construction within a language (Greenberg 1963). Moreover, there are more languages with postposed grams than there are languages with pre- posed grams (Svorou 1994). b. The boundedness of the relational gram. Relational grams may appear free as prepositions or postpositions or bound as prefixes and suffixes. While prepositions and postpositions exhibitlow degree of fusion with the landmark noun forming a syntactic unit with it, prefixes and suffixes may be either agglutinated or fused (Bybee 1985; Svorou 1994). Agglutinated relational grams retain their formal integrity without being affected by the morpho- phonological environment of the host, as in the Abkhaz example in (5). (5) Abkhaz (Hewitt 1979: 130) 1 d @ -s-p @ þn-g @ lo-wþp’ he-me-in.front-stand-stat ‘He is standing in front of me’ Fused grams are characterized by allomorphy conditioned either phonologically or morphologically. They can be exemplified by case affixes found in inflectional languages, like Ancient Greek where the dative case suffixes /-a/, /-e:/, /-o:/, /-i/, /-oi/, /-ais/, /-ois/, /-si/ have locative uses and the accusative case suffixes /-an/, /-e:n/, /-on/, /-a/, /-e/, /-o:/, /-a:s/ have allative uses. c. The host of the relational gram. Bound relational grams may be found in association with the landmark noun phrase or the verb of the sentence. Association with the landmark noun is characteristic of languages with dependent-marking morphology, whereas association with the verb is characteristic of languages with head-marking morphology (Nichols 1986). d. Internal structure of the relational gram. Relational grams vary with re- spect to the number of morphemes that constitute them. They may be monomorphemic, as the English prepositions in, from, and with, or they may be complex forms consisting of a general relational morpheme and a more specific morpheme, as in the English complex prepositions in front of, in back of, and instead of. The internal complexity of the gram is a function of the degree of grammaticalization of the gram, as well as its semantics (Svorou 1994). e. The syntactic relation of governance among the elements of the relational construction. In syntactic theory, the relational notion of ‘‘head’’ is an im- portant one. A head of a construction is the most important constituent that defines the character of the construction and determines the syntac- tic relations of governance, modification, and agreement. Within the gen- erative paradigm, adpositions are considered the ‘‘heads’’ of adpositional phrases. An adposition takes a complement noun phrase and, furthermore, governs the case marking of the noun phrase (Zwicky 1985). Within Func- tional Grammar (Dik 1997) and Role and Reference Grammar (Van Valin and LaPolla 1997), however, in relational constructions that encode an 730 soteria svorou argument of the verb, the noun is considered the head of a relational construction. The status of the adposition as the head of the phrase is then at least problematic. This is further complicated by the fact that languages may express some relational notions with adpositions and others with affixes bound to the noun. For example, Basque, in addition to a set of adpositions, expresses some general locative notions with suffixes, as ex- emplified by the suffix -n: (6) Basque (Houghton 1961: 7) lur-e-a-n etch-e-a-n earth-euph-art-loc house-euph-art-loc ‘on the earth’ ‘in the house’ Since there is semantic equivalence of adpositions and inflectional af- fixes in a given domain, it only makes sense to opt for treating them similarly. In fact, within recent versions of the generative paradigm, inflections are taken to be the heads of adpositional phrases in lan- guages with affixes, just like adpositions are taken to be the heads in languages with adpositions. This approach, however, ignores, on the one hand, the fact that in head-marking languages relational grams are attracted to the verbal head, leaving the landmark noun phrase by itself, and on the other, the diachronic facts, which point to a developmental relation between adpositions and affixes. Relational constructions in languages with head-marking morphosyntax defy constituency. Consider the following examples from Navajo and Abkhaz: (7) Navajo (Young and Morgan 1980: 96) ? an ´ t ? i ? yi-gha ´ - ? na ? fence it-through-crawled ‘He crawled through the fence.’ (8) Abkhaz (Hewitt 1979: 129) a-y o n e ` de-d o e ` þl-c’ @ –yt’ the-house he-out-go/come-Fin ‘He went/came out of the house.’ In both languages, the boldfaced relational gram appears as an affix on the verbal complex with the landmark noun phrase appearing by itself outside the verbal complex. Although the connection between the relational gram and the landmark noun phrase exists, it does not manifest itself in terms of syntactic constituency, at least not adjacency. In these languages, it is rather clear that the primary information- bearing unit (PIBU) is the landmark noun phrase and not the relational gram (Croft 2001). Such facts point to difficulties with considering the relational gram, be it adposition or affix, the head of the phrase. relational constructions in cognitive linguistics 731 Moreover, diachronic observations point to similar difficul- ties. Adpositions grammaticalize and eventually may fuse with the noun as case markers, as in the case of Hungarian, or appear as bound morphology, as in the case of Greek. (9) Hungarian (see Kahr 1976: 119) be ´ l- ‘innards’, bel-el ‘innards-abl’>bel € ol ‘from inside’ postposition c. 1350 (ker)ali magzot-bele(ul) tamadatia royal offspring-from bring.forth 16 c. -b ol is the most frequent form with bel € ol still appearing Modern ha ´ z-b ol ‘from the inside of the house’ Ke ´ z-b € ol ‘from the inside of the hand’ (10) Greek Eis /es/ > /s/ (preposition) Eis tin polin > s-tin poli (cliticization on the article in noun phrase initial ‘To the city’ position) In the process of grammaticalization, the morphosyntactic status of the adposition changes, affecting the whole construction. Adposi- tions seem to behave more like heads in the earlier stages of gram- maticalization, but as they grammaticalize, they may fuse with the noun, thus yielding headhood to the landmark noun. The cross- linguistic picture is characterized by morphosyntactic variation be- cause a functionally equivalent construction, for example, the loca- tive construction, may be in different evolutionary stages. Also, the type of locative construction may be a determinant for the degree of grammaticalization that a construction may reach even across languages. (See section 4 for a more detailed discussion of these points) Relational constructions also express a great variety of relational notions, or of semantic variation. I have compiled a list of meaning components that feature as uses of spatial grams (Svorou 1994). They fall into the following categories: a. Locations (e.g., interior, superior, under, lateral, middle, posterior, con- tiguous) b. Directions (e.g., allative, ablative away, down, up, along, circumferential, through) c. Distance (e.g., proximal, distal) d. Spatiotemporal relations (e.g., anterior-order, posterior-order) e. Temporal locations (e.g., interior-temporal, durative, end point of situation) f. Aspectuals (e.g., continuous, inceptive, every) g. Relations among situations (e.g., purpose, reason, concessive, reality condition) 732 soteria svorou h. Valence (e.g., benefactive, instrumental, source, recipient, comitative, dessive) i. Manner (e.g., comparative, incremental, suddenly, punctual) j. Predicative relations (e.g., sociative, exchange, possessive, partitive, material) k. Conjunctive relations (e.g., inclusion, coordinative) This list constitutes a classification of uses of relational grams that have as one of their uses a spatial one, be it the primary or not. It was created by comparing grammatical forms of the spatial domain across languages. A different way of inquiry is to look into how a particular language expresses a certain domain, for example, location. This view has been adopted by Levinson (1991, 1994, 1996, 2003) and his colleagues of the Language and Cognition Group (Pederson et al. 1998; Levinson, Meria, and the Language and Cognition Group 2003). In studying how different languages express spatial relations, Levinson points out that some languages employ lexical means and others grammatical means for describing the same spatial scene. Such differences may be due to the means avail- able in the language. The findings of this research have been interpreted as sup- porting the linguistic relativity hypothesis. 3. Semantics of Adpositions Early studies of relational grams focused primarily on prepositions in European languages, with particular attention to English (Lindkvist 1950, 1972, 1976; Heaton 1965; Wood 1967;Hill1968;Bennett1972, 1975) and French (Poitier 1961, 1962), although occasional studies of relational grams in other languages exist, for example, Buck (1955) on Mongolian postpositions, Casad (1975) on Cora locationals and di- rectionals, and Friedrich (1969a, 1969b, 1970) on Tarascan suffixes of space. Although conducted within various theoretical frameworks, a common thread to these studies is that they are concerned with providing lists of uses of particular relational grams, with some attempts at analyzing the uses in terms of binary semantic components (for example, Bennett 1975). 2 And while these lists of uses were a sure sign of the polysemous nature of relational grams, it was not until the advent of Cognitive Linguistics that their polysemy was seen as their most interesting and challenging semantic characteristic; witness, for instance, the introduction of polysemy networks as interrelated networks of uses around a prototype and the investigation of the relation between relational grams and context—as relational grams lend themselves to use in various contexts transforming to adapt to the semantic nuances required by the context. Furthermore, in most of these studies, the focus of analysis was on how to represent the linguistic aspects of relational grams. A few studies recognized the need to look at areas beyond the system of language for insights into the structure of relational constructions in cognitive linguistics 733 linguistic categories of relational domains such as space and time. Among them, Talmy’s early work (1972, 1975, 1978) as well as Miller