Resultative secondary predicates in cooking recipes: an empirical study of Germanic and Romance languages

1.1 Germanic resultative constructions

Most prior research on RCs has investigated these constructions in Germanic languages, and more particularly in English. The reasons behind this interest can be summarized as follows. First, English RCs exhibit a wide range of distinct syntactic patterns: they can contain transitive verbs, with subcategorized (cf. [1]) and non-subcategorized objects (cf. [4]), and intransitive verbs, both unaccusative (cf. [3]) and unergative ones (cf. [5]) – also used with non-subcategorized objects.^[2] In other words, these constructions can overrule the valency and the selectional restrictions of the verb: sweep is usually restricted to object noun phrases (NPs) that refer to a surface, and not an instrument (e.g., Sue swept the floor/*the broom) and bark cannot select for a direct object since it is intransitive (e.g., *The dog barked the chicken).

Several analyses have been proposed to account for sentences like (4) and (5). For instance, in ‘Small Clause’ (or SC) analyses (Aarts 1992; Bowers 1997; Chomsky 1981, 1986; Hoekstra 1988, 1992; Hornstein and Lightfoot 1987; Stowell 1981), non-subcategorized object NPs are said to be licensed as subjects of the SC, which is itself lexically selected by the main verb (e.g., The dog [_V barked [ _SC the chickens [_AP awake]]]).^[3] Alternatively, in constructional approaches (Boas 2003; Fillmore et al. 1988; Goldberg 1995; Goldberg and Jackendoff 2004; Iwata 2020), non-subcategorized NPs are argued to be licensed by the (transitive) resultative construction, which is conceived of as a conventionalized pairing of syntactic form (viz. [NP₁ VP NP₂ XP₃]) and meaning (viz. X₁ cause Y₂ [become Z₃] by V-ing). Under this view, the postverbal NP (e.g., the chickens) and the resultative AP (e.g., awake) are solely contributed by the construction, and not by the verb.

Second, English RCs can modify the verb’s aspectual reading (Vendler 1957, 1967), converting activities (i.e., durative and atelic events) into accomplishments (i.e., durative and telic events) (Beavers 2012; Dowty 1979; Rothstein 2004; Tenny 1994). Thus, in (1), the RSP clean provides an endpoint to the event denoted by the verb sweep (i.e., the sweeping event ends when the state of being clean is reached). In the literature, aspect-shifting RCs like (1) are regarded as instances of ‘strong’ RCs (Washio 1997). They are contrasted with what is referred to as ‘weak’ RCs. These RCs contain verbs that denote a telic event (for instance, an accomplishment, e.g., paint in [2]) or an achievement, e.g., break in [3]). Therefore, in such contexts, the RSP only specifies or intensifies a result already incorporated in the verb semantics.

And third, English RCs are also known for their puzzling distribution (Beavers 2002, 2013; Boas 2003; Cappelle 2014; Carrier and Randall 1992; Rodríguez Arrizabalaga 2003; Wechsler 2001, 2005; Wyngaerd 2001). To put it simply, while some “verb + RSP” combinations are allowed, others are not. For instance, the verb drive only accepts RSPs that refer to a pejorative mental state, which include crazy, bonkers, over the edge, and to the brink of lunacy but not happy and to the brink of ecstasy, which are hence ruled out in this context (cf. [6]). The RSP is not only subject to lexical restrictions but is also limited in terms of morphosyntactic categories: thus, with pound, PPs are allowed but not NPs (cf. [7a]). The NP the pancake must be introduced by a preposition. However, with paint, and more generally with chromatic change verbs, it is the other way around: NPs are accepted but not PPs (cf. [7b]).^[4] A similar contrast is observed in example (8): unlike shoot, which can occur with both the AP dead and the PP to death (cf. [8a]), beat can only occur with the PP (cf. [8b]). Given that the RSPs involved in each pair in (7) and (8) are semantically related, one can assume that the choice between one or the other RSP morphosyntactic category largely depends on the verb subcategorization.

Note, however, that the distributional facts illustrated in (8) have also been explained on semantic grounds (Beavers 2002, 2013; Rodríguez Arrizabalaga 2003; Wechsler 2001, 2005; Wyngaerd 2001). The occurrence of either dead or to death in the sentence is argued to correlate with the verb’s aspectual reading: while dead tends to occur with verbs that denote punctual events (e.g., shoot, knock dead), to death tends to appear with verbs that describe durative events (e.g., beat, stab, strangle to death). Interestingly, when used with the PP, the verb shoot receives an iterated reading (i.e., the outlaw shot the miller several times), which is not available with the AP. The correlation between the verb’s aspectual properties and the formal encoding of the RSP will be further explored in Section 4.4.

