper name. We cannot predict the meaning of, for example, the converted naming unit [display.sub.V] if we do not understand the meaning, function, purpose, etc., of the converting [display.sub.N] (of a computer). Similarly, we cannot predict the meaning of the recategorized naming unit [Havel.sub.V] (a famous personality of the Czechoslovak velvet revolution) if we do not know who he is (was) and what the characteristic features of this personality are (were). The knowledge of and/or the experience with the object named is the fundamental meaning predictability condition for both recategorized common and proper names. By implication, like with common names, the predictability of recategorized proper names is not language-dependent.
Stekauer et al. (1997) reported on an experimental study, one part of which examined the ability of language users to predict the meaning of recategorized proper names. The research included 75 undergraduates studying at the Department of British and American Studies, Faculty of Arts, Presov University. The research had the form of an anonymous test consisting of six parts, one of which was focused on recategorized proper names. The subjects were given the following tasks: "(a) explain in detail the meaning of converted proper names in the following sentences; and (b) suggest their Slovak translation. (4) Both in (a) and (b) avoid using the phrase: 'He did it like...'. Try to be more explicit."
The fact that the recategorized proper names were presented in context does not seem to be of high relevance, because the context could not be of much use if the subjects did not know the personalities named. This can be illustrated by the following examples:
(5) a. He Joseph Hellered the situation in that company.
b. He Ben Johnsoned his promising swimming career.
In the test, ten names were offered; nine of them were well-known international personalities at that time, Joseph Heller, Stephen Spielberg, Ben Johnson (a Canadian 100 m world record holder who was later disqualified due to doping), Richard Nixon, Salieri, Maradona, Havel, Madonna, G. Bruno, and one was a well-known Slovak political satirist, Milan Markovic.
The respondents' answers were evaluated as "correct" (if they reflected the characteristic features of the personalities) and "false." (5) The findings indicated serious gaps in extralinguistic knowledge of our students who frequently "missed the point."
As a result, some of the proposed readings make no sense, others are too general. Thus, for example, readings like 'to organize the meeting like R. Nixon' cannot be accepted as "correct" for either American or a Slovak speech community. People are hardly aware of the way Nixon organized the meetings, with the exception of a microcommunity surrounding Richard Nixon at the time of his presidency, which, obviously, was not the case of our respondents. Rather, Nixon is known for his Watergate affair.
For similar reasons, the readings proposed for [Heller.sub.V], such as, 'to criticize', 'to analyze in detail' are too general, in the same way as 'to be successful', or 'to direct' for S. [Spielberg.sub.V]. There are many novelists who 'criticize', who 'analyze in detail', and there are many directors who are 'successful', and, obviously, each director 'directs'. A very low number of our respondents were able to predict the meaning of [Salieri.sub.V], and quite a lot of them confused the fate of G. Bruno with that of Galilei. The test showed that those subjects who had the necessary world knowledge managed to predict the readings of the recategorized proper names, the rest of them gave no proposal, or fell back on too general interpretations. It is not surprising that there were many "correct" interpretations for [Madonna.sub.V] and [Maradona.sub.V].
It was mentioned above that the recategorization-related prediction process can be represented by three steps. To illustrate this point with recategorized proper names, let us take, for example, Ben Johnson. The knowledge of the object to be interpreted, the fact that he used performance-enhancing drugs, and consequently, spoiled his promising career, enabled our subjects to identify one of the most characteristic "behavioral" features of Johnson in the form of a general word formation type [SUBSTANCE.sub.Manner/Pattern]ACTION. The next step identified the manner of action as, roughly, 'to spoil one's career as a result of doping'.
To summarize our discussion, we argue that the word-formation and the meaning-prediction processes, in terms of recategorized common and proper names, are, in principle, the same; they take place in three steps, and they heavily depend on the world knowledge and experiences of language users.
3.3. Seme levels
Stekauer (2005) proposes a subtler classification of semantic components (semes) compared to that mentioned in Section 3.1. It includes five levels of generalization plus the level of word formation type.
Level 1 and the word formation type level reflect the relevant principles of the onomasiological model of word formation (Stekauer 1998, 2001). In particular, Level 1 includes the four most general conceptual categories capable of covering any object of extralinguistic reality to be named. The word formation type is based on the specification of an onomasiological relation;
--Level 2 is represented by "classification semes";
--Level 3 is represented by "specification semes";
--Level 4 is represented by "prototypical semes," identifying the most characteristic, prototypical features of the named objects;
--Level 5 identifies the idiosyncratic features of the named objects.
The individual levels represent different levels of semantic generalization. Levels 4 and 5 differ in their nature. The attribute "prototypical" used to characterize Level 4 semes indicates that they reflect indispensable and fundamental features. On the other hand, the idiosyncratic nature of Level 5 semes refers to one pole of the "abstract-specific" scale. A very specific seme of an object need not be its indispensable and fundamental one. In principle, Level 5 semes have a highly individualizing effect and reflect an acquired, noninherent, idiosyncratic property which does not belong to the defining, prototypical characteristics of the class of particular objects (examples of this kind of feature include 'a spade used by dog for playing' for a possible naming unit dog spade; 'a book in the shape of a baby' for baby book; 'a hammer a part of which is spherical' for ball hammer; 'to wear a yellow sweater and green trousers' for to tulip, etc.).
While the "generalizing" semes of Level 4 contribute to and condition the meaning predictability of naming units, Level 5 semes have an opposite effect. The distinction between Level 4 and Level 5 corresponds to the sometimes criticized approach taken by Katz and Fodor (1963) who distinguish between "semantic markers" and "distinguishers": "The semantic markers assigned to a lexical item in a dictionary entry are intended to reflect whatever systematic semantic relations hold between that item and the rest of the vocabulary of the language. On the other hand, the distinguishers assigned to a lexical item are intended to reflect what is idiosyncratic about the meaning of that item" (Katz and Fodor 1963: 187). In other words: "The distinction between markers and distinguishers is meant to coincide with the distinction between that part of the meaning of a lexical item which is systematic for the language and that part of the meaning of the item which is not" (Katz and Fodor 1963: 188).
For obvious reasons, it is not possible to give an exhaustive list of all the semantic components of the individual levels. In principle, the hierarchical relation in (6) may be regarded as a pattern relation for "filling in" the individual levels with their semes:
(6) Level 1 Substance Level 2 Animate Level 3 Human Level 4 Male/female Level 5 Adult/middle-aged/adolescent/child/baby or Tall/small or Red-haired/blond(e)/black-haired or Decent person/gentleman/evil-doer/killjoy, etc.
This general theoretical framework will now be completed by description of more specific principles underlying any research into the meaning predictability of novel context-free naming units. It will be shown that there are different factors affecting the PREDICTABILITY RATE, and a method of its calculating will be proposed.
4. Experimental research
The experimental research presented below is based on several postulates:
(a) The predictability of novel context-flee naming units correlates with the ACCEPTABILITY of their meanings to listeners/readers. In other words, it is assumed that language users, when facing a naming unit that they have never heard/read before, prefer the interpretation which, in their view, is most acceptable. The most acceptable reading is therefore considered to be the most predictable of all those readings that come to a speaker's mind. By implication, the degree of acceptability of the individual possible readings (valued as scaled points assigned to the individual readings) may be used as an indicator of the predictability rate of the individual readings of a novel naming unit.
(b) Since there is no clear-cut boundary between acceptable and unacceptable readings, the predictability of meanings of a naming unit is a CLINE. Meaning predictability is determined by both objective and subjective factors. The OBJECTIVE FACTOR is represented by the prototypical features of an object named by the naming unit to be interpreted. These features usually underlie new coinages as motivating semes. Since they represent the cognitive core of a particular naming unit, a new converted unit, motivated by such prototypical semes is, by necessity, more easily interpretable, more acceptable (i.e. tends to be assigned more points by respondents), and therefore has, in general, a higher predictability rate. At the same time, meaning predictability is influenced by a SUBJECTIVE FACTOR: each language user is a unique being who perceives the world and the linguistic signs, functioning as labels for the "objects" of extralinguistic reality, in a unique way (certainly within the limits established by the objective factor). Hence, the evaluation of acceptability, expressed as the number of points assigned to the individual readings, may and really does differ with individual respondents. The interaction between the objective and the subjective factors is reflected in different scores assigned to a particular reading by different respondents, on the one hand, and the overall tendency towards a particular meaning acceptable to the majority of the respondents, on the other.
(c) The meanings of a naming unit that turn out to be the most acceptable ones, and thus ex hypothesi the most predictable, are those that are based on the semantic components representing PROTOTYPICAL FEATURES of objects. On the other hand, the motivation based on semes of Levels 1, 2, 3, and/or 5, or the motivation using a non-established shifted (metaphorical) meaning for this purpose, reduces the meaning predictability of a naming unit. This postulate follows from an assumption that the features that dominate the meaning of a word, and are therefore central to its meaning, are the best candidates for the motivation of converted naming units. Consequently, given the interrelation between the word-formation and the word-interpretation processes, they are also the best candidates for motivating the interpretation of such naming units.
