1. Introduction
This article presents a corpus study of up as a locative prefixoid from a usage-based Construction Morphology (Booij Reference Booij2010) perspective. The present study expands Marchand’s (Reference Marchand1969: 109) claim that locative prefixoids function as different syntactic categories and relates this to a relational structure approach to morphosemantics (Gagné & Spalding Reference Gagné and Spalding2015), proposing a hierarchical network model for the polysemy of complex words with up. I combine this structural reading of complex words with locative prefixoids with the Construction Morphology notion of constructional idiom (Booij Reference Booij2010) to model surface generalizations encompassing the locative prefixoid up in words of all syntactic categories such as upstairs, upland, upheaval and uproot.
Locative prefixoids have only rarely been studied in detail, in particular up. The few exceptions commonly focus on prefix verbs alone and do not account for other syntactic categories (Scheible Reference Scheible2005; Schröder Reference Schröder2008, Reference Schröder2011; Kotowski Reference Kotowski2021, Reference Kotowski2023) or give only a broad overview of a group of prefixes (Marchand Reference Marchand1969; Lehrer Reference Lehrer1995; Bauer et al. Reference Bauer, Lieber and Plag2013). Over and out are considered the most productive locatives and have been subject to designated studies that also incorporate analyses of their semantics and selectional restrictions (Börger Reference Börger2007; Kotowski Reference Kotowski2021, Reference Kotowski2023). Most notably, Lieber (Reference Lieber2004) dedicates an entire chapter to providing a formalism for the semantics of locatives from an item-and-arrangement, morpheme-based and generative perspective and demonstrates the merit of such a model for over-. At the time of writing, no research has been carried out on the prefixoid up in all syntactic categories, let alone from a constructional perspective. In his study on verbal out-prefixation, Kotowski (Reference Kotowski2021: 84) concludes that constraint- rather than rule-based approaches such as Construction Morphology are promising starting points for the analysis of locative prefixoids. He emphasizes their capability of ‘spelling out complex and idiosyncratic semantic information and offer the possibilities of fine-grained meaning differentiation, as well as creating networks of related interpretations via hierarchies of schematic lexeme formation rules’ (Kotowski Reference Kotowski2021: 84). Complex words with up are highly polysemous and cannot be accounted for by a single schema. Though up is not as productive as out or over, it serves well to exemplify how modeling a hierarchical network from a Construction Morphology perspective and accounting for its semantics by means of a relational structure approach can be a profitable approach for the analysis of such prefixoids.
First, taking into account all syntactic categories solves a well-known theoretical problem, namely the subcategorization of complex words with locative prefixoids based on their syntactic category. Where prefix verbs are considered alone, converted forms such as upgrade V and upgrade N would be overlooked. This argument is also brought forward by Scheible (Reference Scheible2005: 185), who acknowledges that exclusively examining verbal prefixation disregards conversion and backformation. Categorization by syntactic category of the base such as proposed by the Oxford English Dictionary (OED n.d.) is similarly problematic since syntactic categories cannot always be unequivocally identified in complex words. Based on the righthand head rule (Williams Reference Williams1981), bases of prefixations are often categorized by the syntactic category of the derivative. For example, Bauer et al. (Reference Bauer, Lieber and Plag2013: 342) categorize coming in upcoming as an adjective and cropping in up-cropping as a noun due to the syntactic categories of the respective complex word (see also Kotowski Reference Kotowski2023: 120). Consequently, strict adherence to the righthand head rule would awkwardly classify the base in upgrade V and upgrade N as a verb and noun respectively. Obvious exceptions to this rule are frequently identified, e.g. in to out-absurd someone (Kotowski Reference Kotowski2021: 63), unless prior conversion is assumed to maintain theoretical consistency. Therefore, the present analysis will subcategorize complex words with up based on their polysemous meanings rather than their syntactic category. As will be shown, this approach offers a more nuanced perspective on the syntactic category of the base and complex word.
Marchand accounts for the polysemy of locatives by proposing three ‘preparticle types’ based on the syntactic category of the locatives that roughly correspond to three groups of output categories (list adapted from Marchand Reference Marchand1969: 109):
-
I. Particle with ‘adverbial force’ in verbs (outbid), participials (oncoming, downcast), verbal and deverbal nouns (outgoing, outlook), as well as some adjectives (overanxious)
-
II. Particle with prepositional properties in adjectives or adverbs (upstairs) and some nouns (afternoon)
-
III. Particle with adjectival properties in nouns (outhouse) and adjectives (intoed)
Marchand (Reference Marchand1969: 110) argues that verbal types (I) are derived from their particle verb counterparts. Prepositional combinations (II) are ‘morphologically characterized by the absence of the article in situations where syntactic conditions would require it’, e.g. upstairs from up the stairs, while the locatives in (III) are considered adjectives because some particles can be used as ‘full adjectives’ (Marchand Reference Marchand1969: 112).Footnote 1 As will be shown below, a constructionist approach is capable of providing an analysis of these complex words without relying on transformations. These schemas will be looked at in more detail below.
In combination with this structural reading, the constructional perspective proposed here contributes profitably to the debate about the status of words that have so far been termed prefixoids (see also Börger Reference Börger2007; Hartmann Reference Hartmann2019). Arguing that the semantics of prefixal up differ from its prepositional uses, Bauer et al. (Reference Bauer, Lieber and Plag2013: 340) treat up as a prefix. In contrast, Marchand uses the term ‘locative particle’ and refers to the resulting complex words as compounds due to the proposed syntactic category of the locative but mentions the possibility of classifying them as semi-prefixes (Marchand Reference Marchand1969: 108, 112). The notion of semi-prefix or prefixoid is capable of representing the ambiguous status of such morphemes when we view prefixations and compounds as being located on a continuum of autonomy and dependency on a schema (Tuggy Reference Tuggy2005: 245–8).