English RCs have also been studied from a comparative perspective. Most studies compare English with Romance languages (Legendre 1997; Mateu 2012; Mateu Fontanals 2000; Napoli 1992; Rodríguez Arrizabalaga 2003, 2014, 2016; Washio 1997), and, to a lesser extent, with other Germanic languages, such as German (Boas 2003; Oya 2002) and Dutch (Hoeksema and Napoli 2019; Hoekstra 1988; Lauwers et al. 2021).^[5]

Like in English, Dutch RCs are also fairly flexible: they can contain activity (e.g., schaatsen ‘skate’, cf. [9]) and change of state verbs (e.g., maaien ‘mow’, cf. [10]), adjectival (e.g., kapot ‘broken’, cf. [9]) and prepositional RSPs (e.g., tot dwerg ‘to dwarf’, cf. [11]), subcategorized (cf. [10]) and non-subcategorized objects (cf. [9]). In other words, Dutch allows both strong (i.e., aspect-shifting) and weak (i.e., non-aspect-shifting) RCs.

Although these different sub-types of Dutch RCs have been identified in previous literature, how frequent and productive they are compared to their equivalents in other Germanic languages remains largely unknow.^[6] And yet, findings drawn from more recent empirical studies suggest that Dutch RCs exhibit language-specific properties. For instance, as extensively shown in Gyselinck (2018), fake reflexive RCs associated with a non-literal intensifying meaning (e.g., Lisa danced herself to pieces last night) appear to attract an impressive variety of verbs and intensifiers in Dutch (2018: ix–xvii), which do not always have a counterpart in English, cf. (12).

Similarly, Métairy et al. (2020) have demonstrated that status nouns are always marked by the preposition tot ‘to’ in Dutch resultative nomination constructions, regardless of the verb used (cf. [13]). By contrast, in English, to is usually found with directional verbs (e.g., John was promoted/fast-tracked to manager).^[7]

In this article, I investigate whether there are further intra-Germanic differences in the behavior of RCs, limiting my research to English and Dutch.

Contrary to Germanic languages, Romance are said to lack fully productive RCs (Acedo-Matellán 2012; Aske 1989; Bigolin and Ausensi 2021; Mateu 2012; Mateu Fontanals 2000; Mateu and Rigau 2010; Snyder 2001). The productivity imbalance between Germanic and Romance RCs has been argued to follow from a fundamental typological dichotomy between satellite- versus verb-framed languages, which was originally introduced in Talmy’s (1985, 1991) pioneering work on the syntactic encoding of complex motion events. In satellite-framed languages (e.g., Germanic languages), the path of an event – that is, the goal (cf. [14]) and, by metaphorical extension, the result (cf. [15]) – is encoded by a separate phrase (i.e., a so-called ‘satellite’ [Talmy 1985: 102]), whereas the manner in which the action of an event is carried out is specified in the verbal root:

By contrast, in verb-framed languages (which are prominent in Romance), the path of an event is incorporated in the verbal root, whereas the manner of action is expressed by an adjunct.^[8]

However, scholars have become aware of the relevance of RCs in Romance as well: see, for instance, Napoli (1992), Riccio (2018), and Romagno (2020) for Italian; Riegel (1996, 2001), Legendre (1997), Muller (2000), Dagnac (2009), and Métairy (2020, 2022a, 2022b) for French; Bosque (1990), Demonte and Masullo (1999), and Rodríguez Arrizabalaga (2003, 2014, 2016, 2022) for Spanish; and Farkas (2009, 2011, 2015) and Baciu (2014) for Romanian. Romance RCs will be presented in Section 1.2.

1.2 Romance resultative constructions

This section aims to summarize the main properties that characterize Romance RCs. Given that there is an extensive body of literature that covers this topic in different Romance languages, providing an exhaustive account of RCs in each Romance language individually would be beyond the scope of this paper. See, however, Métairy (2022b: 56–67) for a more detailed overview of these constructions.

Examples of Romance RCs are given below:

In prior literature, Romance RCs are said to exhibit the following properties: first, they only allow transitive verbs (e.g., Spanish: batir las claras ‘beat the egg whites’, cf. [22]) and unaccusative verbs (e.g., French: se briser ‘break’, cf. [19]). Thus, RCs with unergative verbs and non-subcategorized objects (e.g., bark, cf. [5]) do not seem available in Romance (Legendre 1997: 83; Mateu Fontanals 2000: 73; Rodríguez Arrizabalaga 2016: 57).