(d) The predictability of any novel naming unit heavily relies on the conceptual level analysis (i.e. supralinguistic level) and the knowledge of listeners/readers (extralinguistic level). Therefore, experimental RESULTS FOR NATIVE SPEAKERS SHOULD NOT SIGNIFICANTLY DIFFER FROM THOSE FOR NON-NATIVE SPEAKERS. This hypothesis was validated for N+N compounds in Stekauer (2005).
(e) Predictability of the meaning(s), that is, of the individual readings of novel context-free conversions, can be calculated. For that purpose, the notion of PREDICTABILITY RATE (PR) is introduced here. It is calculated as follows:
(7) PR = r/[R.sub.max] * p/[P.sub.max]
--r = the number of respondents identifying a particular meaning as acceptable
--[R.sub.max] = the total number of respondents
--p = the sum total of the points assigned to a given meaning by all respondents (on a scale from 1 to 10, where 10 stands for the highest acceptability of the meaning)
--[P.sub.max] = maximum possible number of points assignable by all respondents
It follows from the equation formula that the maximum possible predictability rate value is 1.0 (if a particular meaning is proposed by all respondents and is assigned 10 points by each of them).
(f) For each novel converted naming unit there are one or two central meanings that are most acceptable to language users and hence most predictable.
4.2. Experimental research
This section describes an experiment aimed at determining the meaning predictability of novel context-free conversions. The respondents were given the following written instruction:
(8) The following are possible English words. Propose as many meanings for each of the words as you can think of, and assign scoring to each of the meanings proposed, with 10 points indicating the highest probability of occurrence in the actual language of the meaning proposed, (and therefore predictable best of all) and 1 point the minimum chance of occurrence in a language of such a meaning.
This instruction was followed by an example:
(9) to milk --to get milk from a cow 10 pts. --to put milk into coffee/tea 8 pts. --to drink milk 4 pts. --to paint an object with milk 1 pt., etc.
The example was followed by ten "test" words. The respondents were asked to do the test in writing. Since I could not take part in testing all the respondents for objective reasons (see below) no time limit for responses was specified. The respondents were asked to do the test in their own time and deliver it to me via e-mail or postal mail. In this way, all of them were given equal conditions for completing the test.
The respondents fell within three basic groups. There was a group of 20 native speakers (NS) and 20 Slovak respondents (NNS). Most of the native speakers were contacted by my friends abroad and by my students. A number of the questionnaires were also completed by native speakers that taught at a language school in Kosice, Slovakia. All in all, the major portion of the native speakers were university undergraduates, and a smaller part university graduates. All non-native Slovak respondents were my students at the Department of British and American Studies, attending either the Introduction to English Linguistics or English Lexicology course.
As indicated above, all the respondents of Group 1 (native speakers) and Group 2 (non-native speakers) were presented the sample naming units in a written form. While this method of research was "enforced" by the fact that I could not get into contact with the native speaker respondents, and the necessity to establish identical conditions for both native and non-native respondents, this circumstance should not be assessed negatively. I surmise that the major part of novel naming units are encountered by language users in a written form, in newspapers, professional and popular journals, on the Internet, documentation to inventions, projects, designs, etc., and therefore the conclusions of the research seem to be relevant in terms of the objectives pursued in this research.
In addition, there was another group of 25 non-native speakers, the so-called "check group" of Polish undergraduates (my students at Rzeszow University, Poland, who studied in the English Department) whose task was a little different. Rather than proposing the greatest possible number of readings, they were asked to propose, for each of the possible naming units, only one single reading which they found most acceptable, that is, the best candidate for the integration in the lexicon of existing words. In addition, they did the experiment under time pressure, having only 60 seconds per naming unit. Students were first given oral instructions, then a naming unit was read to them twice and then written on a blackboard. From this point, I started counting 60 seconds. Then I uttered another potential converted naming unit, and the cycle continued through all ten potential words. At the end, I immediately collected the "tests."
The reason for performing the research with the same sample of non-established conversions under two completely different conditions (my presence/absence during testing; proposal of as many readings as possible/single, most acceptable reading; unlimited time/time pressure) was to verify the hypothesis that for each novel converted naming unit there are one or two central meanings which are most acceptable to language users and hence most predictable. If this assumption is true, the results of the two different experiments should be similar.
The results were processed in the form of tables representing the individual readings (6) and their respective scorings assigned by the individual respondents. In order to assess our hypothesis concerning equal predicting capacity of both native and non-native language users, the results obtained from native speakers and non-native speakers were first analyzed separately. Then, they were added up in order to obtain a sufficiently large sample of 40 respondents.
The test included the following possible/non-established conversions devised by myself according to the criteria following from scheme (10). The purpose was to represent, by selected words, a range of different objects of extralinguistic reality. Hence, the first selection level was the opposition animate vs. inanimate. Animate was subdivided into human, animal, and plant; inanimate into tangible and intangible; tangible into solid and liquid; solid into movable and immovable, etc. Each of the classes obtained in this way was "filled" with one non-established converted word.
(10) ANIMATE --HUMAN boy --ANIMAL lion --PLANT tulip INANIMATE --TANGIBLE --SOLID --MOVABLE cableway --IMMOVABLE planet --LIQUID river --INTANGIBLE --PROCESS conference --FORM triangle LOCATION courtyard TIME morning
4.3. Example of analysis
For spatial reasons, the method of analysis is illustrated here with only one converted naming unit, in particular, with the conversion [boy.sub.N[right arrow]V] whose semantic description is as follows:
Level 4--[male] [-adult] [immature] [characteristic behavior]
However, the final evaluation in Section 4.4 and the subsequent sections reflect the data obtained for all sample naming units.
The scores and the PRs of all individual readings proposed by native and non-native speakers are given in Tables 1 and 2, respectively.
Table 3 illustrates a considerable degree of agreement between the two respondent groups in determining the most acceptable readings. This is borne out by the same ranking of the three most predictable readings, small differences between the top PRs in the two groups (0.071), and the similarity of PR gaps between the first two PRs (0.160 vs. 0.149). The rank 4 readings are insignificant because of their very small PR values.
The three top PR values for all 40 respondents are, respectively, 0.358, 0.203, and 0.035. The PR gaps between them are rather big (the PR of the rank 1 reading is almost twice the PR of the rank 2 reading, and more than ten times that of the rank 3 reading), which contributes to the dominant position of the most predictable reading. The check group (25 respondents) results for boy gear well into the picture obtained for the main group, thus confirming both the order of the two most predictable readings and the dominant position of the reading based on the [-ADULT] vs. [+ADULT] contrast (19 "votes" for rank 1 compared to 4 "votes" for rank 2).
The two most predictable readings are based on a word formation type in which 'boy' as SUBSTANCE functions as a pattern indicating the manner of ACTION (i.e. [SUBSTANCE.sub.Manner/Pattern]ACTION). (7) The ACTION refers to the characteristic behavior of 'boy'. This behavior is mostly conceived of as a whole; there are only a few proposals in which a more specific aspect of behavior is salient (for instance, the characteristic way of speaking, dressing, walking, haircut, tendency to attract attention of girls by force and/or by showing off). Since none of these individual features seem to be more "prototypical" than the other features, the activation of any of them usually requires a contextualized situation: for example, knowing Peter very well, we might say that 'Peter likes to boy when he meets a nice girl', thus referring to his inclination to show off.
The two dominating readings emphasize two different facets of opposition into which 'boy' as [-ADULT] and [+MALE] enters. In particular, reference is made to the characteristic way of boys' behavior or acting that makes them different from (a) adults, especially mature adult men, and (b) girls. While the most predictable reading is based on the [+ADULT] vs. [-ADULT] contrast, (8) the second reading is based on the contrast between [+MALE] and [-MALE]. Rather than on boys themselves (first reading) the second reading lays emphasis on imitation by girls of a boy-like behavior. The relatively high PR of this latter reading is not surprising in the age of emancipation of females in the countries of Western civilization, which is manifested in the way of dressing and hairstyle (unisex), in the growing social and economic opportunities of women and also girls, in their growing chances to excel and to become independent.
An important conclusion that can be drawn from the boy conversion is that the process of conversion can hardly be accounted for by sticking to the linguistic data. These must go hand in hand with the listener's/ reader's extralinguistic knowledge. There is nothing in the semantic structure of 'boy' that would reflect, for example, the present-day "unisex" trends in fashion and/or the emancipation trends.
The reading 'to change the sex from female into male' also makes reference to extralinguistic knowledge, in particular, to the developments in medicine, in surgery and in genetic engineering. Its word formation type is [SUBSTANCE.sup.Result]ACTION. The reading, 'to hire someone to carry luggage' is too much context-bound, and therefore, unpredictable. Carrying luggage is not a prototypical feature of 'boy' in its basic cognitive meaning. 'To behave like a mother's little boy' is an interesting modification of the most predictable reading: it is its negation. To behave like a mother's boy means not to behave like a "true" boy. It may be this fact that caused only a single 1-point occurrence of this reading.
Table 4 summarizes the most predictable readings for all of the converted naming units used in the experiment.