Construction Morphology posits a structural interpretation of affixoids by making use of the notion of constructional idiom (Booij Reference Booij2010: 73–6; Booij & Hüning Reference Booij, Hüning, Boogaart, Colleman and Rutten2014: 103). This is a morphological construction for complex words that contains at least one open slot for a base as well as at least one specified part. Being specified rather than a variable distinguishes the affixoid from compound elements and can account for bound meanings. In addition, the affixoid is co-indexed with the unbound word and bears syntactic category information, distinguishing it from affixes (Booij Reference Booij2010: 57; Booij & Hüning Reference Booij, Hüning, Boogaart, Colleman and Rutten2014: 90). Based on Booij’s (Reference Booij2010: 74) proposal for such a constructional idiom with out, we may posit the following schema for up:
The double arrow represents the interface between the formal and semantic side of the constructional idiom (Booij Reference Booij2010: 14). Lowercase subscripts co-index elements from the form- to the meaning-side of the construction; uppercase subscripts indicate syntactic categories, here mainly substituted by placeholders. Due to the ambiguous status of the affixoid, Booij (Reference Booij2010: 74) classifies out as an adverb. For up, following Marchand’s (Reference Marchand1969) hypothesis about the syntactic categories of locative prefixoids, the syntactic category of the prefixoid will be specified in the subschemas, whereas the mother schema above will only bear the syntactic category label PFX (prefixoid). The syntactic category of the variable X (i.e. the base) and the complex word also remain unspecified in the mother schema to allow for specification in the daughter schemas. The meaning-side of the construction posited above does not specify the relation between the first and second element in tandem with Booij’s (Reference Booij2010: 51) notation of an abstract compound construction, from which this constructional idiom inherits, and awaits further analysis.Footnote 2 Construction Morphology currently lacks an explicit morphosemantic theory, by convention relying on ad-hoc glosses to represent the meanings of constructions in a bottom-up fashion – Booij (Reference Booij2013: 256) acknowledges that the ‘semantic representation is a partially informal one’.
Crucially, the elements of a constructional idiom are defined as being related by a (semantic) ‘relation R’ (Booij Reference Booij2010: 74), as noted above. A psycholinguistic underpinning of this theoretical approach is the relational structure model of morphosemantics and conceptual combination (e.g. Gagné & Shoben Reference Gagné and Shoben1997; Gagné & Spalding Reference Gagné and Spalding2015). Rather than combining semantic properties of the individual component parts, this approach specifies a semantic relation between them (Gagné & Shoben Reference Gagné and Shoben1997: 72) by means of a relational gist structure (Gagné & Spalding Reference Gagné and Spalding2015: 23). Possible relations for NN compounds are a locative ‘in’ relation (mountain stream), an ‘about’ relation (mountain magazine) (Gagné & Shoben Reference Gagné and Shoben1997: 72) or a ‘made of’ relation (stone squirrel) (Gagné Reference Gagné2000: 385). Previous experience with different relational structures and the elements occurring in these structures determines the availability of different structures in conceptual combination (Gagné & Shoben Reference Gagné and Shoben1997). This approach is supported by experimental evidence (Gagné & Shoben Reference Gagné and Shoben1997; Gagné & Spalding Reference Gagné and Spalding2011) and is perfectly compatible with a Construction Morphology account as the relational structure between the elements can be specified by the construction; the elements themselves are variables in open slots or specified parts of the construction.
Like the examples above, these gist-based relational structures are frequently expressed through syntactic structure. Gagné (Reference Gagné2000) connects this relational structure model to earlier works on morphosemantics that compare the structure of compounds to syntactic structure: According to Teall (Reference Teall1892: 5, quoted in Gagné Reference Gagné2000: 366), complex words are formed ‘by omitting minor or connecting parts of a full expression, and using only the principle elements in more or less arbitrary association and frequently in inverted order’, e.g. hat box for ‘box of hats’ or book cover for ‘cover of a book’ (Gagné Reference Gagné2000: 366). In other words, ‘relations may function during conceptual combination, much like syntactic structures function during sentence processing’ (Gagné & Shoben Reference Gagné and Shoben1997: 72). Construction Grammar and Morphology provide a suitable framework to account for this connection since a construction encompasses different levels of form and meaning in the linguistic sign; syntactic and morphological structure is thus considered to have meaning (see, e.g., Goldberg Reference Goldberg1995). Other morphological frameworks likewise acknowledge the connection between syntactic and semantic selectional restrictions: ‘to the extent that particular affixes select for particular syntactic categories of bases, they also show concomitant semantic selection’ (Lieber Reference Lieber2004: 158; see also Plag Reference Plag2004).
Since Gagné’s relational structure model has only been applied to NN compounds and only provides relations for these structures, it is not directly applicable to the construction under investigation. However, the abovementioned analogy to syntactic structure enables a novel extension of the relational structure model to account for locative prefixoids. Marchand’s (Reference Marchand1969) typology can be applied to the construction at hand in conformity with Gagné et al.’s model by specifying new gist-based syntactic relational structures for constructional idioms with up. For example, to say that up in upstairs functions as a preposition is to say that this relational structure is one where up specifies a spatial relation of an entity to a path denoted by the nominal base. When up functions like an adjective in relation to nominal bases like side, this means that the prefixoid is a modifier denoting a property of the base. Up as an adverb modifying a verbal base (uplifting) specifies a direction of movement. Thus, Marchand’s method of subcategorization is here proposed to express the relational structure describing the semantics of these complex words by referring to the syntactic categories that up assumes as a prefixoid as well as in isolation.
This article is structured as follows: after describing the data and methods employed in the present study, I revisit Marchand’s classification of prefixoids and take an explorative approach using corpus data from the British National Corpus (BNC; 2001) and perform collexeme analyses to examine these subschemas. The aim of this method is to identify the lexemes that unify with the constructional idiom in order to formulate selectional restrictions of the base and describe the relation of up to the following element. I focus not only on the formal properties of the construction but also analyze semantic properties of prototypical and extended senses and calculate the productivity of the schemas. I conclude with a discussion of the advantages of a Construction Morphology approach and application of a relational structure model of morphosemantics to locative prefixoids.