Second, Romance RCs are said to mostly – if not exclusively – combine with telic verbs and are, therefore, regarded as weak RCs (Baciu 2014: 65; Bigolin and Ausensi 2021: 6; Mateu 2012: 258; Washio 1997: 26–30). In other words, Romance RCs either involve accomplishments (e.g., Romanian: prăji ‘fry’, cf. [23]) or achievements (e.g., French: se briser ‘break’, cf. [19]). Note that such RCs do not constitute exceptions to Talmy’s (1985, 1991) generalization since the path of the event (that is, the change of state) is already incorporated in the verb and is only further specified by the RSP. In fact, some authors even argue that RCs do not exist in Romance precisely because they would lack RCs that contain atelic verbs (hence, the strong type), which are regarded as true, genuine instances of RCs (Acedo-Matellán 2012; Aske 1989; Bigolin and Ausensi 2021; Mateu 2012; Mateu Fontanals 2000; Mateu and Rigau 2010; Snyder 2001). Nonetheless, in my opinion, one should rather carefully describe the different types of RCs available in a given language than restrict the concept of ‘RCs’ and offer a reductionist definition of these constructions which does not take into account the fact that, even though weak RCs indeed differ from strong RCs, they share common properties with the latter – in other words, they are still resultatives. And third, there seems to be a general consensus to the effect that Romance RCs mostly involve PPs (Farkas 2009: 61; Legendre 1997: 46; Napoli 1992: 84; Rodríguez Arrizabalaga 2016: 72).

That being said, the examples illustrated above call into question some of these generalizations: (i) activity verbs can also be found in Romance, for instance in French (e.g., battre à mort ‘beat to death’, cf. [20]) and in Spanish (e.g., batir a punto de nieve ‘beat to snow’, cf. [22]). Therefore, strong RCs do seem to be possible in these languages as well (Métairy 2022a: 259–261; Rodríguez Arrizabalaga 2014, 2022; Romagno 2020). (ii) APs in Romance RCs are possible (e.g., Italian: asciutti ‘dry’), but tend to be intensified, for instance, via the addition of a degree adverb (e.g., Spanish: bien muertas ‘very dead’, cf. [21]). In Romanian, bare NPs are also attested. Such NPs can refer either to a newly created entity (e.g., scrum ‘ash’, cf. [23]) or to a property when used metaphorically (e.g., lună ‘moon’ and oglindă ‘mirror’, cf. [24]).

3.1 Cleaning phase and research delineation

Several constructions have been excluded from the sample of RCs: (i) phrasal verb constructions, which involve a preposition or an adverb (Aarts 1989; Cappelle 2005; Declerck 1991; Iacobini 2009; Müller 2002: 253–390), cf. (25) and (26):

(ii) caused-motion constructions, which describe a change of location (Beavers et al. 2010; Boas 2003; Goldberg 1995; Iwata 2008), cf. (27)–(30):

And, (iii) adverbial resultative constructions with morphologically marked adverbs (Broccias 2004; Geuder 2000; Iwata 2020; Quirk et al. 1985: 560), such as, en. thinly, fr. finement ‘finely’, sp. finamente ‘finely’, cf. (31)–(33). In such constructions, the adverb indirectly describes a property of the entity created via the verbal process: for instance, in example (31), the event of slicing the apples leads to the creation of apple slices which have the property of being thin. These constructions have been excluded from the analysis on the basis of three criteria. The first one is a morphological criterion: (i) the suffixes -ly in English and -ment(e) in Romance are generally regarded as explicit markers of adverbial status (Hummel 2017a: 14). Thus, although they establish a predicative relationship with a nominal referent (which is here not explicitly expressed in the sentence), marked adverbs cannot be formally analyzed as predicates. The second and third criteria for exclusion relate to their distribution: (ii) unlike prototypical predicates, for instance APs, they can precede the main verb (e.g., Thinly/*thin slice the apples) and (iii) they can only combine with lexical verbs, which exclude causative verbs (e.g., Roll out/*make the dough thinly):

From a semantic point of view, constructions featuring adverbs are not much different from constructions with their adjectival counterparts, e.g., en. thin, fr. fines ‘thin’, sp. finas ‘thin’, cf. (34)–(36).