4.4.1. Native speakers vs. non-native speakers. Let us first compare the data obtained from native and non-native speakers. The most valuable acknowledgment of hypothesis (d) in Section 4.1, saying that the results for native speakers should not significantly differ from those for non-native speakers, comes from almost identical proposals of most predictable readings for the individual converted naming units (in principle, the differences bear on the number and the range of very low-occurrence, usually single-occurrence, readings), and, even more important, from the almost total agreement in their ranking. Thus, the two groups identified the same rank 1 reading for eight of ten conversions. In one (cableway) of the two exceptions, the difference between rank 1 and rank 2 in the NS group is as low as 0.010; in the other case (triangle), it is 0.042.
Even more impressive is the fact that in eight cases, the two groups agree in the identification of the ranking of two topmost readings. Furthermore, the differences between the two groups in terms of the PR value are insignificant: the average difference for rank 1 readings is as low as 0.087, for rank 2 readings it is 0.077, and for rank 3 readings 0.034. The power of these data is significant enough to bear out hypothesis (d).
In addition, the absence of any major differences in the results of the two groups of respondents is an important piece of evidence for the assumption that both native and non-native speakers make the same use of extralinguistic knowledge and experience. Certainly, differences in culturally-similar communities may arise if an object is named that is unknown and/or not used in one of the speech-communities (a case in point is the reading 'a spade used for scooping up dog's excrement' proposed by a number of native speakers for the interpretation of a possible compound dog spade (Stekauer 2005). Since in Slovakia, there is no tradition of 'scooping up dog's excrements' this interpretation was mostly absent in the group of Slovak respondents. No similar case has occurred in my conversion-oriented experiment.
4.4.2. Meaning predictability and dominant readings. The analysis of the data shows that there is only one case in which the PR is above 0.4 ('to take part in a conference'). Let us take this value (for the practical reasons of data evaluation) as the lower limit of high meaning predictability. Three readings exceed 0.3 (and one another is just .001 below it). These are the top readings for planet, boy, cableway. Let us take this to be the lower limit of medium predictability. The PR of nine other readings is above 0.2 which may be labeled as the minimum level of predictability. The great majority of readings fall below 0.2, but none of the top readings. The lowest PR of a top reading is exactly 0.200 ('to swim/bathe in a river'). What strikes one is that the PRs of the third-rank readings are well below 0.1, the only exception being conference whose rank 3 reading's PR is 0.165.
If the predictability levels are defined as above, that is, as steps per 0.1, then, in seven out of ten experimental conversions, there is one very strong reading clearly dominating the remaining readings of a given conversion (the PR gap more than 0.1). In this respect, the "strongest" predictability position can be attributed to 'to grow tulips'. Its PR is higher by 0.257 than that of the rank 2 reading for tulip. The average PR gap between the rank 1 and rank 2 readings of the ten sample naming units is 0.148, that is, more than one predictability level. The average PR gap between the rank 1 and rank 3 readings is immense: 0.258.
An even clearer picture of the dominant position of the most predictable readings is obtained if the rank 1 : rank 2 and the rank 1 : rank 3 ratios are calculated. The rank 1 PR value is twice (exactly, 1.99 times) higher than that of the rank 2 reading. The gap between the most predictable reading and the third rank reading is abysmal: the PR of the former is 5.79 times higher.
All these statistical data provide an unambiguous support to hypothesis (f) in Section 4.1 of one (rarely two) dominant reading for each novel context-free converted naming unit. This conclusion has far-reaching consequences for meaning predictability. In particular, it may be assumed that the difference between the PR of the top reading and the PRs of the lower-rank readings influences the predictability strength of the most predictable reading. Therefore, this variable should be taken into consideration in evaluating the meaning predictability. This issue is further elaborated in Section 4.5 below.
Let us have a closer look at the readings above the high- and medium-predictability levels.
The word formation type (WFT) underlying the reading 'to take part in a conference' (0.427) is [PROCESS.sup.Purpose]ACTION; the WFT of 'to travel to/land on other planets' (0.376) is [SUBSTANCE.sup.Target/Direction]ACTION; the WFT of 'to act or behave the way boys do (immature)' (0.358) is [SUBSTANCE.sup.Manner/Pattern]ACTION, and finally, that of 'to travel by cableway' is [SUBSTANCE.sup.Instrument]ACTION. The diversity of WFTs underlying these readings might indicate that their "success" is not bound to any specific WFT. Rather, the word formation productivity of each of these WFTs establishes general, but necessary, conditions for the interpretation of the respective naming units in the way they are interpreted.
The relation between WFTs and meaning predictability shows certain tendencies. Thus, converting nouns denoting objects with behavior tend to convert in the manner/pattern mode, that is, the converting unit functions as a pattern and determines the manner of action of the converted naming unit. This is the case of the predictable readings of boy and lion. Conversion of naming units denoting natural objects (planet, river) and other artifacts of relevant size (courtyard) usually rests upon the location/direction mode. Artifacts, in general, are also convenient for the factitive-mode conversion (cableway, triangle) and the instrumental mode (cableway). Events are naturally bound to the purpose mode (conference), and time-related naming units usually convert via the temporal mode (morning). It is only in this sense that Ryder's (1994) templates, designed for the field of primary compounds, may perhaps be of some (rather limited) relevance here.
Making any more conclusive generalizations on the relation between the WFT and the PR is, however, preconditioned by implementation of a large-scale research focused on this particular issue. Such a research should answer several basic questions: Do particular nouns (or better, lexically determined groups of nouns) prefer a particular direction of conversion (i.e., WFT)? If so, is there any correlation between this preference and a particular level of meaning predictability? Is the preference for a particular WFT characteristic only of the most predictable reading of such a group of nouns, or is there any broader pattern identifying a typical WFT for each rank within the group? (9)
For more specific reasons behind the dominance of the above-mentioned readings, let us examine the role played by prototypical semes in the interpretation process. Thus, the reading 'to take part in a conference' is motivated by the Level 4 semes [presentation of information] and [exchange of information] that specify the general semantic category of "purpose." While motivated by this top reading, the rank 2 reading of conference, that is, 'to (meet) to talk about/consult/discuss a subject', is semantically shifted and generalized, and hence not bound to the prototypical features of conference (concerning special way of organization, topic selection, course, etc.). In this respect, it is not motivated by a Level 4 seme; rather, the motivation is bound to a higher level--Level 3. Even if the PR of this reading is above the predictability level, the PR gap between this and the top conference reading is more than significant, that is, 0.215.
The reading 'to travel to/land on other planets' outscores the remaining readings of planet in a most significant way. Its basic semantics expressed by the category of direction is specified further by the motivating Level 4 semes [solid] and [[+ or -]reachable]. The '[+ or -]' mark should be interpreted in a relative and conditional way: while only few planets can be reached by mankind at this level of technological development, many others CAN be reached in human dreams, science-fictions, and long-term visions. In this respect, the "reachability," also including the meaning of 'traveling to and landing on a planet', is both a Level 4 seme and, at the same time, a very strong extralinguistic reading-motivation factor, contributing to the relatively high PR of this reading, especially in the era of closely monitored space flights.
It appears that the considerable PR gap between this and the next reading for planet (highlighting the colonization of planets) appears to bear on the aspect of implementability. Language users, when interpreting the naming unit planet may be guided by the 'closer-to-reality' principle, thus preferring the more realistic option and/or assigning it much higher scores.
The reading 'to act or behave the way boys do (immature)' is motivated by the Level 4 [-adult] and [characteristic behavior] which "detail" the category of pattern/manner. Here, however, one cannot account for the significant gap of 0.155 between this and the next reading 'to (try to) look/behave like a boy (clothes, haircut, walk)--of girls' by a different seme level motivation. The latter reading is also motivated by two Level 4 semes, that is, [male] and [characteristic behavior]. The explanation must be sought in extralinguistic reality, our world knowledge and experience. Out of the two contrasting relations [+adult] vs. [-adult] and [+male] and [-male] the latter still appears to be based on a more solid basis, resting upon much longer traditions and the role of the two genders in society. In spite of what has been mentioned above in connection with the emancipation process, the gender distinctions are perceived to be crucial, including the male-female role opposition in everyday life. Therefore, it may be surmised that while the favorable predictability conditions (productive word formation type, Level 4 seme motivation) helped this reading to rise above the predictability level (0.203) it is clearly outscored by the other reading with equally favorable predictability conditions, further supported by the relevant extralinguistic factors.
The gap between the two top readings of cableway is relatively small. The top reading 'to travel by cableway' corresponds with my expectations: the general semantic category "instrument" is supported by the Level 4 seme [means of transportation]. The second-rank reading 'to build/construct a cableway' is motivated by Level 3 [equipment] which implies the seme [constructable]. However, all [artifacts] (Level 3) are constructable, producible. Therefore, it comes to be no surprise that the reading motivated by this general seme played a relatively important role in the meaning prediction process of the native speaker group.
An important role of world knowledge and experience in the interpretation process can be further illustrated by the dominant position of 'to grow tulips' (0.290). The PR of the next lower reading ('to pick tulips') is merely 0.043, which gives the PR gap of as much as 0.247. Interestingly, both of these readings are motivated by Level 3 [growable] and [pickable], respectively. The small PR gap and, simultaneously, the top position of the Level 3 seme motivated readings can be accounted for by the features of tulip and their perception by language users. Like many other flowers, 'tulip' arouses a number of disparate connotations, and functions as a symbol (as do many other flowers). Due to this speaker-bound figurativeness and "connotability," the number of its possible readings (motivated by Level 4 semes) is very high, which results in an extreme scattering of the readings proposed, enabling one central reading (motivated by a Level 3 seme) to take an absolutely dominant position.
