and mini-kitchen is that while in both cases the referents contain a scaling-down element, the diminutive kitchenette evokes positive emotional associations, but mini-kitchen does not necessarily do so (Ungerer 2002: 545–46). This evaluative quality of diminutives has been observed for a number of languages in which diminutives are more widespread than in English, for instance for Italian and German diminutives (Dressler and Barbaresi 1994), for Dutch suffixations with -kje and -tje (Bakema, Defour, and Geeraerts 1993), and for Spanish suffixations with - ito and -illo (Santibanez-Saenz 1999). The latter approach is particularly interesting from a cognitive angle as an attempt, first undertaken by Ruiz de Mendoza Iba ´ n ˜ ez (2000) and developed by Santibanez-Saenz, to explain diminutives as an instance of conceptual fusion of three Idealized Cognitive Models (ICMs): the ICM of ‘size’ (which is responsible for scaling-up and scaling-down prefixation) and the ICMs of ‘control’ and ‘cost-benefit’. This fusion yields two superficially contradictory cor- ollaries, which can explain the existence of affectionate and pejorative meanings observed with the Spanish diminutives -ito and -illo, respectively: a. Small entities are usually manageable; as a result, they may be perceived as likeable. b. Small entities are usually innocuous and unimportant and may be easily ignored; as a result, they may be perceived as unpleasant. (Ruiz de Mendoza Iba ´ n ˜ ez, in Santibanez-Saenz 1999: 176) 3.2. -er Suffixation Of all English suffixes, -er has no doubt attracted most attention because of its frequency and versatility of meaning and use. Figure 25.6. Schema-based explanation of un-prefixation (based on Mettinger 1994: 22, 24) 660 friedrich ungerer The Radial Category Approach This approach, as pursued by Panther and Thornburg (2001), takes up the methods used successfully in describing polysemous lexical concepts in terms of protoypes, metonymic and metaphorical processes (Lindner 1981; Lakoff 1987; Rudzka-Ostyn 1985; Brugman 1981) and transfers them to -er derivation. In their analysis of -er, Panther and Thornburg start, rather than from a simple prototype concept, from a ‘‘basic event schema’’ (or prototypical ‘‘scenario,’’ in their terminology) and then make use of direct, metonymic, and metaphorical extensions to develop a radial category (or schematic network in Langacker’s terminology); they thus explicitly transcend the word class barriers that hamper traditional accounts of -er derivation. As illustrated in figure 25.7, the central sense reflects the nomina agentis notion of a primary participant intentionally acting on a directly affected or effected sec- ondary participant. The ‘‘central sense’’ takes ‘‘habitual occupational performance’’ as a starting point (as in verb-based teacher, baker,oractor); furthermore, it includes cases in which other elements of the prototypical scenario are metonymically expressed by nominal base concepts—for example, by the patient role (as in hatter ‘someone who creates hats’), by the instrument role (as in driftnetter ‘someone using a certain type of net in fishing’), or by the setting, such as that of the occupation (as in Senator). At varying distances from this prototypical scenario are five types of extensions, the first two sharing agentivity, but replacing habitualness by charac- teristic engagement (runner, snorer, or less dynamically, owner) or action-oriented disposition (left-hander, hetero-sexer, hardliner). The remaining extensions are ei- ther tied to habitualness in the sense of exhibiting an enduring attribute (widower, six-footer, murderer) or—somewhat more distant—try to integrate instances of context-dependent actions (caller, keynoter, thanker). A further metaphorical ex- tension links animal and plant concepts (retriever, creamer) and certain object concepts (gas-guzzler, sky-scraper) to the scenario by making use of the personifi- cation metaphor (nonhumans are humans). Banking on the conceptual closenesss of instruments and agents, the proto- typical scenario for nominals with human referents is complemented by an ‘‘in- strument scenario,’’ complete with verbal base and metonymic nonverbal base variants (as in can-opener and freighter, whaler, respectively), as well as extensions which try to integrate -er-formations based on lexical items denoting articles of clothing (sweater, pedal-pushers); locations (diner ‘dining-car’), and even patients (broiler ‘chicken’, reader ‘collection of texts’). The third large component of this radial category is made up of -er-formations expressing events (such as thriller, stomach churner, brain-teaser, etc.), which are explained as metaphorical extensions of the agent and instrument scenarios, re- spectively, based on the reification metaphor (events are objects). An interesting aspect of Panther and Thornburg’s approach is that they establish an intermediate category of ‘‘-er Gestalts’’ to account for borderline cases