Indeed, on a par with marked adverbs, these APs are not predicated of the direct object’s referent, but of the entity that emerges from the verbal event. Consequently, in previous literature, this type of adjectival constructions has been referred to as ‘spurious’ (Washio 1997: 17), ‘pseudo’ (Demonte and Masullo 1999: 2493; Levinson 2010: 138), or even ‘adverbial’ RCs (Mateu Fontanals 2000: 89), taking the ‘non-object’ scope of the adjectival form as a symptom of adverbhood, assimilating it to a manner-of-process interpretation. In my view, this conclusion is a bit too extreme. That we cannot simply discard these ‘pseudo’ RCs from the domain of resultative predication is also advocated by Broccias (2013), who argues that this phenomenon is only a by-product of the semantics of some verbs used in the RC. When the verb describes an event that inherently results in the creation of an entity (or set of entities), which is the case of chop, grind, or slice, the RSP can only be predicated of this new entity.^[10] Therefore, following this author, we have considered adjectival constructions like those in (34)–(36) as instances of RCs and have included them in our dataset.

Note that the APs finas ‘thin’ in Spanish (cf. [36]) and fines ‘thin’ in French (cf. [36]) agree in gender and number with the postverbal NP. In Romance, agreement is generally observed with predicative APs (cf. [37]) but not with APs used for adverbial functions (cf. [38]).^[11] Agreement morphology has been taken as evidence for non-adverbial status.

Then, in order to be in line with the Romance system, (iv) constructions with uninflected APs (cf. [39]) and APs with no apparent agreement marking (cf. [40]) in Spanish have been removed from the final dataset.

Things are quite different in Dutch since there is no morphological difference between adjectives (41) and adverbs (42) in this language. In other words, contrary to what is observed in English, French, and Spanish, Dutch adverbs are never marked by a specific adverbial morpheme, such as English -ly, French -ment, and Spanish -mente. This typological contrast is known as the “flexible” versus “differentiated” languages dichotomy, cf. Hengeveld (1992: 15, 20–21, 62–72); Hengeveld et al. (2004) and Hengeveld and van Lier (2010), cited in Hummel (2017a: 18).

That is why, in cases of ‘pseudo’ RCs, it is impossible to decide whether the AP in Dutch (e.g., fijn ‘thin’, cf. [43] and grof ‘coarse’, cf. [44]) is used as a predicate or as an adverb. Therefore, such APs have been included in the analysis and annotated as bicategorical ‘adjective-adverb’ items.

Table 1 summarizes our analysis of adjectival and adverbial constructions in the four languages under study.

3.2 Corpus annotation

As a second step, 1,000 hits of the RC were retrieved manually in the four languages studied (that is, 4,000 in total) from the corpus of cooking recipes. RCs turn out to be more frequent in Germanic languages, and particularly in Dutch, than in Romance: as shown in Table 2, a higher number of tokens had to be read in French and Spanish in order to collect 1,000 hits of the construction than in Dutch or English. This is indicated by the normalized token frequency calculated for each language: per 1,000 tokens, 12.6 RCs were found in Dutch, 6.7 in English, 4.2 in French, and 2.4 in Spanish.

Each occurrence of an RC was annotated according to several variables. These variables concern the morphosyntactic category of the RSP (viz. PPs [e.g., into pieces], APs [e.g., dry]), the lexeme of the preposition (e.g., into, to, until, etc.) as well as the formal make-up of the complement within the PP – that is, whether it is encoded by an AP (e.g., until smooth) or an NP (e.g., until a golden caramel). Further important variables were considered during the annotation process, such as the verb lexeme (viz. cut, whisk, fry, etc.) and the semantic verb class. Each verb class will be presented in Section 4.

4.1 Verb classification

In total, 234 verbs were found with an RSP in our 4,000-token dataset, i.e., Dutch: 55, English: 101, French: 59, Spanish: 19 (see Tables 10–13 in the Appendix). These verbs have been divided into nine semantic classes, which are presented below:^[12]

1. Cut verbs (e.g., en. tear, nl. verdelen ‘divide’, fr. tailler ‘carve’, sp. cortar ‘cut’, cf. [45]–[48]) describe processes that lead to a change in the material integrity of some entity (Levin 1993: 245). They refer to a separation which can be done with a bladed instrument (e.g., knife, scissors) or using one’s hands and the actions result in the creation of discrete entities of a certain size and shape (e.g., slices, chunks, cubes). From an aspectual point of view, cut verbs denote achievements (i.e., punctual and telic events). However, when such events are iterated (as is usually the case in cooking recipes), this iteration yields an accomplishment reading. Consider, for instance, the event of slicing a cucumber: the slicing event ends when there is no more cucumber left (that is, when there are only slices of cucumber).

   (45)

   Tear the hearts into quarters .

   (46)

   Verdeel de broccoli in roosjes .

   ‘Divide the broccoli into florets.’