2. Data and method
The usage-based approach taken here requires the use of corpus data to inductively draw conclusions based on a set of exemplars that is as large and balanced as possible. The data for the present analysis was retrieved from the BNC (2001), which contains 96,134,547 words from predominantly written sources of British English from the end of the twentieth century.Footnote 3 The corpus was accessed via Sketch Engine (n.d.; Sketch Engine: British National Corpus n.d.; Kilgarriff et al. Reference Kilgarriff, Baisa, Bušta, Jakubíček, Kovář, Michelfeit, Rychlý and Suchomel2014).
A search for words starting with up was conducted in the BNC.Footnote 4 False positives, i.e. types that are not instantiations of the construction and where the following element was not a recognizable word (e.g. upper, Upton), were identified by manual inspection and deleted. Orthographic variants with and without hyphen were counted as instantiations of the same type since hyphenation was found to be highly irregular (see Bauer et al. Reference Bauer, Lieber and Plag2013: 55). Instances where the word is simultaneously an instantiation of the up-x construction and some other morphological construction, e.g. derived nouns (upgradeability) or adjectives (uploadable), were counted towards the same variable as the corresponding non-derived types (grade, load).
A collexeme analysis, part of the family of collostructional analyses (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003), was conducted. This statistical test is the constructionist counterpart of a collocation analysis using a contingency table. However, instead of measuring the association strength between two words in syntagmatic relation, it measures the association strength between a lexeme and an open slot in a partly schematic construction. In this case, this measure describes the attraction of a base to the morphological construction. The advantage of this method over considering raw frequencies is that it takes into account the overall frequency of a word in a corpus (Gries Reference Gries2012):
Since collostructional analysis goes beyond raw frequencies of occurrence, it identifies not only the expressions which are frequent in particular constructions’ slots; rather, it computes the degree of association between the collexeme and the collostruction, determining what in psychological research has become known as one of the strongest determinants of prototype formation, namely cue validity, in this case, of a particular collexeme for a particular construction. That is, collostructional analysis provides the analyst with those expressions which are highly characteristic of the construction’s semantics and which, therefore, are also relevant to the learner. (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003: 237)
Collexemes occurring in the construction more often than expected are said to be attracted by the construction, those less often are said to be repelled. Ranking the variables according to their association strength reveals ‘constructional semantic prototypes’, based on which coherent clusters of meaning can be identified (Stefanowitsch & Gries Reference Stefanowitsch and Gries2005: 34), as these are the most strongly associated with the meaning of a construction (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003: 228). This allows for the identification of semantic classes of variables that cluster around one or multiple central member(s), characterizing constraints on the construction’s open slot, and enables a description of the meaning of the construction as a whole (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003: 211).
Hence, collostructional analysis is easily integrated into a usage-based account (Stefanowitsch Reference Stefanowitsch2013: 290), which assumes that frequent constructions, i.e. combinations of a schema with a base, form the basis for (analogical) extensions of the schema. For example, Bybee (Reference Bybee2013), analyzing the high token frequency of drive NP crazy, argues that the prototypicality of this specific type facilitates the coinage of new types with similar meanings to crazy, e.g. bananas or up the wall: ‘items with higher token frequency within the construction serve as central members of the categories that form for schematic slots within the construction. … Extensions of the construction will be based on these central members’ (Bybee Reference Bybee2013: 61). Conversely, selected examples of repelled and non-significantly attracted types will demonstrate that these deviate from the constructional semantic prototype. However, high-frequency types may also be entrenched as lexicalized chunks, not easily allowing for extension. This is a particular caveat when applying collostructional analyses to morphological constructions, requiring careful interpretation of the results.
The collexeme analysis was conducted in R using an open-source script developed by Susanne Flach (Reference Flach2021) with the log-likelihood value as the index of association strength, calculated by Dunning’s (Reference Dunning1993) log-likelihood coefficient. This is one of the most frequently used association measures (Flach Reference Flach2015; Hartmann Reference Hartmann2019) and regarded as more reliable than the χ2 test in dealing with small frequencies (Stefanowitsch Reference Stefanowitsch2020: 227). This is especially relevant since the construction under investigation is relatively infrequent in comparison to constructions at the level of syntax, which collostructional analyses have mostly been applied to. Still, a simple collexeme analysis also yields valuable results for constructions with lower type frequencies since this analysis is mainly meant to provide a ranking of collexemes, i.e. bases (Gries Reference Gries2012; Reference Gries2015: 520).
The bottom right cell in the co-occurrence table, which lists the number of constructions of the same class in the corpus that could potentially occur in this slot, has been the subject of much debate due to its uncertain reference (see Bybee Reference Bybee2010: 98; Gries Reference Gries2012; Schmid & Küchenhoff Reference Schmid and Küchenhoff2013: 541–4; Gries Reference Gries2015; Küchenhoff & Schmid Reference Küchenhoff and Schmid2015: 545–6 for a discussion). The point of reference taken here is not the number of verbs in the corpus, as is convention in analyses about constructions on the level of syntax, but the size of the BNC, representing the level of morphological constructions.Footnote 5
In order to compute the relative frequency of a variable in the construction, the overall frequency of the base in the BNC was determined. This includes the base in isolation as well as other complex words containing the base (for upriver, e.g. river, riverbed, riverbank). Variables were considered part of a polysemous category encompassing multiple syntactic categories (for upgrade, e.g. grade N, grade V). The search was adjusted to only show types containing the base with a frequency equal or greater than 10 to discount marginal instantiations and reduce the results to a manageable size for manual examination.