between genuine -er word-formation 661 derivatives and unanalyzable lexicalized items (plumber, tailor, trousers, hammer) (2001: 189). While Panther and Thornburg’s approach is synchronic, it does make use of diachronic evidence in the selection of the central sense, and its radial extensions can be seen as possible paths along which the diversification of the -er-formations may have developed and should therefore be submitted to empirical diachronic verifi- cation. A weakness of this model is its limited predictive force. Although Panther and Thornburg claim that productivity decreases as one moves away from the central sense, this is no more than a rule of the thumb and is not necessarily borne out by a comparison of more central and more marginal subcategories. The Cognitive/Pragmatic Constraints Approach Compared with Panther and Thornburg, who introduce their basic scenarios without much reference to traditional argument-structure analysis, Ryder’s (1999) study springs from detailed criticism of this approach, in particular of Rappaport and Levin’s (1992) interpretation. 7 While admitting the substantial achievements of argument-structure analysis, Ryder (1999) claims that it is not comprehensive en- ough to cope with the multiplicity of -er derivations, because the only thing that an Figure 25.7. The major categories of -er formation (compiled from Panther and Thorn- burg 2001, simplified) 662 friedrich ungerer -er suffix reliably indicates in present-day English is ‘‘that the derivative is a noun’’ (278), without restricting the underlying word class, let alone specific argument patterns, however dominant some of them may be. While this generalization is easily captured as a high level schema in Langacker’s system (see figure 25.2b), it is ob- viously not satisfactory as a differentiating explanation. This is why Ryder suggests two kinds of constraints: restrictions acting on the base of the -er items and re- strictions acting on the referent of the derivative (the target concept); see table 25.2. As far as the first two constraints of table 25.2 are concerned, Ryder’s proposal is close to the traditional notion that, prototypically, -er forms are derived from actions verbs with agentive or instrumental subjects which are also capable of rendering the durativity required by nominalizations. Yet Ryder differs in that she does not attempt to pin down -er derivation to specific argument patterns (agents, certain kinds of instruments), and this permits the inclusion of -er forms reflecting an underlying locative, patient, or circumstantial relation (as in diner ‘dining car’, scratcher ‘a lottery ticket that is scratched’, or clothing items like rompers and sleepers; Ryder 1999: 270). The third constraint is related to Ryder’s bottom-up approach to noun-noun compounds discussed below (section 3.4), in that it as- signs priority to specific rather than general event schemas. This restriction ex- plains, for instance, why -er forms like doer and maker are rare due to the vagueness of the underlying event schemas, but become quite acceptable when restricted by a noun or numeral, as in dress-maker, evil-doer,orsix-footer (Ryder 1999: 281). With the last three constraints, Ryder moves from the analysis of the semantic structure into the domain of psychological parameters of entrenchment (applied to base concepts), salience, and, as a kind of precondition of salience, identifiability (both applied to the target concept and its referent). While the first parameter (salience) requires little comment (although it may have to be supported by se- mantic frequency counts), salience and identifiability of the target concept are par- ticularly important where a less typical underlying relationship overrides agent or instrumental relationships. In this respect, consider roaster, which is nowadays understood to express the patient relationship of ‘roasted chicken’ rather than an Table 25.2. Cognitive/pragmatic constraints on the formation of -er derivatives (based on Ryder 1999) Restrictions acting on the base concept of -er derivates 1. Bases must have few, preferably only a single event schema (a condition most easily met by action verbs). 2. Event schemas must be capable of being applied to durative and habitual actions. 3. Events schemas must be specific (e.g., supported by verbs describing specific actions or nouns and adjectives providing a specific context). 4. Event schemas must be highly entrenched. Restrictions acting on the target concepts of the -er derivatives 5. Target concepts and their referents must be salient. 