and Johnson-Laird (1976) figure prominently. These works sketched out how language and cognition interact in the expression of relational notions involving motion, space, and time. A seminal study for the treatment of relational categories and for Cognitive Linguistics in general is Talmy (1983). In this study, Talmy provided a sketch of how language structures spatial categories along with the levels of analysis we need to consider for understanding this mapping of language with the cognitive cate- gory of spatial orientation. Talmy’s proposals are in line with the views of Hers- kovits (1982, 1986), Jackendoff ( 1983), Lakoff (1987), Langacker (1987, 1990a, 1990b, 1991), and Vandeloise (1991) in that relational grams, such as prepositions, express construals of the experiential situation by a conceptualizer. Speakers, as concep- tualizers, use a number of mechanisms to construe a scene linguistically. Although it is not clear whether these mechanisms are processing mechanisms or mecha- nisms that lead to the creation of lexical and grammatical patterns in language, as Rice, Sandra, and Vanrespaille (1999) have pointed out, they can affect the lin- guistic encoding of a scene. Some of the proposed mechanisms are schematization, perspective, and idealization and abstraction, and are discussed below. A fundamental consideration for the analysis of meaning is the understanding that language involves schematization, ‘‘a process that involves the systematic se- lection of certain aspects of a referent scene to represent the whole, while dis- regarding the remaining aspects’’ (Talmy 1983: 225). This process leads to the ad- justment of the level of specificity employed by the conceptualizer. Compare This is here to The book is in front of me. Individual forms of language, such as relational grams, represent particular schematic abstractions called ‘‘schemas.’’ Discussions on the nature of schemas elaborating relational grams have centered on spatial grams. Are spatial grams and the schemas that underlie them characterizable in terms of geometric, topological properties or in terms of functional properties? Ear- lier studies have stressed the geometric properties of language (Bennett 1975; Miller and Johnson-Laird 1976; Herskovits 1982, 1986; Talmy 1983). In this view, linguistic forms encode aspects of the geometric relations that exist between and among objects as if these relations existed in a world in which conceptualizers were only perceivers and not users of objects and experiencers of usage events. Each form is associated with a set of spatial primitives such as ‘enclosure’ for the representation of English in or ‘spatial contiguity’ for the representation of En- glish on. Certain functional concepts, such as ‘region’, proposed by Miller and Johnson-Laird (1976), or ‘support’, which is included in Herskovits’s (1986) and Miller and Johnson-Laird’s (1976) accounts, appear only as secondary or in sup- port of the more central and important geometric properties. Later studies provide evidence for the importance of functional, in addition to geometric, properties of spatial descriptions; that is, they include knowledge that is gained from our functional experience with objects (Vandeloise 1986, 1991; Talmy’s 1988). Casad and Langacker (1985) provide an analysis of two Cora particles, u ‘inside’ and a ‘out- side’, in terms of topological as well as functional properties such as ‘accessibility’. 734 soteria svorou Vandeloise (1986) proposes that the French prepositions are best explained in terms of functional relations such as ‘containment’ and ‘support’. Exploring the attempts to describe the French preposition dans as well as equivalent prepositions in En- glish, Dutch, and German, Vandeloise (1994) points out the difficulties that these geometric or topological proposals have and instead proposes the notion of ‘force’ as the controlling factor in the choice of preposition to describe a spatial scene. This notion has to do with the force that the container exerts on the content and is very similar to what Garrod and Sanford (1989) and Garrod, Ferrier, and Campbell (1999) have called ‘‘location control.’’ Coventry, Carmichael, and Garrod (1994) and Garrod, Ferrier, and Campbell (1999) provide experimental evidence for functional categories of containment and support associated with the English prepositions in, on, over, and beside, according to which speakers would rely on functional information for location control in cases where the spatial scenes were less than prototypical examples of containment and support. This finding led them to the conclusion that the semantic representation of prepositions should include geometric as well as functional information. This view of schemas as rich with both geometric and functional information, put forth primarily by looking at the se- mantics of spatial grams in very few languages, is further corroborated by my cross- linguistic comparison of spatial grams (Svorou 1994). Conceptualizers have the ability to view a spatial scene from different per- spectives, thus producing different linguistic descriptions reflecting these perspec- tives. For example, the sentence The apple is