The significance of productivity for the constructional status of a pattern is much debated. Though unproductive patterns are not considered constructions in some Construction Grammars (Kay Reference Kay2013: 37–8), redundant and usage-based models (Hilpert Reference Hilpert2019: 67–8; Delhem & Marty Reference Delhem and Marty2020: 30) argue that constructions can still be posited even if they lack productivity. Construction Morphology (Booij Reference Booij2013: 258; Booij & Hüning Reference Booij, Hüning, Boogaart, Colleman and Rutten2014: 86) takes the same stance. Productivity is a crucial piece of information to be marked directly within the notational schema in related approaches like Relational Morphology (Jackendoff & Audring Reference Jackendoff and Audring2020: 41) since knowledge of whether a construction can be extended to new instances must be part of its mental representation. Since quantitative measures of morphological productivity are highly contested in the literature (Plag Reference Plag1999: 5–35; Bauer Reference Bauer2005: 315–16; Schröder & Mühleisen Reference Schröder and Mühleisen2010: 44), two measures will be calculated and subsequently compared. I will first determine the construction’s potential productivity, which is calculated by dividing the hapax legomena of a schema by the number of its tokens (Baayen Reference Baayen, Lüdeling and Kytö2009: 902). The second measure to be calculated is the type-token ratio, calculated by dividing the number of types by the number of tokens (Goldberg Reference Goldberg2019: 68). These ratios take into account the negative effect of highly frequent and entrenched types to the productivity of a schema. The schematicity or variability of the open slot, i.e. the range of words that are attested to unify with the construction, is another important factor in a construction’s productivity as it may restrict potential new bases (Bybee Reference Bybee2010: 95). The selectional restrictions will be identified through a qualitative bottom-up semantic analysis of existing types since restrictions are very fine-grained and not made apparent through applying an already existing typology, though Levin’s (Reference Levin1993) classification of verbs will be made use of where possible. Extended senses of the construction, which are instantiated by, e.g. conceptual metaphor or polysemy (Lakoff & Johnson Reference Lakoff and Johnson2003; Lehrer Reference Lehrer, Nerlich, Todd, Herman and Clarke2003; Bauer et al. Reference Bauer, Lieber and Plag2013: 641), will similarly be identified in a bottom-up manner and related to the concrete senses of the construction through metaphor or polysemy links (Goldberg Reference Goldberg1995: 72–81; Hilpert Reference Hilpert2019: 57–65).
3. The up-x construction
3.1. Preliminary remarks
Marchand’s classification of preparticle types can directly be applied to up but requires an additional schema to be able to account for all types. For the first mentioned schema, Marchand (Reference Marchand1969: 112) argues that the prepositional type is derived from a periphrastic phrasal expression, e.g. up the stairs in case of upstairs. Bauer et al. (Reference Bauer, Lieber and Plag2013: 344), too, argue that these words ‘are distributionally identical to prepositions or prepositional phrases’. However, from a constructionist perspective that favors surface generalizations (Goldberg Reference Goldberg2006: 25), we may assume that speakers have stored a surface pattern that can be extended to new instantiations without having to assume derivations from periphrastic variants – some instantiations cannot be accounted for by derivation (?up the breeze). However, it is easily discernible that this subschema exclusively selects nouns for its open slot: In the periphrastic variant, up is the head of a prepositional phrase, which prototypically embeds noun phrases headed by a noun, the second element of the construction:Footnote 6
The adjectival schema can easily be paraphrased, as well. Marchand (Reference Marchand1969: 121) suggests that this type can be subdivided into two schemas where up assumes different meanings, namely ‘upper’ (upside) and, as an extension, ‘going upward’ (uproad). Here, up modifies a noun:
In the ‘adverbial force’ type, the prefixoid approximates the function of an adverb in relation to the following verb, similar to particle verbs:
A function of up that Marchand does not distinguish is its verbal function meaning ‘raise’, which is also attested in unbound uses (up the ante). When up functions as a verb, the variable must have a nominal predication and describe the entity that is being raised, similar to a direct object with theme-semantics in a transitive construction:
The posited schemas also distinguish meaning:
The first examples are instantiations of the prepositional schema respectively: upslope (6a) and uptide (7a) construe the nouns as a path along which an entity moves. Upsloping (6b), as an instantiation of the adverbial schema, selects the verbal member of the polysemous category slope; up-tide (7b) is part of the adjectival schema since up functions like an attributive adjective to tide (‘high tide’).
Based on Marchand’s categorization and the fourth subschema suggested above, in figure 1 we assume a hierarchical network model of the up-x construction.
Figure 1. Hierarchical network model of the up-x construction
3.2. Data extraction
The search in the BNC for words starting with up yielded 263,046 tokens. After false positives and instances of unbound up were removed, 20,916 tokens remained. All tokens were then categorized into the four schemas given above by assessing which syntactic relational structure proposed by Marchand can account for their semantics, as shown in figure 2.
Figure 2. Distribution of tokens and types into the subschemas of the up-x constructional idiom
About half of the types and tokens found in the BNC instantiate the adverbial up subschema, accounting for prefix verbs (uplift) as well as verb-based nouns (upheaval, upbringing) and participles (uplifting, upset). Almost a quarter of the tokens and 10 percent of types are prefix verbs (uproot) and verb-based nouns (upgrade) of the verbal up subschema. The prepositional up subschema is instantiated by 21 percent of types and tokens, made up by mostly adjectives and adverbs (upstairs, upriver). The adjectival up subschema is the smallest of the four, made up by only 5 percent of the tokens but 17 percent of the types, which are mostly nouns (upland, upside). A total of 112 variables were found to unify with the constructional idiom, four of which occur in two schemas and have different meanings respectively.Footnote 7 Two types (upmost, uptight) with 98 tokens in total elude classification. I now discuss the four subschemas individually.
3.3. Prepositional up: type upstairs
This subschema is instantiated by 24 types, including 7 hapax legomena and 4,317 tokens in the BNC. Depending on their usage, instantiations of this subschema can function as adjectives (upstairs bedroom) or adverbs (go upstairs); some can also be used as nouns (the upstairs). Let us now turn to the nouns that are attested in this subschema and the strength with which they are attracted to the construction. Table 1 shows the frequency of each base in the corpus, its observed frequency in the up-x construction, its expected frequency in the construction if the collexeme were distributed equally in the corpus, the association (attraction or repulsion), the log-likelihood value indicating the collostructional strength as well as its significance.