Let us finally comment on the smallest PR gap between two top readings. This is the case of 'to draw a triangle' (0.203) and 'to make a triangle from something/to make triangle-shaped things' (0.184). The gap of only 0.019 contradicts the above-indicated tendency of one clearly dominating reading for each first-encountered converted naming unit. This finding cannot be accounted for by the identical word formation type [QUALITY.sup.Result] ACTION, based on the factitive relation, because, for example, the above-discussed difference between 'to act or behave the boys do (immature)' and 'to (try to) look/behave like a boy (clothes, haircut, walk)--of girls', also characterized by the same WFT, is more than one predictability level (0.155). With triangle, the decisive factor appears to be the competition of more or less equally probable readings.
4.4.3. Seme level. Nine of ten most predictable readings are motivated by a Level 4 seme. This seems to justify the hypothesis of the important role of Level 4 semes in the meaning prediction process.
The only top reading motivated by Level 3 is 'to grow tulips'. The result (tulip) of action (growing) follows from the feature [growable]. Certainly, this feature is not typical of tulips only. It may be assumed that the reading based on Level 3 seme [growable] would be a preferred direction for the majority of conversions in the lexical group of plants.
4.4.4. Summary. The comments on Table 4 may be summarized in the following tendencies for the meaning predictability of novel context-free conversions:
--Tendency 1: One dominant reading.
--Tendency 2: Its PR is above 0.250 (the average value for the ten top readings in the sample is 0.280).
--Tendency 3: The gap between the first and the second readings usually exceeds one predictability level, that is, the PR of 0.1 (the average PR gap for the ten top readings in the sample is 0.148). From a different point of view, the rank 1 PR is twice the value of rank 2. The PR values of the rank 3 and lower rank readings are insignificant and negligible (the average PR for rank 3 readings in the sample is 0.027). (10)
--Tendency 4: An important (not sufficient, though) condition for meaning predictability is Level 4 of the motivating seme; in other words, a "good" PR tends to be conditioned by prototypical semes.
--Tendency 5: Predictability is influenced by extralinguistic factors, such as world knowledge and experience.
4.4.5. Main group vs. cheek group. Table 5 compares the results obtained from the main group and the check group of respondents.
In spite of the fact that the research techniques applied to these two groups differed substantially, Table 5 reveals a significant ranking identity of the most predictable readings in the main and the check groups. As for the most predictable readings, the agreement between the two groups obtains in nine out of ten cases. The only exception is triangle where the two most predictable readings have swapped their positions. This is not surprising, as the difference in the PRs of the two readings in the main group is very small (0.019). The agreement is especially remarkable with conference in which case the respective rankings of the top three predictable readings fully coincide.
A big gap between the main group of respondents and the check group seems to exist in the case of 'to build/construct a cableway' (rank 2 in the main group, rank 5-8 in the check group); however, a closer view reveals that all the readings in the check group, except for the unambiguously dominating 'to travel by cableway', feature minimum frequency of occurrence: the readings 'to move up, to progress and advance' and 'to haul/ help somebody during a climb', which share the second position, occurred three times, and the rank 4 'to help somebody make progress in a particular activity' occurred only twice.
4.5. Objectified predictability rate
The predictability rates calculated for the individual readings in the previous section take into consideration two significant variables. First, the number of respondents who adduced a particular reading for a given naming unit, thus indicating that the reading was acceptable to them. Second, the score assigned to the individual readings by the respondents determines the degree to which a particular selected reading was acceptable to them. The value obtained is important to show the meaning predictability situation inside a naming unit. However, what the PR value, calculated in this manner, does not reflect is the mutual influence of the individual predictable readings. It may be assumed that the predictability of a particular reading is influenced by the predictability rate of the next lower readings and the number of readings with a (relatively) significant PR value. It is only by taking into account this factor that we can mutually compare the PRs of top readings of various naming units.
The idea of PREDICTABILITY AS A COMPETITION-BASED RELATION leads us to the introduction of the notion of OBJECTIFIED PREDICTABILITY RATE (OPR). OPR is applicable to the best predictable readings, and makes it possible to interrelate the best predictable readings of various naming units. The idea of "competition" is not new. It was applied, for example, by Gagne and Shoben (1997), although in a different context. They speak of the competition between various thematic relations in their CARIN model (Competition Among Relations in Nominals) to demonstrate that the thematic relation most readily available, that is, the relation with greatest strength, is the one preferred in the interpretation of combined concepts (compounds). Gagne and Shoben's idea of competition should be highly appreciated, because it points out a GRADABLE nature of the ACCEPTABILITY to language users of VARIOUS POSSIBLE MEANINGS.
In Stekauer (1998), I use the same term for accounting and calculating the relative productivity of various WFTs belonging to the same conceptually defined cluster (agents, instruments, actions, etc.). Since, as it is assumed above, the word-FORMATION and word-INTERPRETATION processes are closely interrelated, the notion of competition is, naturally, reintroduced into the discussion of meaning predictability. While in Gagne and Shoben's approach, the competition pertains to thematic relations available for the interpretation of compound words, the present theory of meaning predictability applies this notion to unequal predictability rates of various potential readings of a particular novel naming unit.
The concept of predictability rate thus makes it possible to identify the reading with the highest chances to be selected by a language user out of a number of possible candidates when such a naming unit is encountered outside context for the first time. Importantly, the bigger the PR GAP the higher the OPR. This kind of relation can be advantageously calculated by Luce's (1959) choice rule "in which the strength of the first choice is weighed against the strength of other competing choices" (Gagne and Shoben 1997: 81). In terms of the present research, this formula should be read as follows: "the PR of the most predictable reading is weighed against the PR of other competing readings." As the analysis of my experimental data indicates the PR values of rank 4 and lower readings are generally very low and insignificant, and, by implication, have little effect on the PR of the top reading. For this reason, the number of PR values that are included in the denominator for the calculation of OPRs is limited to three.
The formula, adapted for the calculation of the OPR, is therefore as follows:
(12) OPR = [PR.sup.top] / [PR.sup.top] + [PR.sup.top-1] + [PR.sup.top-2]
The OPRs for the individual sample naming units are given in Table 6.
It follows from Table 6 that, out of the ten naming units used in my research, the best predictability can be attributed to tulip whose top reading's OPR is 0.831, closely followed by planet whose top reading's OPR is 0.795. These two readings clearly outscore the other readings. For the explanation of this situation, we must refer to Table 4. Much has already been indicated in Section 4.4 when commenting on the data in Table 4. The PR gaps between the top readings and the second rank readings of tulip and planet, respectively, are the greatest from among all sample naming units. The high PR gap is reflected in the strength of the top readings, which is further supported by the very low PRs of the third-rank readings. By implication, it is not the absolute PR value which is decisive for the actual predictability expressed as an OPR value (although it obviously plays an important role). For this reason, for example, the reading 'to take part in a conference' which has by far the highest PR (0.427) ranks as low as ninth in terms of OPR. This is caused by a tough competition on the part of the other two readings, especially the extraordinarily high PR of the rank 3 reading (0.165). On the other hand, 'to swim/bathe in a river', which features the lowest PR from among the ten top readings in the sample (0.200) ranks third according to the OPR (0.634). On the other hand, the high PR value of 'to grow tulips' (rank 2 among PR values) has been confirmed by the OPR value (also rank 2 among the OPR values) thanks to a very high PR gap (0.247). An overview of the interrelation between the PR value, the dominance of a top reading represented as the PR gap, and the OPR is summarized in Table 6.
To sum up, the objectified predictability rate gives a more objective picture of actual meaning predictability, and makes it possible to relate and compare the predictability of top readings of the individual naming units by reflecting the role of both the PREDICTABILITY RATE GAP and the STRENGTH OF THE LOWER RANK READINGS.
5. Meaning predictability and associative meaning: the experimental results in the light of free association of words
The experimental data discussed above were obtained in two different ways and under two completely different conditions which reflect different possibilities of encountering a new naming unit by a language user. The central part of the data was obtained without any time limit imposed on the respondents, reflecting a situation when a language user has enough time to think over the different possible meanings and to select the most acceptable to him/her. The "check group" interpretations were obtained under time pressure, a situation not infrequently experienced by language users. Furthermore, while the former group of respondents was asked to propose the greatest possible number of potential readings for each sample naming unit and rate them according to (the degree of) their acceptability, the latter group was asked to propose just one reading that came immediately to a respondent's mind.
The idea behind choosing two completely different experimental conditions was--in addition to reflecting two different ways of encountering new naming units by language users--to identify the role of the time factor in identifying the most predictable readings, that is, to determine whether sufficient time space for identifying and rating the most acceptable (predictable) meaning(s) is a significant predictability-related factor. The comparison of the main group and the check group indicates that the time factor does not play a crucial role and has minimal impact on the decision-making of language users in regard of identifying the predictable readings. This might indicate two things:
1. The dominant position of certain combination(s) of particular prototypical features, captured in the concepts of the objects that are related in the concept of an object to be named, is so striking that its identification is not time-dependent. This mainly applies to those cases characterized by the positive influence of the "predictability-boosting conditions" (i.e. high productivity of a WF type, single "strong" reading, motivation by prototypical seme[s], general knowledge, etc.)