in the plate presents a particular con- strual of an imagined or actual situation in which the apple and the plate, although both statically present in the scene, are presented asymmetrically. The apple is construed as the trajector (Figure), the entity to be located, and the plate as the landmark (Ground), the entity with respect to which the trajector is located. In the possible, but unlikely, sentence The plate is under the apple, the plate is construed as the trajector and the apple as the landmark. The latter construal is unlikely or rare because the larger in size plate can generally function as a better landmark than the smaller in size apple. The spatial and temporal relational domains are two of a number of other relational domains that exhibit this asymmetric organization, such as possession and causation. The meaning of relational grams can be best analyzed by considering the role of the conceptualizer and the role of the ground landmark. In order for a reference object to be construed as a landmark, it has to un- dergo a process of idealization and abstraction by which ‘‘familiar objects, in all their bulk and physicality, are differentially ‘boiled down’ to match ascribed schemas’’ (Talmy 2000: 220). Other aspects of the objects are ignored. For example, the plate is idealized as a container in The apple is in the plate but as a surface in The apple is on the plate matching the requirements of the in schema versus the on schema. Moreover, encoding a spatial scene with a relational gram of con- tainment (in) requires the idealization of the landmark as a container, whether it has prototypical characteristics of containers (in the cup, in the lake) or not (in the rain). relational constructions in cognitive linguistics 735 Schemas have been employed starting with Brugman (1981) to represent the multiple uses of prepositions, such as English over. These uses, distinguished into more or less prototypical, have been related to one another in the form of a poly- semy network, a configuration of the various senses of a lexical item such as a preposition. While speakers have a good idea of what constitutes a use (Colombo and Flores D’Arcais 1984), what constitutes a use for the analyst may be contro- versial (Sandra and Rice 1995). Still, this particular solution to the problem of po- lysemy gained in popularity and led to a number of analyses of relational grams across languages in terms of polysemy networks, for example, Vandeloise (1986, 1991, 1994) and Cuyckens (1993b) for French, Cuyckens (1991, 1993a) and Geeraerts (1992) for Dutch, Rice (1996) for English, Bacz (1997) and Dancygier (2000) for Polish, Smith (1993) for German accusative and dative cases and Bellavia (1996) for German € uber, Delbecque (1996) for Spanish para and por, Schulze (1993) for English around, Dewell (1994) and Kreitzer (1997) for English over, and Kristoffersen (2001) for Norwegian mot. In these studies, the network consists of a central, prototypical schema and a set of related senses in a particular configuration. The granularity of analysis of networks and the level of schematization of the prototype continue to challenge cognitive linguists, as the re-telling of the story of over in Tyler and Evans (2001, 2003) indicates. What is generally held and argued for, on the theoretical level (among others, Brugman 1981; Lakoff 1987; Rice 1992) and shown on the experi- mental level (Rice, Sandra, and Vanrespaille 1999), is the validity of a polysemic approach to the representation of relational grams in contrast to a monosemic approach (Ruhl 1989). The relations among the senses in a polysemy network are seen as a result of semantic extension from a prototypical schema. On the synchronic level, similar to the diachronic level to be discussed in section 4, the processes that lead to semantic extension are generally believed to be metaphor and pragmatic inferencing (Lakoff 1987; Schwenter and Traugott 1995, among others). Reflecting localistic accounts of cases (Anderson 1971), space is considered to be primary and the basis of semantic extension to domains such as time and causality (Radden 1985). The extension of originally spatial grams to the domain of time has been seen by many as a result of metaphor—the time is space metaphor (Lakoff and Johnson 1980; Claudi and Heine 1986; Lakoff 1987). Indeed, this type of extension is responsible for the de- velopment of the majority of temporal grams in many languages, as Haspelmath’s (1997) detailed study on a sample of fifty languages has shown. He concluded that the time is space metaphor is in fact universal. Despite the alleged ubiquity of this metaphor, the question remains whether it is a live metaphor in the minds of the speakers that they use in the online construction of meaning or whether it is an epiphenomenon of historical processes that has resulted in the development of temporal grams out of spatial grams and is no longer available to the consciousness of the speakers. This question has been addressed in an experimental study by Rice, Sandra, and Vanrespaille (1999), who found that for many English and Dutch pre- positions the time is space metaphor is fading away in the minds