Table 1. Results of the simple collexeme analysis for the prepositional up subschema
Some of the most significantly attracted and repelled collexemes suffice to illustrate the meaning of this schema. Stairs is not only the word with the highest raw frequency but it is also the word with the highest collostructional strength. This is not surprising, considering that stairs are prototypically vertical. The prefixoid up specifies the direction of the path along this vertical axis. What follows is that the variable must code some vertical path. Other types of this kind include uphill and upslope.
Verticality gives rise to metaphorical extensions (see Goldberg Reference Goldberg1995), here licensed by the good is up metaphor (Lakoff & Johnson Reference Lakoff and Johnson2003: 18):
Similarly to entities with concrete elevation like stairs or hill, market (8) and scale (9) are understood as metaphorical vertical axes.Footnote 8 Here, up describes the direction along a vertical axis of quality where the topmost point denotes the highest possible quality. This can be a reference to a higher point along this vertical path (8) or the topmost point along this path (9).
In contrast, there are three types which also have a high degree of collostructional strength and do not describe a vertical but a horizontal path. First, in instantiations like uptown, upcountry or upstate, up means ‘north’, which is motivated by the way that cardinal directions are typically represented on maps.Footnote 9 Consequently, nouns that occur in this subschema are bounded (political) locations with an unambiguous north on maps. Second, in upstream and upriver, the uppermost point is not the northernmost point but the source – a point located upstream is therefore closer to the source and simultaneously at a higher altitude. However, because moving upstream does not necessarily entail vertical movement, moving upstream more directly entails horizontal movement against the current. Cruse (Reference Cruse1986: 224) explains the experiential basis for this metaphor by arguing that ‘the direction of the tractive force of the water is seen as analogous to the pull of gravity’. The same applies to currents of air: upwind and upbreeze do not entail an elevated source; moving upwind also means moving against the wind. This cluster therefore selects moving bodies of water or air with a current.
Consequently, the fact that water is repelled by this constructional schema is not unexpected. The only instantiation in the BNC is from a poem:
Even creative instantiations often comply with the constraints of a schema.Footnote 10 Gathering from the context, upwater can be read as upriver. However, water is a mass noun that does not profile a source or current. The construction coerces water into a topographical interpretation as a moving body of water; concrete vertical movement such as movement from the ground to the surface of the water is not expressed.
The third extension of this schema that does not code verticality is instantiated by types like upstage and up-field. In upstage, the top of the stage is construed as the point farthest away from the audience. This is motivated because the horizon – and, in this case, the back of a stage from the perspective of the audience – appear farther up:
Whereas the only possible perspective of upstage (11) is that of the audience, the perspective of up-field (12), where field designates a playing field, is less fixed. It is typically that of the speaker, i.e. the audience or commentator, or the moving entity, i.e. the players on the field. Movement up-field entails movement towards the opponent’s half of the playing field and therefore farther away from the subject’s point of origin. Uplake is also an instantiation of this cluster, though repelled. This can be accounted for by the fact that a lake does not have any natural perspective to take, unlike stage or field. Here, up simply denotes a point farther away from the point of origin, which is understood through the context. Like with upwater, the base is coerced to fit the semantics of the construction.
Unlike the other types, upstage is also attested as a verb in the sense of ‘steal the show’, accounting for its high token frequency and high degree of collostructional strength. The verbal meaning is diachronically related to ‘up the stage’, since, according to the OED, it originally meant ‘[t]o move upstage of (another actor), forcing him to face away from the audience’. Synchronically, however, the verb is likely accessed whole by speakers and not fully compositional, mirrored by the fact that verbal use of this subschema is not attested for other types in the BNC.
Upfront and upsides are, despite the high frequency of the former, cases that are not sufficiently compositional. While formulating the periphrastic variant is not possible, their categorization as prepositional types is supported by the fact that they can be roughly paraphrased by the preposition at where the second elements act as reference points.
Based on the prototype upstairs, we can now posit the following schema with a prototypical meaning and meaning extensions, where each extension has its own constraints and selects variables from distinct semantic clusters:
Note that the syntactic category of the complex word is not identical to the syntactic category of the base, contrary to the predictions of the righthand head rule, but is specified independently. Constraints on the word class of the base emerge based on the assumed word class of the prefixoid or, in other words, the relational structure of the subschema. The prepositional subschema requires a nominal base for the reasons described above. Corpus data show that the complex words instantiating this subschema are mainly adjectives and adverbs. While distinct groups of output syntactic categories can be assigned to each of the subschemas (see Marchand’s original classification in section 1), they do not preclude idiosyncratic uses or conversions – for example, upstairs is often found in nominal and upstage in verbal uses.
3.4. Adjectival up: type upland
The adjectival up subschema is composed of 20 types, 7 of which are hapax legomena, and 1,006 tokens in total (see table 2). The relatively large number of low-frequency types can be taken as an indicator of a high degree of productivity, which will be examined below.
Table 2. Results of the simple collexeme analysis for the adjectival up subschema
Upland, the instantiation with the highest token frequency and the highest degree of attraction, is an instantiation where up means ‘upper’ (Marchand Reference Marchand1969: 121) and describes lands of high altitude (14). The same applies to the highly attracted upside, which designates the upper side of some entity, though most often used metaphorically as ‘advantage, positive side’ or in combination with down.
The question may arise why upland is not an instantiation of the upstairs subschema, considering that land also denotes a location. Two examples – one from the BNC and one from the Timestamped JSI Web Corpus 2014–2021 (Sketch Engine: Timestamped JSI Web Corpus n.d.; Bušta et al. Reference Bušta and Herman2017) – illustrate this:
In the BNC, upland (14) is not attested in the sense of moving ‘up the land’. In contrast, up-land (15) indeed makes reference to a path toward some higher part of the land and is attested in low frequencies in bigger corpora. We may consider this instantiation of the prepositional up schema evidence of its availability and the fact that land loosely fits the constraints of this schema. However, low frequencies are to be expected since land neither clearly encodes elevation like hill, licensing vertical movement, or is a clearly bounded (political) territory like country, licensing the north-interpretation.