6. Target concepts and their referents must be identifiable. word-formation 663 instrumental device used for roasting, while the agentive reading ‘a person ha- bitually engaged in roasting meat’ is practically excluded (Ryder 1999: 289). Ryder’s claim that her account integrates -er forms with nonverbal bases is not all that convincing considering that it is only supported by a few explanatory remarks (Ryder 1999: 290) and an appendix of only roughly classified material. What is missing is a systematic link between underlying nouns and adjectives and event schemas that meets her own constraints of action and durativity. In this sense, Ryder does not take the issue of productivity constraints any further than Panther and Thornburg do. 3.3. Conversion Conversion and Figure/Ground Alignment The fact that most conversions involve a verbal processual structure (either as source or target) suggests a description in terms of Figure and Ground or trajector and landmark spread over time, an idea taken up by Twardzisz (1997). For verb-noun conversion, his interpretation is very much in line with Langacker’s explanation of -er derivation (see figure 25.2b). The verbal concept cheat is combined with the schematic nominalizing morpheme nr (for nominalizer) to create the agent noun cheat (Twardsisz 1997: 130). The interpretation proposed for noun-verb conversions is more original. Here, Twardzisz assumes that the concept underlying the converted noun may be added, as a ‘‘secondary’’ landmark, to the (clausal) processual rela- tionship between trajector and primary landmark. As illustrated in figure 25.8, the secondary landmark ‘‘interferes’’ with the processual relationship between trajector and primary landmark, thus affecting a change of meaning. For example, the sec- ondary landmark ‘salt’ is added to the processual relationship between the trajector ‘Peter’ and primary landmark ‘soup’, as lexicalized in Peter treated the soup in some way, which then assumes the meaning Peter salted the soup. Doubtless, Langacker’s framework is ingeniously extended in Twardzisz’s ap- proach to cover an important domain of word-formation. However, there remains a gap between his fairly general and abstract application of the trajector-landmark contrast and his very concrete observations concerning examples from individual domains like food, tools, animals, and human occupations: it is only at this more specific level that Twardzisz addresses and explains questions of productivity, mainly in terms of entrenchment. What is missing is a link with the more traditional explanation based on semantic roles, as supplied by Dirven (1999) in his event- schema-based approach. Noun-Verb Conversion and Event Schemas Event schemas are an area where the connection between Cognitive Linguistics and traditional semantic analysis is most obvious, especially if one takes Dirven’s lead, who was one of the earliest adherents of Fillmore’s Case Grammar. Dirven 664 friedrich ungerer (1999) develops his analysis of noun-verb conversions against the background of six major event schemas, which have sprung from related configurations of pre- dicator and semantic roles, as listed in table 25.3. The aim is to show which of these schemas can be understood as input strat- egies for conversion processes. If one selects the action schema ‘Agent-Patient’, as in X paints a picture, Dirven would claim that the Patient ‘picture’ functions as a metonymy for the painting action as a whole (or its metaphorical meaning of creating an image or representation). While this fact could have been stated within an argument-structure analysis, the explanation why this process works as it does depends on the cognitive notion of salience (see also Schmid, this volume, chapter 5). Following Langacker’s approach that an event schema can be seen as iconically reflecting the energy transmission in action chains, we find that the energy flow is triggered by the head of the chain (the agent) and is transmitted to the second major element, the patient, which is affected, or even changed, by the energy input. This is why the patient receives the highest degree of salience in the chain, which turns it into the ‘‘metonymic focus,’’ as Dirven (1999: 280) calls it. Apart from the patient, the other nonagentive roles in the action schema, namely, instrument and occasionally manner, can assume the metonymic focus; thus, for the fishing schema we find the conversions He was harpooning fish (< He was catching fish with a harpoon, focusing on the instrument) and He was fishing pearls (< He was collecting them like fish, focusing on manner). Another well-known event schema is the motion schema, for which Dirven singles out the goal element as the metonymic focus of conversion; he stresses, however, that motion must not be understood in a literal sense, but may evoke specific scenes, such as the notion of food preservation (as in to bottle), of shelter (as in to jail), of turning something into a specific shape (as in to slice), or of creating a new artifact (as in the verbs to book, to map,orto register). The third source of verb conversion is the essive schema, which describes class membership or a special attribute of the patient, as in He authored the book (< He was the author)orHe volunteered to give blood (< He was a volunteer). In addition to these basic schemas, there are