of the speakers. Other recent studies have shown semantic extension to be the result of pragmatic 736 soteria svorou inferencing or context-induced reinterpretation of the meaning of grams on the synchronic (Tyler and Evans 2001, 2003) as well as diachronic level (Svorou 1994; Schwenter and Traugott 1995; Heine 1997). If semantic change takes place by con- ventionalizing inferences induced by the context in which a relational gram appears, then the same mechanism should be available for making online adjustments of meaning on the synchronic level. In this view, then, metaphor is the result of the process, and not the process itself. The above issues on the semantics of relational grams (the mechanisms of construal, polysemy networks, and semantic extension) have been investigated in many studies, and yet many questions remain on some of the issues raised. One such issue has to do with whether to treat meaning as associated with individual forms in the lexicon or whether to treat meaning as associated with individual forms in specific constructions. Attempts to analyze the meaning of adpositions in the 1970s have concentrated on the meaning of prepositions in languages like English and have assumed that the preposition by itself is the form that provides all the relational meaning in a rela- tional construction such as the locative construction. For example, in The baby is in the tub, the preposition in, and nothing else, is responsible for specifying the spatial relation between the baby and the tub. The work of cognitive linguists, however, has shown that elements other than the preposition may contribute to the construal of a scene. Moreover, as discussed above, the various senses of a relational gram come from the context in which it may be found, and before they can be recognized as uses of a certain gram, these senses are emergent or latent in particular sentences. In order for the meaning of the gram to be extended, creating a polysemous structure, elements of the context have to contribute to the new meaning of the gram. In other words, the new sense of the gram is a result of rearranging semantic bits within a construction. A prerequisite for semantic extension, then, is the existence of context for the interpretation of grams and the assumption that meaning is distributed over several elements in a construction. For the locative construction, this view has been advocated by Sinha and Kuteva (1995) for English, Dutch, Japanese, and Bulgarian and by Ameka ( 1990, 1995) for Ewe. It finds support in studies of various languages, and it is purported to be a universal. Some examples follow. German two-way prepositions, like in, hinter, unter, auf, and an, exhibit a va- riety of senses depending on whether the landmark noun carries accusative or dative marking (Smith 1988, 1993). The senses go beyond the traditional motion versus location interpretation to include senses motivated by these prototypical mean- ings. In all cases, the construal of the spatial situation is a function of the specific preposition used, the landmark noun, and the case marking of the landmark noun. Similarly, Polish (Dancygier 2000) uses three interacting substems to express spatial information: (i) Direction nouns, such as g ora ‘up’ and dol ‘down’ or prz od ‘front’ and tyl ‘back’, describe regions in space or regions of landmarks. They combine with prepositions and appropriate case markings to indicate various spatial construals such as direction of movement, location of objects with respect to the observer, and so on. (ii) Prepositions, such as w ‘in’, which combines with the relational constructions in cognitive linguistics 737 accusative or the locative, or do ‘to’ and od ‘from’, which combine with the genitive, indicate dynamic spatial construals such as path/goal or source/path. (iii) Case markings, including genitive, locative, accusative, and instrumental, usually com- bine with either prepositions or direction nouns to provide spatial information, except for the instrumental, which can mark landmark nouns for the expression of path and path-related notions. In any given construction, it is the interaction of all or some of these subsystems with the specific landmark noun that determines the specific construal of the situation at hand. A relevant issue in the discussion of the interaction of preposition and case in case languages is the question of whether the preposition or the case is the primary element, the one that ‘‘governs’’ the other. Bacz (2000) provides a succinct dis- cussion of this issue as far as Polish data goes. The Polish data pose a problem for any monolithic account of preposition and case because different prepositions re- late to case in different ways. She proposes a solution to the problem by adopting Kuryłowicz’s historical explanation whereby the role of the preposition changes throughout the history of the locative construction. That is, a preposition can be said to govern a case when that preposition can be combined with only one case. Prepositions that can be combined with a variety of cases are considered to be at an earlier historical stage where the case morpheme plays a more fundamental role in