The ‘upper’ subschema has a variety of metaphorical extensions. For instance, uptempo and uptime are licensed by heightened activity is up (Lee Reference Lee2001: 36). The repulsion of time may be due to the fact that uptime is a technical term used in programming. This metaphor or, alternatively, happy is up, may also account for the highly attracted upbeat. Uplevel is licensed by good is up where levels are construed as grades on a vertical scale.
The second subschema that Marchand (Reference Marchand1969: 121) notes, the ‘upwards’ schema, is connected to the ‘upper’ schema by a polysemous relation where the endpoint of the path is related to the path itself (Börger Reference Börger2007: 63). Instead of specifying an ‘upper’ part of an entity, up here describes an upward-directed entity. This subschema is instantiated by, amongst others, up-draught (16), uplight (17) and up-escalator (18). Though superficially similar to upstairs, up-escalator refers to an escalator going upwards, up here specifying a property of the base, rather than a higher floor of a building, construing the escalator as a path. Similar to the ‘upper’ schema, the nouns that unify with this subschema are very heterogeneous and do not allow for a delimitation of a semantic class for this open slot, resulting in high semantic variability. The similarities they share is that they are nouns that describe entities that extend in space and whose orientation can vary:
We can now posit the following schema with meaning extensions:
3.5. Adverbial up: types upheave and upset
The last type that Marchand identifies, which he terms ‘adverbial force’ and assigns to verb-related syntactic categories, makes up 50 percent of the overall type and token frequency of the construction. This cluster includes verbs, verbal nouns, deverbal impersonal and personal nouns, as well as past and present participial adjectives (Marchand Reference Marchand1969: 109). It is also the only subschema that selects verbs for its open slot. In the BNC, it consists of 58 types, 11 of which are hapax legomena, with 10,517 tokens (see table 3).
Table 3. Results of the simple collexeme analysis for the adverbial up subschema (types where verbal use only is attested)
Marchand (Reference Marchand1969: 109) only classifies those words as participial adjectives and verbal and deverbal nouns where the corresponding verbal use is not attested, such as downcast, oncoming and outlook. In the present article, too, nouns and participles that exist alongside a verb are considered part of the type ‘verb’, such as upload V, upload N and uploading N. This type is closely related to particle verbs: Marchand (Reference Marchand1969: 110) claims that participial adjectives as well as verb-related nouns are transposed particle verbs and Bauer et al. (Reference Bauer, Lieber and Plag2013: 442) argue that participial adjectives are created through inversion: For example, particle verbs in relative clauses (knees that are raised up) are easily rephrased as participial adjectives with prefixal up (upraised knees). I will only consider instantiations that also occur in verbal use (e.g. I upraise my knees) for the collostructional analysis to warrant a semantic analysis that is as little contaminated by particle verbs as possible. Though a constructionist account rejects derivations and focuses on surface forms (Goldberg Reference Goldberg2006: 25), an analysis in terms of overlapping paradigms forming the same participial adjectives may indeed be possible and remains a topic for future research.Footnote 11
Two polysemous subschemas become apparent, the first instantiated by the highly attracted upset, the second by upheave and uplift. The latter is the most semantically transparent despite its lower collostructional strengths. The collexeme that best exemplifies the semantics of this subschema is heave, which occurs predominantly in the derived noun upheaval but is also attested in verbal use. Heave can be classified as coding the cause of some motion without specifying a direction (Levin Reference Levin1993: 136) and relates ‘to the exertion of a force’ (Levin Reference Levin1993: 137). The direction of movement – an upward path – is specified by the first part of the construction. Both the verb heave as well as the meaning component of up involve physical effort for upward movement against the forces of gravity. Transitive and intransitive verbs of inherently directed motion like lift or rise (Levin Reference Levin1993: 114, 264) can also unify with this construction. Like Scheible (Reference Scheible2005: 191) notes, verbs that contradict the upward path specified by the prefixoid cannot unify with this construction (e.g. *uplower).
Thus, we can posit a subschema where the base codes the means of upward movement:
This accounts for the attraction of hold, which describes ‘prolonged contact with an entity’ (Levin Reference Levin1993: 145). The construction profiles a meaning of hold that includes an exertion of force against the forces of gravity, like the prototypical variable heave. In other words, an entity is held or sustained against an inverse force – gravity or social forces – though not moved upward.
The repelled variable in this subschema, build, as well as the least attracted collexeme chat do not describe upward movement, which may be taken as a reason for their lack of attraction to the constructional schema. In upchat, up approximates the meaning of ‘approach’. This type also exists as a particle verb (chat up). The classification of upbuild into the group of verbs must be taken with a grain of salt since the only occurrence in verbal use is a biblical quote:
The other exemplars of this type are verbal nouns and present participial adjectives. However, build as a verb of creation (Levin Reference Levin1993: 174) can readily be construed as the means of upward movement of the constructed entity.
Some of the most attracted collexemes, i.e. upset, upturn and upright, instantiate a polysemous extension of this subschema that does not entail upward movement:
Upset, upturned and uprighted entities do not experience vertical movement per se. Instead, only a part of the entity, here the legs of the table (22), are moved upward, resulting in the rotation of the entity. We can thus posit a polysemous extension of the kind pars pro toto:
The highly attracted upset exhibits limited compositionality and is primarily used metaphorically. The subschema’s opacity is reflected in its low variability and small number of high-frequency types, hindering the creation of new forms. This illustrates the challenge for collostructional analyses of morphological constructions: despite its strong association, upset is a poor central member or constructional semantic prototype based on which the schema can be extended, let alone sufficiently described.