combinations of the ac- tion and the essive schemas (reminiscent of combining a transitive and a subject complement pattern into an object complement pattern), which elevate to salience either class membership, as in He was knighted (< make someone a knight), or, much Figure 25.8. The role of ‘‘secondary’’ landmarks in the word-formation process of conversion (based on Twardsisz 1997: 93, 120) word-formation 665 more frequently, an attribute expressed by an adjective, as in clean the table (< make the table clean). A further aim of Dirven’s analysis, again in line with argument-structure dis- cussions, is not just to describe verb-creating conversions, but to explain what transformational linguists would have called constraints; to that purpose, he adds three more schemas, the experiencing, possession, and transfer schemas, and shows why they do not seem to be involved in verb conversion (see table 25.3). The reason they are not is that they feature human Experiencer, Recipient, or Possessor roles, which, just like the Agent role in an action schema, are the focus of attention and are thus excluded from triggering conversions. Metonymic focus, on which conversion is based, is therefore mainly reserved for nonhuman roles, unless hu- mans are assigned to Patient roles and treated like nonhuman entities, as in to author or to nurse (Dirven 1999: 285). What Dirven offers in his article is a consistent description, which also pro- vides a convincing explanation of the constraints of these word-formation patterns (and in which Twardzisz’s secondary landmark cases mostly reemerge as conver- sions based on Instrumental and Manner roles). 3.4. Noun-Noun Compounding Multilevel Templates and the Bottom-Up Approach Noun-noun compounding is another area of word-formation, where a long tra- dition of argument-structure-based analysis beckons for the use of semantic rela- tionships (see Levi 1978). 8 The problem is that it is difficult to explain all the existing compounds in such a framework, let alone predict the meaning of novel forma- tions. Here, Ryder (1994) offers a solution by providing a cognitive description that makes use of two of Langacker’s basic tenets, the notions of multilevel schematicity and of analyzability (see section 2.1). What Ryder suggests is that the explanation of noun-noun compounds should not be tied to a single, fairly general level of se- mantic relationships such as favored by Levi and others, but that it should consider Table 25.3. Event schemas used as input for noun-verb conversion (based on Dirven 1999) Event schemas and associated semantic roles action schema Agent, Patient, Instrument, Manner, Result location/motion schema Place, Source, Path, Goal essive schema Class Membership, Attribute experiencing schema Experiencer, Stimulus possession schema Possessor, Possession transfer schema Recipient, Beneficiary 666 friedrich ungerer a range of ‘‘templates’’ starting from the level of simple lexical items and rising to more general levels of semantic relationship, as shown in figure 25.9. Following Langacker’s notion of analyzability, this scale of schematicity should not be seen as a systematic classification of noun-noun compounds, but as a de- scription of the hearer’s bottom-up actualization, in which the hearer attempts to link the templates with suitable encyclopedic knowledge, which Ryder conceives as organized in event, entity, and feature schemas. This means that if hearers, upon encountering a compound, do not find the item already fully entrenched and if the context does not provide a plausible interpretation right away, they will Figure 25.9. Schema-based templates for the interpretation of noun-noun compounds (based on Ryder 1994) word-formation 667 proceed from the most specific to the most general template: from information supplied by the specific meanings of the constituents to core words, that is, words frequently occurring in compounds, and on to constructions linked to core words, and so on, until finally only the explanation contained in the general templates and the most general semantic information is left for accessing the compound— compare the bottom-up arrangements of the examples in figure 25.9. No doubt this permits a much more flexible approach than earlier attempts. To support her claims, Ryder conducted psycholinguistic tests in which participants were asked to define fabricated compounds, thus taking up a line of experimental compound analysis introduced by Downing (1977). The Attribute-Listing and Matching Approach Another psychological tradition that can be applied to compounding stems from the attribute-listing experiments initiated by Rosch for simplex lexical concepts (Rosch 1975). The basic idea is to collect the attributes named by participants when confronted with lexical concepts and to rank them according to frequency. Using this method for the study of compounds requires separate attribute listings for the constituents of the compound and for the