2. One also might suspect that there are certain more or less firm connections between the meanings of lexical units stored in the mental lexicon which, in connection with predictability-boosting conditions, might contribute to speaker's meaning prediction capacity.
The view of the interrelatedness of meanings of lexical units stored in the mental lexicon is not new. Lexical semanticists and morphologists came up with the idea of lexical, semantic, and conceptual fields, based on the paradigmatic relations of hyponymy/hyperonymy, antonymy, synonymy, etc. Recent psycholinguistic research into associated meanings has confirmed the existence of such connections of different strength. In addition, recent research in the field of pragmatic corpus-based lexicography has emphasized the requirement of well-balanced paradigmatic and syntagmatic relations in determining the meanings of lexical units (Sinclair 1998). With regard to the associative meaning principles, this requirement would, in an ideal case, mean that
(a) for complex words with two stems, the most predictable meaning of a novel naming unit results from the intersection of the strongest associative meanings (targets) bound to syntagmatically related words (cues), or, at least, from an associative meaning bound to one of the constituents;
(b) for complex words with a stem and an affix, the most predictable meaning(s) of a novel naming unit results from the intersection of the strongest associative meaning and the general meaning of the affix; or, at least from an associative meaning bound to one of the constituents; and finally,
(c) for converted naming units, the most predictable meaning(s) of a novel naming unit should be identified by the strongest connection between cue and target.
Therefore, the objectives of this section are to check whether there is any relation between the predictability of meaning(s) of new naming units and the associative connections between lexical units stored in our mental lexicon.
Based on the previous discussion, the research into the relation between meaning predictability and association was implemented with two predictions in mind: (a) there should be at least some overlap between associates and the experiment-generated meanings; and (b) rankings of associates and meanings should correlate.
For that purpose, I used an associate connectivity database by Nelson, McEvoy, and Schreiber available on the Internet. Appendix A of the database includes pairs of 5,019 normed words (cues) and the 72,176 responses (targets). The database provides 31 different types of data which can be variously combined, depending on a user's needs. For the purpose of this research, I made use of the FSG parameter (column 6) indicating the "forward strength," that is, the CUE-TO-TARGET STRENGTH (i.e. the probability that one word produces another word) (Nelson et al. 2001: 1153), calculated by dividing #P (the number of participants producing a particular response [column 5]) by #G (the number of participants in the group norming the word [column 4]).
Importantly, the general free association procedure used to identify a word's associates and the strength of connections is similar to that used for the identification of most predictable meanings by the check group: "Initially, a word is presented to a large group of participants who are asked to produce the first word to come to mind. The probability of a particular associate is determined by dividing its frequency as a response by the sample size" (Nelson et al. 2001: 1147).
Since conversion is based on a single word stem, it lends itself better than two-stem units for the examination of any links between meaning predictability and associative meanings. Therefore, the top PR readings of converted naming units of this experiment were confronted with the database associates. I was looking for correlations between the PR and FSG values and between the respective rankings of the two groups of meanings. The following is a brief review, including the best predictable readings and the corresponding associates (if any).
5.2. Predictability rate versus associative meaning
The meanings proposed by the respondents for the individual sample words are compared with the associative meanings as given in the above discussed database.
1. to act or behave the way boys do (immature)
2. to (try to) look/behave like a boy (clothes, haircut, walk)--of girls
3. to change the sex from girl into boy
Associates: FSG Rank 'boy' [right arrow] girl 0.701 1 [right arrow] man 0.27 4 [right arrow] child 0.11 7
There is a partial correlation between the results. Thus, the top-rank association 'girl' corresponds with the second-rank PR reading, and the rank 4 associate 'man' corresponds with the top PR reading. A partial and indirect relation between 'to bring up a boy-child' and the 'immature behavior' contained in the top PR reading may be partly identified in the associate 'child'. Other associates, included in the Nelson database--namely, 'blue', 'George', 'cute', 'friend', 'dog', 'hot', 'scout'--have no counterparts among the readings in my research.
1. to roar/shout
2. to have the power and rule/to act as a leader
3. to catch/hunt/kill a lion
4. to behave/act like a lion
5. to be brave
6. to protect somebody as lioness protects her cubs (only three occurrences, none in the native-speakers or check groups)
Associates: FSG Rank 'lion' [right arrow] cub 0.74 3 [right arrow] roar 0.614 5 [right arrow] king 0.21 11
'King' can be associated with the PR reading 'to have the power and rule/to act as a leader'. It is, however, surprising that there is no associate indicating lion's physical characteristics, such as 'strength', its look (mane), and behavior (brave, lordly), that is, the prototypical features that one would expect to appear in an association database. On the other hand, the readings based on these prototypical features were not infrequent in my experimental research.
1. to grow tulips
2. to pick tulips
3. to decorate with tulips
4. to shape something like a tulip
5. to look beautiful
Associates FSG Rank 'tulip' [right arrow] flower 0.782 1 [right arrow] pretty 0.027 3 [right arrow] red 0.020 5 [right arrow] Holland 0.014 6 [right arrow] plant 0.014 7
It seems that one can expect some relation between association meanings and meaning predictability in cases pertaining to appearance and behavior. This is clear with animate beings like the experimental naming units to boy and to lion. Since tulips have no behavior, an important factor of potential association-inspired interpretation is lost. What remains is appearance, which is reflected in the relation between a low PR 'to look beautiful' and a high-rank (3) but very low-FSG associate 'pretty'. The associates based on the hyperonym-hyponym relation do not seem to have any significant influence on the meaning interpretation. This influence is highly limited and indirect only: any flower (but not any plant) can be picked; any flower and plant are growable and/or can be used for decoration. In this connection, it should be noted that the readings in which a converting noun becomes an object of action can hardly be expected to have any counterparts among the associates due to the universal nature of the relevant action (pick, grow, decorate).
In the check group, 3 out of 25 respondents selected the reading 'to be shy, to blush with shame' which may be related to the 'red' associate, but apparently, this link is rather loose taking into account the FSG of the associate and the low frequency of occurrence of the PR reading. This poor connection is only supported by four occurrences of a related reading in the NS group--'to turn red in face' PR.
The 'Holland' associate (its FSG approaching zero) has even weaker links to a single-occurrence reading 'to go to Amsterdam' in the NS group and a two-occurrence reading 'to go to Holland' in the NNS.
Since the naming unit cableway does not occur in the database, no evaluation is possible.
1. to travel to other planets/to land on a planet
2. to populate/colonize another planet
3. to be interested in planets and space
4. to discover a planet
Associates FSG Rank 'planet' [right arrow] space 0.042 4 [right arrow] universe 0.018 7 planets [right arrow] orbit 0.013 18
The link is indirect and loose in this case. The high PR readings concerning space travel, colonization, and also discovery of other planets may be related freely to the associates 'space' and 'universe' indicating the remoteness, thus establishing a kind of general framework for the identification of more specialized readings. The lowest FSG associate 'orbit' is related to the corresponding action by two occurrences with very low rating.
1. to swim/bathe in a fiver
2. to travel down a fiver (in a canoe, kayak, etc.)
3. to direct river into an area in order to irrigate it
4. to flow, to spread
Associates FSG Rank 'river' [right arrow] stream 0.118 2 [right arrow] flow 0.063 4 [right arrow] boat 0.055 5 [right arrow] canoe 0.055 6 [right arrow] swim 0.016 19
The only strong associate 'stream' may be related to the low PR 'to flow, to spread'; the highest PR 'to swim/bathe in a river' relates to the 'swim' associate the FSG of which approaches zero, and the rank of which is far from corresponding to the significance of the PR reading. The second-rank PR reading 'to travel down a river (in a canoe, kayak, etc.)' does not give more significant support from the associates 'boat' and 'canoe' either. The top-rank associate 'lake' has no PR counterpart.
1. to take part in a conference
2. to meet to talk about/consult/discuss a subject
3. to organize a conference
Associates FSG Rank 'conference' [right arrow] meeting 0.390 1 [right arrow] talk 0.082 3 [right arrow] people 0.027 6
The 'meeting' associate seems to be linked with both of the highest PR readings. The connection between 'to take part in a conference' and 'people' is not strong enough due to the very low FSG of the latter. There seems to be a correlation between the second strongest associate 'talk' and 'to meet to talk ...' in terms of their respective rankings. Unfortunately, the FSG of the former is not very high.
1. to draw a triangle
2. to make a triangle from something/to make triangle-shaped things
3. to be a part of a triangle of lovers
4. to move between three points
Associates FSG Rank 'triangle' [right arrow] geometry 0.095 2 [right arrow] shape 0.027 8 [right arrow] three 0.027 9 [right arrow] love 0.014 12
Neither triangle confirms a correlation between the meaning prediction and the association processes. The 'geometry' and 'shape' associates are too general to directly motivate the 'drawing' and 'making' readings of triangle. The only direct connection seems to be in the case of 'love'. However, the rank and the FSG-value of this associate are very low, comparatively much lower than the PR of 'to be a part of a triangle of lowers', to be of any major significance.