3.6. Verbal up: type uproot
The fourth subschema comprises 12 types, 3 of which are hapax legomena, and 4,978 tokens in total (see table 4). It was argued above that up has verbal properties while the second element functions as a direct object with theme-semantics. Indeed, in cases where the second element may be a conversion, only the meaning of the nominal counterpart seems to be referred to. For example, the OED suggests that grade in upgrade is a noun. Similarly, as Marchand (Reference Marchand1969: 121) argues, upend originates from the phrase ‘up end’, meaning ‘turn the end upwards’.Footnote 12 Let us now turn to their collostructional strengths.
Table 4. Results of the simple collexeme analysis for the verbal up subschema
The most attracted collexeme with a non-metaphorical meaning is root, which lends itself well to illustrate the semantics of this subschema:
Though often used metaphorically, uproot has the concrete sense of ‘raising the roots (of a plant)’ (24). Despite the fact that both schemas output verbs, participials and verbal and deverbal nouns, the difference between the adverbial and verbal up schemas is that in the former, the variable describes the means of an upward movement whereas the variable in the latter describes the affected entity that experiences upward movement.
Metaphor links within this construction are licensed by, amongst others, upgrade, which is strongly attracted by the construction. The OED even defines upgrade along the lines of the schema proposed here, as ‘[t]o raise (something, esp. equipment or facilities) from one grade to another’. The fact that a higher grade is equated with higher quality is licensed by the metaphor good is up, which construes improvement as an upward movement. The same applies to upskill, uprate and upvalue. Similarly, upsize is licensed by more is up.
Having claimed that the variable is the entity being raised and has object-like status, the fact that some output verbs themselves allow for a direct object requires an explanation. Take, for instance, uproot (24). Here, the roots are a part of the uprooted dandelion. This also accounts for the attraction of upend, where the variable inherently codes a part of an unspecified whole. We can assume that the entity being raised prototypically describes a part of the direct object affected by the action. This is similar in metaphorical extensions, which are often motivated by their source domain, here physical entities: Variables in metaphorical instantiations like upgrade describe a characteristic or quality of the affected entity.
Update, though with the highest degree of attraction, does not align with this schema and is not fully compositional. The OED relates update to ‘up to date’. As a non-transformational account, Construction Morphology does not posit a derivation from up to date to update. Its categorization into this subschema is merited only due to the lexical category of the base and semantic interpretation of the complex word; it is the date that is affected by some action coded by up.
Upskirt, which is not highly attracted to the construction, is also an especially interesting case and can be argued to have the potential to belong to multiple schemas. The single occurrence attested in the BNC is an instantiation of the uproot pattern:
Here, being upskirted denotes the action of one’s skirt being raised by the wind; up has the non-metaphorical meaning ‘raise’, like in uproot. This example differs from modern uses of upskirt as the practice of taking photographs under clothing as a form of sexual harassment (26). This meaning of upskirt is an instantiation of the upstairs schema:
This instantiation can be paraphrased by ‘up the skirt’; the skirt acts as a vertical path along which the entity moves. The different subschemas of the up-x construction where the prefixoid functions like different syntactic categories can therefore select from the same semantic clusters of words but require different semantic relations between the prefixoid and the second element.
Taking these results into account, we can posit the following schema with metaphorical extensions:
4. Productivity
Quantitative measures of productivity can give a good estimation of the extensibility of the subschemas, as seen in table 5 and figure 3.
Table 5. Productivity measures of the subschemas in comparison
Figure 3. Type–token ratio and potential productivity of the subschemas
Quantitative measures of productivity cannot be compared with similar measures from different corpora, nor are there thresholds that would enable us to deem a subschema productive. Nonetheless, both quantitative measures used in the present study show the same tendency and can readily be accounted for by the semantic clusters identified above: the adjectival up subschema is the most productive of the four despite its relatively low type frequency. As was shown above, this subschema has a high variability as it selects from a heterogeneous group of nouns that can be imagined as a loose semantic cluster with high schematicity. This confirms the claim made by Marchand (Reference Marchand1969: 112) that word-formation with up in adjectival functions is fairly productive.
The prepositional up subschema, though with a slightly higher type frequency, is less than half as productive. The nouns that are selected by this subschema are less heterogeneous and more semantically similar since they must code a location, resulting in smaller semantic variability. Bauer et al.’s (Reference Bauer, Lieber and Plag2013: 344) argument in favor of the productivity of prefixal up with ‘topographical nouns’ is mirrored in the results of this subschema having a medium degree of productivity relative to the other subschemas – this schema can be extended, but only within the schema’s tight selectional restrictions. This is a case in point for theories arguing that schemas are productive when the category of existing and possible bases is relatively well defined (Bybee Reference Bybee2010: 69), even if their type frequency is relatively low (Barðdal Reference Barðdal2008). To illustrate, the category of bases of bounded political territories (uptown, upstate) is relatively small but is easily extensible using semantically similar nouns (e.g. up-province, JSI 208780758). This form of productivity arising from schemas with high semantic coherence and low type frequency is often termed analogy (Barðdal Reference Barðdal2008: 38; Bybee Reference Bybee2010: 69).
The adverbial up subschema is approximately as productive as the prepositional up subschema. It may be only moderately productive due to its relation to particle verbs, blocking the formation of prefixed verbs, which must remain subject for future research. As was described above, it selects verbs that can be construed as encoding a means of movement, making this class relatively open. As predicted by Barðdal (Reference Barðdal2008: 38), relatively open schemas with a high type frequency like the given subschema may be equally as productive as schemas with a lower type frequency and a higher degree of semantic coherence, like the prepositional up subschema.
Finally, as was to be expected, the least productive subschema is the verbal up subschema, which only has twelve types and a high token frequency per type, resulting in low variability.
5. Conclusion and outlook
As this article has shown, the locative prefixoid up has a wide array of idiosyncrasies that remain undetected when examined only in the context of prefix verbs. By carefully teasing apart constructional schemas with different morphosemantics and identifying polysemous and metaphorical meaning extensions of verticality, we can determine clusters of words whose properties allow unification with the construction.