compound as a whole. Comparison of these lists shows conceptual overlap and differences between the compound and its constituents and indicates where compounds attract ‘‘free’’ attributes from other domains. This method may help to distinguish between different stages in the lexical- ization process, as represented by the compounds apple juice, wheelchair , and news- paper. As illustrated in table 25.4, specifying compounds such as apple juice share most of their attributes with their constituents and attract only few ‘‘free’’ attri- butes. Enriched compounds like wheelchair draw on the attributes of their con- stituents to a lesser extent and mainly rely on free attributes imported from do- mains like ‘illness’ and ‘hospital’; this ratio between constituent-derived and free attributes becomes even more extreme in the case of fully lexicalized compounds like newspaper. The findings documented in table 25.4 also support the intuition that with partonymic compounds like coat collar, it is the first constituent (and not the second, as suggested by the traditional modifier-head analysis) that dom- inates the compound conceptually. To some extent, this shift of dominance even applies to nonpartonymic compounds like rain coat, where largely associative at- tributes like ‘wetness’, ‘bad weather’, ‘thunderstorm‘, and ‘umbrella’ suggest closer links with the constituent rain than with coat. 3.5. Lexical Blends: A Schematic Nonmorphemic View Making use of a new collection of neologisms and claiming to disregard the pro- cessing aspect, 9 Kemmer (2003) finds that a crucial aspect of a satisfactory account of the huge variety of blending patterns is a schematic view of shared phonological 668 friedrich ungerer similarities. In the case of the most frequent subgroup, overlapping blends, the overlap may involve shared segments of different size (dumbsizing from dumb and downsizing, glitterati from glitter and literati); of at least equal importance is the syllable structure the blend shares with one or both of the source lexemes (com- pare dumbsizing and downsizing). Substitution blends, in which one part of one source lexeme is substituted by a complete second source lexeme (as in carjack- ing ; from car x hijacking), can be seen as a subgroup of overlapping blends, while ‘‘intercalative’’ blends (Kemmer 2003: 72), in which the two source elements are not represented contiguously in the blend (as in Lewis Caroll’s famous coining slithy, from slimy and lithe), can be regarded as more marginal. All these variants find a place in Kemmer’s account, which she summarizes as follows: Phonological similarity of the blend with part or whole source lexemes increases the likelihood or felicity (the ‘goodness’) of a blend. Similarity can range from seg- mental identity through segmental similarity to same or similar syllable structure; and the similarity can range from identity/similarity of the blend with both source lexemes, to one source lexeme, or to parts of these. (Kemmer 2003: 75–76) That this phonological variation cannot be captured by a hard-and-fast rule is obvious; what is needed are generalizations which are flexible enough to cover more or less specific segmental elements as well as a syllable structure that have been repeatedly experienced by language users, and which have thus come to be ac- cepted as sanctioning a certain type of blend. For instance, for the blend swooshtika (< swoosh ‘the Nike logo’ þ swastika, with its Nazi associations), Kemmer posits a schema comprising the phonological form /swVS/ (V for vowel, S for sibi- lant), which is realized by swoosh, SWAStika, and SWOOSHtika alike; in addition, the schema includes the three-syllable structure of swastika, which is also taken up Table 25.4. Semantic attributes illustrating conceptual overlap and differences between compounds and their constituents (based on Ungerer and Schmid 1998; c1 ¼ first constituent, c2¼ second constituent, cpd ¼ compound; total of attributes refers to attributes listed by at least 4 out of a sample of about 30 informants) Type of Compound Specifying Compound Enriched Compound Lexicalized Compound Partonymic Compound ‘‘Associative’’ Compound apple juice wheelchair newspaper coat collar raincoat C1 þ CPD þ C2 4 002 1 C1 þ CPD 6257 8 CPD þ C292154 CPD ONLY 21421 7 6 TOTAL OF ATTRIBUTES 13 18 27 17 17 word-formation 669 . suggests two kinds of constraints: restrictions acting on the base of the -er items and re- strictions acting on the referent of the derivative (the target concept); see table 25.2. As far as the first. input. This is why the patient receives the highest degree of salience in the chain, which turns it into the ‘‘metonymic focus,’’ as Dirven (1999: 280) calls it. Apart from the patient, the other nonagentive. chains, we find that the energy flow is triggered by the head of the chain (the agent) and is transmitted to the second major element, the patient, which is affected, or even changed, by the energy input. This