The structure of the associates suggests a process leading in a completely different direction compared to the prediction process. While top ranks among the associates are assumed by incompatible geometrical figures, namely, square, circle, rectangle, they, because of their incompatibility with triangle, cannot and do not play any role in the prediction process of triangle.
Courtyard is not included in the association database.
1. to get up early in the morning
2. to do one's morning routine
3. to greet somebody in the morning/to say 'good morning'
4. to work in the morning
5. to make/have breakfast or morning coffee
6. to wake up somebody in the morning
Associates FSG Rank 'morning' [right arrow] early 0.176 2 [right arrow] coffee 0.034 8 [right arrow] wake 0.014 18
While the associate 'early' points to the topmost PR reading, a more direct association with 'to get up early in the morning' is provided by the associate 'wake'. Unfortunately, its rank and FSG are too low. Since the verb to morning necessarily implies a kind of activity done by an agent in the morning, the associate 'coffee' is a good candidate for correlation with one's morning 'agenda'. The associates like 'sun' and 'sunrise' do not pertain to the agent's activity, and therefore can hardly motivate the interpretation of the verb to morning.
The picture obtained from the analysis of associates and predictable readings indicates that the first of the above-mentioned predictions, expecting at least some degree of overlap between associates and the most predictable readings, has been borne out for some of the readings. The other prediction concerning the correlation between rankings of associates and most predictable readings has not been substantiated.
The answer can be sought in the basic principles of the meaning predictability theory outlined in this article, in particular, the role of the PROTOTYPICAL SEMES of naming units (reflecting the prototypical features of the corresponding referents).
Association seems to be based on a variety of relations. A significant portion of them are represented by the lexical relations of antonymy (oppositeness) and hyponymy/hyperonymy. The associates based on the former (e.g. square, circle, for 'triangle'), cannot correlate with predictable meanings for the simple reason that the prototypical features of antonymous words are incompatible, mutually excluding, and therefore cannot motivate the interpretation of new naming units. The same is true of the associates constituting what Cruse (1986) calls helices as a special type of nonhierarchical lexical configurations (night, evening, afternoon, day in relation to 'morning').
The hyponymy/hyperonymy relations must be excluded for a different reason. If the meaning of the word in question is a hyponym of a corresponding hyperonym, the latter necessarily functions as an identification or a classification seme of the former. My basic assumption is, however, that it is the prototypical rather than the too general identification and classification seines that motivate the meaning of a novel naming unit. It is for this reason that the motivating capacity of hyperonyms ('animal' and 'beast', for example, with regard to to lion, and 'man' [in the sense of human being] and 'child' with regard to boy) is limited. Co-hyponyms are not good 'motivators' either (for instance, 'tiger' and 'cat' with regard to to lion).
A different situation involves those antonyms which, in the interpretation process, are not perceived from contrastive perspective; instead, the target and the cue enter into the relation of SIMILARITY. A case in point is the boy--girl "pseudo-antonymy" where the behavior of boys serves rather as a PATTERN for the behavior of girls. The other relevant associates, 'man' and 'child' show that the PATTERN-based (SIMILARITY) relation may establish a certain correlation of results. Similarly, one of the associates of to lion is 'to roar' which is based on the PATTERN (SIMILARITY) relation (similar behavior), in the same way as the 'king' associate (similar power/strength).
Another important observation is that the associative principle does not seem to permit more specific interpretations, such as 'to give birth to a boy' and 'to change the sex from girl to boy' where the cue is the result of action. The reason is that this kind of reading combines two dimensions, action and its result. Contrary to this, association appears to be a single-dimension phenomenon. In principle, since the associative relation between cue and target is single-dimensional it precludes the associates from providing the LEVEL OF DETAIL necessary for the interpretation of novel context-free naming units.
The purpose of this article has been to point out some aspects of a theory of novel context-free naming units, to propose a method of calculating the predictability rate and objectified predictability rate for these naming units, and to apply this method to converted naming units.
There are two different kinds of predictability: the context-free predictability and the context-bound predictability which might also be labeled as system-level predictability and speech-level predictability, respectively. These two kinds of predictability require different methods of research and calculation. The present research has been focused on the former type, that is, on (the factors affecting) the ability of speakers of a language to predict the meaning of a productively and regularly coined naming unit when they encounter it for the first time without any context support.
It should be noted that any statements presented in this article, summarizing the research results, should be taken as an indication of certain prevailing tendencies. Their relevance may be increased by a series of analogical research projects. Moreover, as in any other research in which subjective assessment plays an important role (respondents) the results may have been influenced by several factors. They include:
(a) Attention paid to completing the task by the individual respondents, that is, the degree of their involvement. A comparison between the Slovak and the native-speaker groups indicates that the native speakers generally proposed a smaller number of readings for the individual converted naming units, while the number of proposals obtained from Slovak students was higher. My students appear to have been more enthusiastic and have taken the research as a kind of creative game. On the other hand, I could not personally influence native speakers' attitudes to the research because I did not come into contact with them.
(b) Linguistic awareness of the individual respondents.
(c) Their imagination.
(d) Their personal experience and extralinguistic knowledge.
6.2. Basic conclusions
The most significant conclusions may be summarized as follows:
1. PROTOTYPICALITY OF A MOTIVATING FEATURE (SEMANTIC COMPONENT) IS AN IMPORTANT, NOT A SUFFICIENT CONDITION THOUGH.
With two exceptions, the most predictable readings obtained for the ten non-established conversions are based on the activation of Level 4 semes. It may be therefore assumed that meaning predictability is conditioned by the inherence of features, by their prototypical nature; any individualization, trespassing on the prototypical feature limits, appears to become a serious obstacle to predictability. Figurative meanings rarely exceed the predictability level (an exception is, e.g., 'to roar/shout; to be aggressive', 'to have the power and rule/to act as a leader' for lion). It should be, however, noted that the activation of a Level 4 seme itself does not guarantee a high PR.
2. There is a tendency for novel conversions to have oNE CENTRAL AND DOMINATING READING whose PR is usually above 0.250.
3. The PR GAP BETWEEN THE TOP AND THE SECOND-RANK READINGS USUALLY EXCEEDS ONE PREDICTABILITY LEVEL (the PR of 0.1), with the PR value of the most predictable reading being twice the PR value of the rank 2 reading. PR values of the rank 3 and lower rank readings are usually insignificant.
4. The predictability of the most predictable readings of various naming units can be compared by introducing the notion of an OBJECTIFIED PREDICTABILITY RATE taking into account the predictability rate gap and the PR values of the lower-rank readings.
5. There is no single factor conditioning the predictability of novel context-free naming units. On the contrary, it is possible to speak of the INTERPLAY OF SEVERAL FACTORS, some of them boosting and others reducing the meaning predictability. The most important of them (in the case of converted naming units) include
(a) the productivity of a word formation type
(b) activated seme level
(c) extralinguistic knowledge (knowledge of the world and experiences)
(d) the number and the strength of competing readings.
6.3. Concluding remarks
In the epilogue of The Indo-European Lexicon: A Full Synchronic Theory, R. Beard (1981) compares his approach to the range of possible meanings of a derivative with that developed by Clark and Clark (1979). While the Clarks restrict the range of possible meanings of what they call "contextuals" (my "conceptual recategorization") by the range of logical possibilities, Beard's answer is based on the system of Indo-European categories. While the Clarks base their approach on performance, Beard concentrates on competence-bound linguistic regularities, in particular on lexical-syntactic paradigms. Beard admits that these regularities can be overridden by performance. That is why those readings of, for example, to bottle and to teapot which deviate from those predicted by the respective rules are treated in his approach as exceptions.
In any case, apart from the differences resulting from different points of view, the two theories are very close to each other:
The Clarks' generic knowledge, underlying the interpretation of "contextuals," is based on the classification of extralinguistic objects and relations in terms of placeables, agents, results, instruments, etc., a system which is similar to Beard's Indo-European case function system.
Both Clark and Clark's and Beard's classifications specify possible interpretations of novel naming units. While Clark and Clark emphasize extralinguistic knowledge, Beard puts emphasis on linguistic categories. The idea of the prototypical semes bridges over these two approaches because it assumes that prototypical semes as semantic primitives at the level of language (the meaning facet of linguistic signs) are direct reflections of conceptual categories (logical predicates) at the level of human mind which, in turn, reflect the prototypical features of an object named. From this it follows that meaning predictability, crucially depending on PROTOTYPICAL SEMES, is influenced and conditioned by both LINGUISTIC AND EXTRALINGUISTIC FACTORS.
The aim of this article has been to demonstrate that--as noted above--the predictability of novel context-free naming units is a matter of COMPETENCE, both linguistic and extralinguistic. In this respect, it differs from the predictability of contextualized novel naming units which falls within the scope of performance.
Received 9 April 2003
Revised version received 6 July 2004
Allen, Margaret R. (1978). Morphological investigations. Unpublished doctoral dissertation, University of Connecticut.
Beard, Robert E. (1981). The Indo-European Lexicon: A Full Synchronic Theory. North-Holland Linguistic Series 44. Amsterdam: North-Holland.