   (47)

   À l’aide d’une râpe, taillez -la en fines lamelles comme des spaghettis.

   ‘Using a grater, carve it into thin strips like spaghetti.’

   (48)

   Pelar y cortar las zanahorias a rodajas .

   ‘Peel and cut the carrots into slices.’

2. Grind verbs (e.g., en. crush, nl. malen ‘grind’, fr. mixer ‘grind with a blender’, sp. triturar ‘mash’, cf. [49]–[52]) are also verbs of disintegration. However, the processes denoted by these verbs are carried out with different types of instruments, such as a fork, a mortar, or an electric food processor, and lead to the creation of a mass entity (e.g., purée, coulis, powder). Grind verbs describe accomplishments, which are durative, and telic events: the event of grinding, crushing, mashing, etc. ends when the pieces contained in the processed substance have reached the desired size. Thus, grind verbs denote continuous processes which do not consist of a series of punctual subevents.

   (49)

   […] then crush the garlic and anchovy to a coarse paste .

   (50)

   Maal de gepelde amandelen fijn in de blender of een hakmolen.

   ‘Grind the peeled almonds until fine in a blender or chopper.’

   (51)

   Mixez en coulis la chair de la mangue avec un peu de sucre dans un blender.

   ‘Blend the flesh of the mango to a coulis with a little sugar in a blender.’

   (52)

   Cuando estén bien blandos, retirar y triturarlos en puré .

   ‘When they are soft, remove them and mash them to a purée.’

3. Beat verbs (e.g., en. press, nl. kloppen ‘beat’, fr. fouetter ‘whisk’, sp. batir ‘beat’, cf. [53]–[56]) refer to processes that involve a repeated contact with some entity. This repeated contact, which varies in terms of intensity (e.g., pat vs. pound) and the instrument used (e.g., roll vs. whisk), subsequently impacts this entity. Nonetheless, it does not necessarily lead to a determined change of state: for instance, one can whisk cream for a while without causing it to reach a specific consistency. In this sense, beat verbs describe activities (i.e., durative and atelic events).

   (53)

   Press flat with the back of a fork and carefully drop into the hot oil.

   (54)

   Klop de slagroom stijf met wat bloemsuiker.

   ‘Beat the cream until stiff with some icing sugar.’

   (55)

   Fouettez la crème en chantilly et ajoutez le sucre glace.

   ‘Whisk the cream to whipped cream and add the icing sugar.’

   (56)

   En otro bol, bate las claras a punto de nieve y después agrega las yemas.

   ‘In another bowl, beat the egg whites to snow and then add the yolks.’

4. Shape verbs (e.g., en. flatten, nl. vouwen ‘fold’, fr. façonner ‘shape’, sp. montar ‘raise’, cf. [57]–[60]) describe processes that alter the initial shape or state of an entity (this change is not obtained via material disintegration or cooking). This class includes verbs associated with distinct semantic properties: while some verbs provide information about the new shape of the entity in question (e.g., en. flatten the dough → make the dough into a flat object; sp. montar las claras ‘(lit.) raise the egg whites’ → get the egg whites foamy resulting in an increase of their volume), others simply entail that a change of shape is taking place (e.g., en. shape, mould, and form, fr. façonner ‘shape’, former ‘form’, and mouler ‘mould’). This second subset of verbs obligatorily requires an RSP (e.g., shape the dough *[into a ball] vs. flatten the dough [into a disk]). Shape verbs mostly denote accomplishments (viz. durative and telic events).

   (57)

   Use your fingers to flatten the dough into a disk .

   (58)

   Leg de vulling op 1 helft van het deeg en vouw dicht .

   ‘Place the filling on 1 half of the dough and fold tight.’

   (59)

   Façonnez chacune d’elles en boule .

   ‘Shape each one of them into a ball.’

   (60)

   Luego, montamos las claras a punto de nieve con el resto del azúcar.

   ‘Then, raise the egg whites to snow with the rest of the sugar.’

5. Cook verbs (e.g., en. toast, nl. bakken ‘fry’, fr. cuire ‘cook’, cf. [61]–[63]) refer to various methods of cooking ingredients. They also denote accomplishments: the cooking event ends when the ingredients are cooked.

   (61)

   Lightly toast the coconut until golden .

   (62)

   Verhit een pan en bak de spekblokjes krokant .

   ‘Heat a pan and fry the cubes of bacon until crispy.’

   (63)

   Faites cuire les pâtes al dente dans un grand volume d’eau salée.

   ‘Cook the pasta until al dente in plenty of salted water.’