Collostructional analyses have primarily been applied to syntactic rather than morphological constructions. A limitation of this method to complex words is that these are often not fully compositional when compositionality is defined as ‘the degree of predictability of the meaning of the whole from the meaning of the component parts’ (Bybee Reference Bybee2010: 45) rather than the mere recognition of its component parts. Marslen-Wilson et al. (Reference Marslen-Wilson, Tyler, Waksler and Older1994: 21) show that in cross-modal repetition priming tasks, synchronic semantic transparency between stem and prefixation is a prerequisite for priming and is therefore necessary to posit a link between collexeme and construction. Non-compositional types (e.g. upstage) may be stored as monomorphemic words and may not contribute to the schema’s productivity. High-frequency types, often ranked highly in collostructional analyses, may also be entrenched and accessed as lexicalized chunks, interfering with their being central members. Hence, a qualitative interpretation of the results and identification of possible constructional semantic prototypes is crucial. Even though collostructional analysis, particularly when applied to morphological constructions, faces challenges regarding the non-compositionality of highly attracted types, it remains an established method in constructionist research (see Hartmann Reference Hartmann2019 for an application to morphological constructions; more recently Van Goethem & Norde Reference Van Goethem and Norde2020; Hartmann & Willich Reference Hartmann and Willich2024) despite the harsh criticism it has faced (e.g. Bybee Reference Bybee2010; Schmid & Küchenhoff Reference Schmid and Küchenhoff2013; Küchenhoff & Schmid Reference Küchenhoff and Schmid2015; see rebuttals in Gries Reference Gries2012; Reference Gries2015). While ranking the types by simple frequency of occurrence, which is often cited as a preferable alternative (e.g. Bybee Reference Bybee2010), would also yield possibly non-compositional prototypes, considering expected frequencies of bases aligns with cognitive linguistic principles such as conditional probabilities and associative learning measures (Gries Reference Gries2012: 496). Ultimately, rather than claiming to describe psychological reality, the method aims to provide statistical support for a theoretical model informed by insights from cognitive and usage-based linguistics, such as frequency effects. Further psycholinguistic research is needed to address the compositionality of the up-x construction as well as the implications of this shortcoming for collostructional analyses in general.
As was shown by the instantiations potentially belonging to multiple subschemas, like upland and upskirt, the subschemas do not select from distinct semantic clusters but rather refer to distinct semantic components from their bases. This type of coercion has its limits, though, and this phenomenon is not very frequent in the data (see also Kawaletz & Plag Reference Kawaletz and Plag2015 for a discussion of how different polysemous meanings of derivatives select for a different pool of bases). Nonetheless, as suggested by a usage-based exemplar model (Bybee Reference Bybee2010: 88, 90), it is more profitable to account for the meaning of the subschemas as well as the selectional restrictions of the base in fuzzy clusters of meanings built around one or multiple central members rather than a single abstract meaning.
A constructional perspective on locative prefixoids has distinct advantages over derivational models. From a theoretical point of view, word-based approaches allow for the definition of properties of open slots of a subschema based on existing instantiations without positing a rule that is by definition productive (Jackendoff & Audring Reference Jackendoff and Audring2020: 32–3). In the present analysis, for instance, this allows us to define selectional restrictions for the verbal up subschema even without the existence of many novel coinages.
In addition, by combining the structural interpretation of prefixoids as specified parts of constructional idioms with Marchand’s (Reference Marchand1969) reading of prefixoids as functioning as different syntactic categories and relating this to a relational structure approach to morphosemantics, this article can profitably contribute to the question raised by, amongst others, Bauer et al. as to whether locative prefixoids are category-changing in adjectival premodifiers like extra-household activities (Bauer et al. Reference Bauer, Lieber and Plag2013: 338) or upstairs bedroom (see also Kotowski Reference Kotowski2021). It was shown that positing category change of the variable based on prefixation patterns or the righthand head rule or having to account for category change in upgrade V and upgrade N is not necessary. Though these models make correct predictions for the regular cases, it is a fundamental idea of Construction Grammar to posit a construction that is powerful enough to account for the exceptional, peripheral cases; the regular cases automatically follow (Fillmore Reference Fillmore1985; Fillmore et al. Reference Fillmore, Kay and O’Connor1988). Instead, we can acknowledge the necessity of a base of a certain syntactic category by referring to the syntactic relational structure between the elements. In the up-x construction, we can specify the syntactic category of the second element independently of the syntactic category of the complex word. More precisely, constraints on the word class of the base and likely output categories arise based on the semantic relational structure of the relevant subschema of the construction. Contrary to the assumptions of the righthand head rule, which emerges as too simplistic, there is a more complex interplay between the word class of output and base.
The interdependency of unbound and bound uses of up, where properties pertaining to semantics and syntactic category of the former recur in the latter, provide an interesting perspective on the discussion on the status of affixoids. By assuming that the syntactic category of the prefixoid determines the syntactic category of the following element based on semantics and by comparing morphological patterns to syntactic structures, we can also bridge the gap between morphology and syntax. While these similarities may be considered transformations and, in consequence, ‘a syntactic problem’ (Marchand Reference Marchand1969: 112), Construction Morphology can propose surface representations for these cases. In the future, the factors determining the choice between the constructional idiom and its periphrastic counterpart should be investigated in more detail. This applies not only to the alternation of prefix and particle verbs (uplift, lift up) but likewise to the other subschemas (e.g. upstairs, up the stairs). An approach focusing on alternations would provide a promising avenue for further research that would shed light not only on the constructions individually but also on the complex relations between them.
Acknowledgments
I would like to express my sincere gratitude to Alexander Bergs, Lara Höttecke, Meike Pentrel, three anonymous reviewers and the journal’s statistical consultant for their insightful feedback and valuable suggestions. All remaining errors are mine.
Open access