--(1995). Lexeme-Morpheme Base Morphology. A General Theory of Inflection and Word Formation. SUNY Series in Linguistics. Albany, NY: State University of New York Press.
Clark, Eve V.; and Clark, Herbert H. (1979). When nouns surface as verbs. Language 55(4), 767-811.
Cohen, Benjamin; and Murphy, Gregory L. (1984). Models of concepts. Cognitive Science 8, 27-58.
Coolen, Riet; van Jaarsveld, Henk J.; and Schreuder, Robert (1991). The interpretation of isolated novel nominal compounds. Memory and Cognition 19, 341-352.
--; van Jaarsveld, Henk J.; and Schreuder, Robert (1993). Processing novel compounds: evidence for interactive meaning activation of ambiguous nouns. Memory and Cognition 21, 235-246.
Cruse, Alan D. (1986). Lexical Semantics. Cambridge: Cambridge University Press.
Derwing, Bruce L.; and Skousen, Royal (1989). Morphology in the mental lexicon: a new look at analogy. In Yearbook of Morphology 2, G. Booij and J. Van Marle (eds.), 55-71. Dordrecht: Foils.
Downing, Pamela (1977). On the creation and use of English compound nouns. Language 53(4), 810-842.
Finin, Timothy W. (1980). The Semantic Interpretation of Compound Nominals. Coordinated Science Laboratory, Report T-96. Urbana, IL: University of Illinois.
Gagne, Christina L. (2001). Relation and lexical priming during the interpretation of noun-noun combinations. Journal of Experimental Psychology: Learning, Memory, and Cognition 27(1), 236-254.
--; and Shoben, Edward J. (1997). Influence of thematic relations on the comprehension of modifier-noun combinations. Journal of Experimental Psychology: Learning, Memory, and Cognition 23(1), 71-87.
Gleitman, Lila R.; and Gleitman, Henry (1970). Phrase and Paraphrase: Some Innovative Uses of Language. New York: W. W. Norton.
Hampton, James A. (1983). A composite prototype model of conceptual conjunction. Unpublished manuscript, The City University, London.
--(1987). Inheritance of attributes in natural concept conjunctions. Memory & Cognition 15, 55-71.
Katz, Jerrold J.; and Fodor, Jerry A. (1963). The structure of a semantic theory. Language 39, 170-210.
Lees, Robert B. (1960). The Grammar of English Nominalizations. Bloomington, IN: Indiana University Press.
Levi, Judith N. (1978). The Syntax and Semantics of Complex Nominals. New York: Academic Press.
Li, Charles (1971). Semantics and Structure of Compounds in Chinese. Berkeley: University of California.
Luce, R. (1959). Individual Choice Behaviour. New York: Wiley.
Mathesius, Vilem (1975). A Functional Analysis of Present Day English. The Hague: Mouton.
Meyer, Ralf (1993). Compound Comprehension in Isolation and in Context. The Contribution of Conceptual and Discourse Knowledge to the Comprehension of German Novel NounNoun Compounds. Tubingen: Niemeyer.
Murphy, Gregory L. (1988). Comprehending complex concepts. Cognitive Science 12, 529-562.
Nelson, Douglas L.; McEvoy, Cathy L.; and Schreiber, Thomas A. (1999). The University of South Florida word association, rhyme, and word fragment norms. [http://www.usf. edu/FreeAssociation/].
--; Zhang, Nan; and McKinney, Vanesa M. (2001). The ties that bind what is known to the recognition of what is new. Journal of Experimental Psychology: Learning, Memory and Cognition 27, 1147-1159.
Ondrus, Pavel; Horecky, Jan; and Furdik, Juraj (1980). Sucasny slovensky spisovny jazyk. Lexikologia. Bratislava: SPN.
Ryder, Mary E. (1994). Ordered Chaos: The Interpretation of English Noun-Noun Compounds. Berkeley: University of California Press.
Sinclair, John (1998). The lexical item. In Contrastive Lexical Semantics, E. Weigand (ed.), 1-24. Amsterdam: John Benjamins.
Skousen, Royal (1989). Analogical Modelling of Language. Dordrecht: Kluwer.
Smith, Edward E.; Osherson, Daniel N.; Rips, Lance J.; and Keane, Margaret (1988). Combining prototypes: a selective modification model. Cognitive Science 12, 485-527.
Stekauer, Pavol (1996). A Theory of Conversion in English. Frankfurt am Main: Peter Lang.
--(1998). An Onomasiological Theory of English Word-Formation. Amsterdam: John Benjamins.
--(2001). Fundamental principles of an onomasiological theory of English word-formation. Onomasiology Online 2, 1-42 [www.onomasiology.de].
--(2005). Meaning Predictability in Word Formation. Amsterdam: John Benjamins.
--; Franko, Stefan; Horecky, Jan; Staskova, Jaroslava; and Sutherland-Smith, James (1997). Getting beyond the Rules of Language. Presov: Acta Facultatis Philosophicae Universitatis Presovensis.
van Jaarsveld, Henk J.; Coolen, Riet; and Schreuder, Robert (1994). The role of analogy in the interpretation of novel compounds. Journal of Psycholinguistic Research 23(2), 111-137.
van Lint, Trudeke (1982). The interpretation of compound nouns. In Linguistics in the Netherlands 1982, S. Daalder and M. Gerritsen (eds.), 135-145. Amsterdam: North-Holland.
Wisniewski, E. J. (1996). Construal and similarity in conceptual combination. Journal of Memory and Language 35, 424-453.
Zimmer, Karl E. (1971). Some general observations about nominal compounds. Working Papers on Language Universals 5, 1-21.
--(1972). Appropriateness conditions for nominal compounds. Working Papers on Language Universals 8, 3-20.
* Correspondence address: Katedra anglistiky a amerikanistiky, Presovska Univerzita, U1 17, novembra c. 1,081 16 Presov, Slovakia. E-mail: firstname.lastname@example.org.
(1.) The term naming unit refers to units generated in the word-formation component, that is, units coined by productive rules of word formation. This term was introduced by V. Mathesius (1975). In my approach, it substitutes for terms like word, lexeme, etc., because of their inconsistent use and varying connotations in linguistic literature.
(2.) They are treated in the same way as derivations with overt suffixes, for example, scar: to scarify = mark: to mark, because the former has the meaning 'cause to have scars' and the meaning of the latter is based on the same formula, 'cause to have marks'. Similarly, to crystallize from crystal may mean either 'come to have crystals' or 'cause to have crystals'; the same meanings are carried by the denominal to clot: 'come to have clots' or 'cause to have clots' (Beard 1995: 180).
(3.) A terminological note: "functional derivation" changes the basic lexical meaning by means of rules operating over "grammatical" functions like subject, object, locus, means, manner, possession, possessive, origin, etc. "Transposition," on the other hand, reflects the capability of the lexicon to change the lexical class of a lexeme without changing the basic lexical meaning. A complete list of these universal functions, including examples, is given in Beard (1995).
(4.) This task is irrelevant for our present purpose. It tested the translation ability of our students.
(5.) Obviously, the terms "correct" and "false" should be understood in the context of the research referred to, in particular, the role of general knowledge in the ability to understand converted proper names. Disregarding this particular context, one cannot speak about "correct" and "incorrect" meaning because the word formation process should be understood as creativity within productivity constraints. This means that even the least probable meaning is "correct" if a naming unit has been formed by a coiner with that particular meaning in his mind. What a theory of predictability can refer to is the predictability rate as defined at the outset of this article.
(6.) The classification of the individual readings was, in some cases, accompanied by the problem of whether certain readings should be grouped together or classified separately in two different cohorts. This task proved to be demanding and, necessarily, subjective to a certain degree. Since there are no hard-and-fast criteria identifying the borderline between related readings (due to their fuzzy edges), and since the situation is highly case-specific, I mostly relied on my judgment and intuition. This is not to say that the classification was haphazard or voluntaristic. The basic criterion of grouping various readings was their semantic identity (the same interpretation expressed in different words) or close semantic cognation; next, it was the same degree of generalization (for this reason, the readings, such as 'to act or behave the boys do' and 'to speak like a boy' were not grouped together); third, important help has also come from the informants themselves: if two semantically cognate readings were proposed separately as two different readings by one and the same reader, they were classified separately.
I am aware of Coolen et al.'s (1991) experiment in which they employ three independent judges to classify the readings, proposed by the respondents, according to the classification criteria adopted from Levi (1978). Murphy (1988) also makes use of judgments of two "raters" to identify the number of meanings proposed by his respondents. While this solution seems to be more objective, I rejected it for the reason that Levi's classification is too general and cannot capture the subtle semantic differences in the proposed readings. As a result, it does not lend itself for the predictability research within the proposed framework.
(7.) There are also some other types of readings, like 'to bring up a boy child' ([SUBSTANCE.sup.Object]ACTION), 'to change the sex from girl into boy' and 'to organize a team or a group where there are only boys' ([SUBSTANCE.sup.Result]ACTION), etc. Interestingly, the reading 'to staff with boys', based on the analogy with the central meaning of many, that is, 'to furnish with a staff', gained minimum support.
(8.) The contrast based on the seme [ADULT] is much more significant for this reading than that based on the seme [MATURE] as it much broader, and as such, it reflects a number of other differences between 'boy' and 'man'.