determining the semantics of a construction. The historical perspective of adpo- sitions in general is discussed in section 4 below. This ‘‘distributed’’ view of meaning constitutes a shift from previous accounts that have focused on the meaning of a single element, such as a preposition, without regard of the elements that it collocates with frequently and the influence of such elements on the meaning of the preposition. Such accounts are partly a result of analyzing languages with no significant case systems and partly of analyzing pre- positions as being the head of the construction they participate in. 4. The Grammaticalization of Relational Grams In the past two decades, a cross-fertilization of functional and cognitive approaches with diachronic and cross-linguistic perspectives gave us a deeper understanding of the nature of relational grams, illuminating their semantic as well as their formal aspect. A number of language specific and cross-linguistic studies prepared the ground by providing the seed ideas. Papers by Traugott (1975, 1978) and Kahr (1975, 1976) investigate the histori- cal development of spatial adpositions, locationals, and case markers. Specifically, Traugott (1975, 1978) and Kahr (1975) point to body-part terms and relational nouns as the ultimate sources of spatial adpositions in many different languages. In 738 soteria svorou these studies, the focus is on the description of the formal identity and semantic similarity of spatial adpositions, on the one hand, and body-part nouns and rela- tional nouns, on the other. Kahr (1976) provides evidence for the historical processes underlying synchronic case systems in languages from five families, documenting the postpositional origin of case morphemes. A common thread in all these studies is the observation that the historically old and new forms and functions may coexist at a particular synchronic period, blurring the distinction of grammatical categories and of the lexicon and grammar. These cross-linguistic studies complemented by some language-specific studies, such as Friedrich’s (1969a, 1969b, 1970) studies on Tarascan spatial suffixes, point to the historical relation between lexical sets, such as body-part terms and spatial adpositions and affixes. Nominal expressions are not the only source of adpositions. In a number of African and Asian languages, verbs in certain constructions, such as serial-verb constructions and participial constructions, function as ‘‘co-verbs,’’ or ‘‘verbids,’’ that is, as adpositions but with some verbal characteristics. Studies by Li and Thompson (1973, 1974)onChinese,Givo ´ n(1975) on Niger-Congo, and M. Clark (1978)on Vietnamese provide evidence for a verbal source of adpositions. As the 1980s approach, the following statements can summarize the received knowledge: a. Prepositions and postpositions have their historical sources in two lexi- cal categories: (i) nouns expressing body parts or relational notions and (ii) verbs of motion or existence. b. Postpositions may be the source of case morphemes. c. Prepositions and postpositions may coexist synchronically with their nom- inal or verbal sources resulting in blurring of grammatical distinctions and fuzziness of categories. Although there was evidence for the historical relationship of adpositions to lexical sources such as nouns and verbs, there were few satisfactory ideas as to how to account for this relationship in an explanatory framework. The theoretical frame- work in vogue, the generative paradigm, was embracing a separatist view of grammar and lexicon and any historical connections were viewed as a result of some ‘‘radical reanalysis’’ (Lightfoot 1979). In the case of adpositions, this approach fails to explain why several languages could have the same form functioning as a noun and an adposition or verb and adposition at the same synchronic point, if any ‘‘radical reanalysis’’ were to take place. The most promising ideas were put forth by Givo ´ n’s (1971, 1975) proposal of looking at the process of grammaticalization to understand the synchronic facts, as in the case of adpositions rising out of serial verbs. Givo ´ n’s proposals, corroborated by Lehmann’s (1982) thoughts on gra mmati - calization and Heine and Reh’s (1984) study on grammaticalization in African lan- guages, were the beginning of a renewed interest in the development of grammati- cal material. Although grammaticalization was neither a new concept nor a new phenomenon 3 —Kuryłowicz (1975: 52) had already discussed it as ‘‘a process which relational constructions in cognitive linguistics 739 . svorou argument of the verb, the noun is considered the head of a relational construction. The status of the adposition as the head of the phrase is then at least problematic. This is further complicated. morpheme, as in the English complex prepositions in front of, in back of, and instead of. The internal complexity of the gram is a function of the degree of grammaticalization of the gram, as well. In order for the meaning of the gram to be extended, creating a polysemous structure, elements of the context have to contribute to the new meaning of the gram. In other words, the new sense of the gram