6. Blend verbs (e.g., en. bring together, nl. roeren ‘stir’, fr. ficeler ‘tie’ [64]–[66]) denote processes whereby things are mixed or attached together. This semantic class includes both telic (result-oriented) verbs (e.g., en. bring together) and atelic (manner-oriented) verbs (e.g., nl. roeren ‘stir’).

   (64)

   Using your hands, bring the dough together into a ball .

   (65)

   Roer de geweekte mastellen tot een dikke broodbrij .

   ‘Stir the soaked mastels to a thick bread paste.’

   (66)

   Bouquet garni: choix de plantes aromatiques, ficelées en petit fagot […].

   ‘Bouquet garni: a selection of aromatic herbs, tied into a small bundle […].’

7. Wipe verbs (e.g., en. scrub, cf. [67], nl. borstelen ‘brush’, cf. [68]) describe how things can be removed from surfaces and objects. Equivalent verbs exist in Romance (e.g., fr. frotter ‘scrub’, brosser ‘brush’, sp. fregar ‘scrub’, cepillar ‘brush’) but none of them have been found with an RSP in the Romance corpus sample. One possible explanation for this observation is that wipe verbs rarely combine with PPs in cooking recipes. Recall that, in this particular genre, prepositional RCs usually entail the creation of an object (or set of objects). However, it is difficult to imagine a situation where an event of wiping, scrubbing, brushing, etc. leads to this kind of outcome. Thus, we could assume that PPs are not found with wipe verbs because the types of events they denote do not alter the material integrity or shape of the affected entity, which is necessary in the creation of a new object. Given that the RSP is preferably encoded by a PP in Romance, especially when it occurs with activity verbs, this can explain why wipe verbs have not been attested in these languages.

   (67)

   Scrub the potatoes clean and get rid of any gnarly bits.

   (68)

   Borstel de champignons schoon en snij ze in kwartjes.

   ‘Brush the mushrooms clean and cut them into quarters.’

8. Put verbs (e.g., en. arrange, cf. [69], fr. déposer ‘lay’, cf. [70]) describe how things are placed with respect to some location.

   (69)

   Slice pears and arrange fanned out on the pastry.

   (70)

   Déposez sur le dessus les lamelles de radis en écailles .

   ‘Lay the radish slices on top in a scale-like pattern.’

9. Finally, inherently resultative verbs (e.g., en. get, nl. brengen ‘bring’, fr. rendre ‘make’, sp. hacer ‘make’, cf. [71]–[74]) describe processes that solely consist in a change of state. The means through which this change of state is achieved is not specified. Semantically underspecified, these verbs cannot occur in a sentence without the RSP (e.g., make the dough *[into a ball] vs. roll the dough [into a ball]). In the literature, they have been referred to as ‘causative operators’ (Gross 1981: 23, 1998; Muller 2000: 22) or ‘support verbs’ (Demonte and Masullo 1999: 2507) in order to account for their semantic underspecification.

4.2 General overview of the RSP morphosyntactic categories

Table 3 provides a general overview of the morphosyntactic categories attested in the RSP slot in Dutch, English, French, and Spanish.^[13]

As shown in Table 3, the RSP can be encoded by (i) a PP (e.g., into slices, to a paste, until fine) or (ii) an AP (e.g., flat, dry, thin).^[14] PPs are more frequent in the languages of our sample (cf. nl: 53 %, en: 96.7 %, fr: 98 %, sp: 94.8 %). They are followed by APs (cf. nl: 47 %, en: 3.3 %, fr: 2 %, sp: 5.2 %). Interesting cross-linguistic contrasts can be observed: first, APs are much more frequent in Dutch than in English (cf. 47 vs. 3.3 %) – although both languages are satellite-framed. Second, even in Romance, APs are not distributed evenly: they are more frequent in Spanish than in French (cf. 5.2 vs. 2 %). Quite surprisingly, the frequency of APs is even higher in Spanish than in English (cf. 5.2 vs. 3.3 %), where this morphosyntactic category is expected to occur more freely. Nonetheless, as will be seen in Section 4.3, Romance APs are limited in terms of number and diversity of lexical types. Moreover, they turn out to be quite different from the ones found in Germanic languages since they tend to be intensified via the addition of degree modifiers.

In the following sections, each RSP morphosyntactic category is thoroughly looked into, starting with APs (Section 4.3) and then PPs (Section 4.4).