(9.) For obvious reasons, such a research goes beyond the scope of this study.
(10.) The evaluation of meaning predictability disregards the number of readings proposed for the individual naming units. As it follows from the evaluation, only few of a relatively large number of readings are worth discussing. Single occurrence readings are of little use for meaning predictability. As pointed out by Nelson et al. (1999: 4) in the introduction to Appendix A of their "associate" database, the criterion "two or more" participants "was chosen many years ago on the assumption that idiosyncratic responses given by a single participant would tend to be 'off the wall'." In other words, the readings whose frequency of occurrence in the sample is very low may only indicate the possibility to interpret novel context-free in various bizarre ways.
University of Presov
Table 1. The readings identified for and the scores assigned to [boy.sub.v] by native speakers Readings/ respondents 1 2 3 4 5 6 7 8 9 10 11 12 to act or behave 5 3 4 10 8 10 10 8 the way boys do (immature) to (try to) look/ 9 8 7 2 8 10 8 behave like a boy (clothes,haircut, walk)--of girls to exclaim "boy!" to call someone 8 a boy to speak like a boy to go to the pub 1 to serve someone 6 to 'staff' with boys 10 to undergo plastic surgery in order to look younger to give birth to 5 a boy to flirt with boys to kill boys 1 to change the sex 5 6 4 from female into male to assign "a fag" to an older boy to acquire a boyfriend to pursue pedophiliac activity to treat like a boy 2 2 to bully to buy or sell boys to play with to tease to attribute boyish 3 characteristics to someone Readings/ respondents 13 14 15 16 17 18 19 20 PR to act or behave 3 10 2 6 8 5 0.322 the way boys do (immature) to (try to) look/ 10 10 0.162 behave like a boy (clothes,haircut, walk)--of girls to exclaim "boy!" 5 0.0013 to call someone 0.002 a boy to speak like a boy 3 0.0008 to go to the pub 0.0003 to serve someone 0.0015 to 'staff' with boys 10 0.010 to undergo plastic 8 0.002 surgery in order to look younger to give birth to 10 7 0.0165 a boy to flirt with boys 7 0.0017 to kill boys 0.0002 to change the sex 3 6 0.030 from female into male to assign "a fag" 7 0.0018 to an older boy to acquire a 3 0.0008 boyfriend to pursue 9 0.0025 pedophiliac activity to treat like a boy 5 0.0068 to bully 8 0.002 to buy or sell boys 4 0.001 to play with 2 0.0005 to tease 10 0.0025 to attribute boyish 0.0008 characteristics to someone Table 2. The readings identified for and the scores assigned to [boy.sub.v] by L2 speakers Readings/ respondents 1 2 3 4 5 6 7 8 9 10 11 to act or behave 9 10 10 9 10 10 the way boys do (immature) to (try) to look/ 5 3 8 8 5 1 8 behave like a boy (clothes, haircut, walk)--of girls to take somebody 3 2 as a boy instead of taking him as an adult to hire someone 2 to carry luggage to change 1 someone's character from adult's character into youth's character to be interested 9 in boy's occupations to staff with 4 boys to bring up a boy 10 10 8 child to play 2 football very bad to never grow up 10 to be foolish/ 5 to fool around to be childish 9 to change the 3 10 8 sex from girl into boy to behave like 1 another's little boy to become 4 less sensitive to joke like 7 the boys do to fight like 9 a boy to be naughty 4 to have friendships with boys to give a girl a boyish name to think like a boy to something for a boy Readings/ respondents 12 13 14 15 16 17 18 19 20 PR to act or behave 10 10 10 9 5 10 9 0.3933 the way boys do (immature) to (try) to look/ 7 9 4 7 7 3 0.2438 behave like a boy (clothes, haircut, walk)--of girls to take somebody 0.0025 as a boy instead of taking him as an adult to hire someone 0.0005 to carry luggage to change 0.0003 someone's character from adult's character into youth's character to be interested 0.0023 in boy's occupations to staff with 0.001 boys to bring up a boy 0.021 child to play 0.0005 football very bad to never grow up 0.0025 to be foolish/ 9 0.007 to fool around to be childish 10 0.0095 to change the 5 5 0.0388 sex from girl into boy to behave like 0.0003 another's little boy to become 0.001 less sensitive to joke like 8 0.0075 the boys do to fight like 0.0023 a boy to be naughty 0.001 to have 2 0.0005 friendships with boys to give a girl a 5 0.0013 boyish name to think like 9 0.0023 a boy to something 6 0.0015 for a boy Table 3. Comparison of the most predictable readings for boy Reading Native speakers Non-native speakers Ranking PR Ranking PR to act or behave the way 1 0.322 1 0.393 boys do (immature) to (try to) look/behave 2 0.162 2 0.244 like a boy (clothes, haircut, walk)--of girls to change the sex from female 3 0.030 3 0.039 to male to 'staff' with boys 4 0.010 -- to bring up a boy child -- 4 0.021 Table 4. An overview of most predictable readings Naming unit Predictability rate Native Non-native speakers speakers Total boy to act or behave the way boys do 0.322 0.393 0.358 (immature) to (try to) look/behave like a boy 0.162 0.244 0.203 (clothes, haircut, walk)--of girls to change the sex from female 0.030 0.039 0.035 into male lion to roar like a lion; to be 0.143 0.330 0.237 aggressive to have the power and rule as a 0.106 0.144 0.124 lion to behave/act like a lion 0.066 0.031 0.047 tulip to grow tulips 0.350 0.233 0.290 to pick tulips 0.038 0.049 0.043 to decorate with tulips 0.024 0.009 0.016 to look beautiful -- 0.028 cableway to travel by cableway 0.309 0.309 0.309!!! to build/construct a cableway 0.319 0.160 0.232 to lay cables 0.026 -- 0.007 to work one's way up -- 0.023 planet to travel to/land on other planets 0.444 0.312 0.376 to populate/colonize a planet 0.078 0.100 0.089 to discover a planet 0.008 0.007 0.008 river to swim/bathe in a river 0.144 0.264 0.200 to travel down a river (e.g. in a 0.116 0.098 0.107 canoe) to direct river into an area in 0.030 -- 0.008 order to irrigate it to shed tears 0.001 0.027 conference to take part in a conference 0.380 0.476 0.427 to (meet to) talk about/consult/ 0.294 0.140 0.212 discuss a subject to organize a conference 0.231 0.110 0.165 triangle to draw a triangle 0.158 0.253 0.203 to make a triangle from 0.190 0.178 0.184 something/to make triangle- shaped things to be a part of a triangle of 0.030 0.025 0.028 lovers courtyard to play/spend time in the 0.302 0.296 0.299 courtyard to build a courtyard/to make an 0.280 0.112 0.186 area into a courtyard to enclose a space 0.039 -- 0.010 to keep something in a courtyard -- 0.009 morning to get up early in the morning 0.246 0.294 0.270 to do one's morning routine/to 0.128 0.095 0.111 perform usual morning activities to greet somebody in the 0.044 0.098 0.068 morning/to say 'good morning' Table 5. Comparison of the two main groups and the check-group: rankings of the predict-able readings Naming unit/reading Ranking Main Check group group boy 1. to act or behave the way boys do (immature) 1 1 2. to (try to) look/behave like a boy (clothes, 2 2 haircut, walk)--of girls lion 1. to roar/shout; to be aggressive 1 1 2. to have the power and rule/to act as a leader 2 3 tulip 1. to grow tulips 1 1 2. to pick tulips 2 -- cableway 1. to travel by cableway 1 1 2. to build/construct a cableway 2 5-8 planet 1. to travel to other planets/to land on a planet 1 1 2. to populate/colonize another planet 2 2 river 1. to swim in a river 1 1 2. to travel down a river (e.g. in a canoe) 2 2 conference 1. to take part in a conference 1 1 2. to meet to talk about/consult/discuss a 2 2 subject triangle 1. to draw a triangle 1 2-4 2. to make a triangle from something/to make 2 1 triangle-shaped things courtyard 1. to play/spend time in the courtyard (with 1 1 friends) 2. to build a courtyard/to make an area into a 2 -- courtyard morning 1. to get up early in the morning 1 1 2. to do one's morning routine/to perform 2 2 usual morning activities Table 6. Top reading ranking, PR gaps, objectified predictability rates Most predictable PR PR-based PR gap OPR OPR-based reading for each NU ranking ranking boy 0.322 3 0.155 0.601 5/6 lion 0.237 8 0.113 0.581 7 tulip 0.290 6 0.247 0.831 1 cableway 0.309 4 0.077 0.564 8 planet 0.376 2 0.287 0.795 2 river 0.200 10 0.093 0.634 3 conference 0.427 1 0.215 0.531 9 triangle 0.203 9 0.019 0.489 10 courtyard 0.299 5 0.113 0.604 4 morning 0.270 7 0.111 0.601 5/6
|Printer friendly Cite/link Email Feedback|
|Publication:||Linguistics: an interdisciplinary journal of the language sciences|
|Date:||May 1, 2006|
|Previous Article:||Case stacking in realizational morphology.|
|Next Article:||No "double plurals" in Dominican Spanish: an optimality-theoretic account *.|