4.3 Adjectival phrases

Examples of adjectival RCs in the four languages under scrutiny (i.e., Dutch, English, French, and Spanish) are provided in (75)–(78):

As pointed out earlier, the frequency of APs is much higher in Dutch (47 %) than in English (3.3 %), French (2 %), and Spanish (5.2 %) (Table 3). Given that Dutch and English are both satellite-framed languages, one would expect to observe a comparable frequency of APs in these two languages. A question that naturally arises at this point is: why APs are so frequent in Dutch in particular? The first reason is that Dutch adverbs are not morphologically marked and are therefore formally identical to adjectives (Section 3.1). Some Dutch lexical items, for instance fijn ‘thin’ in snij de ui fijn ‘cut the onion thin’ and grof ‘coarse’ in hak de kappertjes grof ‘chop the capers coarse’, which were annotated as APs, may correspond to two distinct morphosyntactic categories in English, namely APs (e.g., thin) and ly-adverbs (e.g., thinly). The latter is, however, not included in the 1,000-token sample. Therefore, the fact that adjectives are used both for adjectival and adverbial functions in Dutch partly explains why APs are more frequent in this language. Nonetheless, even if we remove these ambiguous items, the frequency of APs is still much higher in Dutch (439−196 = 239) than in English.

Another parameter seems to play a role in the cross-linguistic distribution of APs: in English, most APs are headed by the preposition until. Indeed, out of 317 APs found in the English sample, 285 (90 %) are preceded by until and therefore regarded as PPs, as against 32 (10 %) occurring with no preposition. This is a very interesting finding considering that until-RCs have been overlooked in the literature (see, however, Beavers 2008; Rodríguez Arrizabalaga 2014, 2022). In fact, the lack of interest in until-RCs is quite a paradox since the preposition until is sometimes used to account for the meaning of RCs (i.e., water the tulips [until they are] flat [Carrier and Randall 1992]). This type of RC will be investigated in Section 4.4.3. By contrast, Dutch APs are almost never introduced by a preposition: only one AP has been attested with the preposition tot ‘to/until’, cf. (79). It could be argued that tot ‘to/until’ is added to the sentence precisely to mark the adjectival RSP gaar ‘cooked’, which occurs at a long distance from the main verb bakken ‘fry’, following several adjuncts. However, other examples with a similar structure are attested without a preposition, cf. (80).

If we look at the adjectival type frequency in Table 4, we can see that adjectival RCs are more productive in Germanic languages than in Romance. Dutch exhibits the highest type frequency with 23 types. It is followed by English with nine types, then by Spanish and French with only four types each. Thus, despite a relatively higher token frequency (52 tokens), Spanish allows a limited number of types, which also belong to the same semantic field (viz. fino ‘thin’, pequeño ‘small’, menudo ‘thin’). By contrast, English shows a lower token frequency (32 tokens) but allows a higher variety of types (viz. dry, clean, open, flat, small, smooth, etc.). This low diversity of types observed in Spanish is very likely due to the fact that the Spanish RC is almost restricted to one semantic verb class, viz. cut verbs. Although RCs that contain cut verbs are very frequent in the four languages studied, they happen to be even more frequent in Spanish (sp. 94 %, fr. 78.6 %, nl. 64.5 %, en. 38.3 %). The limited productivity of APs could then be correlated with the limited productivity of the verb slot in this language.

It should also be pointed out that Romance APs are very often associated with intensifying degree adverbs, and more precisely, so-called ‘boosters’ (Quirk et al. 1985) (viz. fr. très ‘very’, bien ‘well’; sp. muy ‘very’, bien ‘well’, cf. [81]–[83]).^[15] In Spanish, these adverbs can be associated with a diminutive (e.g., bien picadita ‘very chopped’ [83]), which has the same intensifying function (Armstrong 2012; Bosque 1990; Rodríguez Arrizabalaga 2016). Spanish diminutives consist of the adjectival root and the -ito/a suffix.

Modified APs are indeed far more frequent in Spanish (40/53, 75.5 %), than in Dutch (24/439, 5.5 %), and in English (2/33, 6 %). In French, the situation is slightly different: al dente ‘al dente’, which is the most frequent AP (17/21, 81 %), tends to resist adverbial modification, although it is not entirely excluded (e.g., cuire très al dente ‘cook very al dente’ [cf. Note 14]). This is probably due to the fact that this term refers to a specific level of cooking, which, in principle, cannot be construed as gradable. However, the other APs found in the French sample are also modified by a degree adverb. This observation is in line with previous findings on Romance RCs and is corroborated by French native speakers’ intuitions: according to my informants, sentences (84) and (85) are more acceptable when the AP is preceded by très ‘very’:

The question that arises at this point is: why can Romance APs only occur in RCs if they are intensified? One could argue that adding a degree adverb (or reduplicating the adjective, as observed in Italian, e.g., stirare la camicia piatta piatta ‘iron the shirt very flat’ [cf. Napoli 1992: 75]) is one way of foregrounding the resulting state denoted by the AP by making it syntactically heavier, and thus deviating attention from the primary predication (Armstrong 2012: 21; Napoli 1992: 75–76; Rodríguez Arrizabalaga 2016: 74). However, in my view, this might be more a consequence of the phenomenon rather than an explanation. Instead, I believe that Romance adjectival RSPs are more readily accepted when the state they denote is interpreted as exceeding an assumed norm or standard (Quirk et al. 1985: 589): thus, ‘the egg whites are very stiff’ entails that the consistency of the egg whites is even stiffer than what is usually regarded as stiff. In other words, when expressing a high degree of a given property, the resulting state becomes sufficiently marked to be overtly mentioned. Conversely, if the AP refers to a property that is perceived as ‘neutral’, for instance if the egg whites must be just stiff, perhaps one does not need to be that explicit and can make use of other strategies to encode this information, for instance with verb modifiers (e.g., bien battre les blancs ‘beat the egg whites well’).

This condition might also apply to Romanian nominal RSPs, such as lună ‘moon’ and oglindă ‘mirror’ (cf. [24], repeated in [86] for convenience), which refer metaphorically to the property of being shiny. This example is then interpreted as follows: the girl scrubbed the floor in such a way that it became as shiny as the moon or a mirror. When speaking of the floor, the adjective shiny entails a very high degree of cleanliness. Note that reaching this degree presumably requires more ‘scrubbing’ on the part of the agent. Then, freca lună/oglindă ‘(lit.) scrub moon/mirror’ in Romanian and scrub clean in English, while being semantically very similar, seem to serve distinct communicative functions: one is to intensify the action denoted by the verb, the other is to encode a complex change of state.

In addition, Romance APs are almost always found in weak RCs: most of them combine with verbs that describe events that inherently lead to a change of state. In French, these include (i) verbs of disintegration (e.g., couper ‘cut’, cf. [35]), (ii) verbs of cooking (e.g., cuire ‘cook’, cf. [63]), (iii) verbs of change of shape (e.g., étaler ‘spread out’, cf. [81]), and (iv) verbs of combining and attaching (e.g., ficeler ‘tie’, cf. [82]). In Spanish, APs are only attested with verbs of disintegration (e.g., picar ‘chop’, cf. [78], cortar ‘cut’, cf. [83]). By contrast, Germanic APs also occur activity verbs (e.g., nl. roeren ‘stir’, cf. [74] and borstelen ‘brush’, cf. [68]; en. press, cf. [53] and scrub, cf. [67]). See Table 14 in Appendix.

There is one example in French that could be considered as an instance of strong RCs, namely battre les blancs très fermes (cf. [77]). When used without the AP très fermes ‘very stiff’, the verb battre ‘beat’ can be interpreted either as an activity or as an accomplishment, hence its compatibility with pendant ‘for’ and en ‘in’ (cf. [87]). However, when the AP is added to the sentence, en ‘in’ is preferred over pendant ‘for’ (cf. [88]), which suggests that this element provides an explicit endpoint to the beating event. This aspectual ambiguity may follow from the fact that, in this particular expression (viz. battre les blancs ‘beat the egg whites’), the result of the event is conventionally expected and therefore entailed in the VP: when beating egg whites, especially with an electric whisk, one usually intends to obtain a foam with a rather stiff consistency. In other expressions where the result is not as strongly entrenched, battre ‘beat’ is only construed as denoting an activity (e.g., battre un tapis pendant/??une heure ‘beat a carpet for/??in an hour’).

To summarize, we have shown in this section that the RSP can also be encoded by an AP in Romance. However, Romance adjectival RSPs are characterized by a certain number of properties: first, they are not very frequent and limited to a few (semantically related) lexical items. Second, they are very often associated with intensifiers, suggesting that APs are acceptable in Romance RCs provided that they express a high degree of the property denoted by the adjectival head. And third, they almost always occur in weak RCs: that is, the AP merely specifies a result that is already incorporated in the verb semantics. We have also observed a very interesting contrast within the Germanic languages: adjectival RSPs turn out to be much more frequent in Dutch than in English, where APs are mostly headed by the preposition until (Section 4.4.3).

4.4 Prepositional phrases

This section deals with prepositional RSPs. Recall that PPs constitute the most frequent RSP morphosyntactic category in both Germanic and Romance languages (nl. 53 %, en. 96.7 %, fr. 98 %, sp. 94.8 %, Table 3).