Non-Maximal Definites in Romance

Mainly based on data from Old Spanish and Modern Francoprovençal, this paper discusses a hitherto underresearched use of the Romance definite article that cannot straightforwardly be explained by recurring to any of the standard analyses of semantic definiteness


Introduction
Drawing on data from Old Spanish (1) and Modern Francoprovençal (2), this paper discusses a hitherto underresearched use of the definite article that cannot straightforwardly be explained by recurring to any of the standard analyses of semantic definiteness, i.e., maximality (Russell 1905;Hawkins 1978;Heim 2011) and/or familiarity (Christophersen 1939;Heim 1982 As reflected by the symbol < ? > preceding the English renderings, both (1) and (2) display an apparent mismatch between morphology and semantics. In both cases, the morphological definiteness of the recipe ingredient-denoting nouns el léboro negro ' ? the black juniper' and l=evə tsˈadɐ ' ? the lukewarm water' does not straightforwardly correspond to a semantically definite interpretation of the corresponding nominal. Instead, both (1) and (2) appear in semantically indefinite contexts where neither maximality nor familiarity is fulfilled. The two nominals are not contained in any kind of recipe-preceding ingredient list either. Hence, an intuitively more appropriate English rendering seems to be one that uses bare nouns instead of definite articles, i.e., 'Take ø black juniper' and 'One must not clean the jugs with ø lukewarm water'.
Non-maximal definite articles of type (1) and (2) are also widely attested in Modern Italian (and Modern Corsican, see Chiorboli 1994). Modern Italian (3a), in which the definite article is ambiguous between a maximal and/or familiar and a semantically indefinite reading, is taken from Cardinaletti & Giusti (2018). The same authors show that in this language, semantic indefiniteness can also be expressed by so-called partitive articles (3b) and bare nouns (3c), though with diatopic and, in some cases, fine-grained semantic differences (cf., also Lebani &Giusti 2022 andPoletto 2022a,b). 1 In the remainder of this paper, we will not explicitly discuss Modern Italian data, but instead focus on Old Spanish (1) and Modern Francoprovençal (2). In the context of Modern Italian (3), however, note already (see Section 3.2) that in Old Spanish, the encoding of semantic indefiniteness, besides non-maximal definite articles (1), is also possible by means of bare nouns, while bare nouns are blocked in Modern Francoprovençal. In turn, this latter language, besides non-maximal definite articles (2), also possesses partitive articles (4a), as well as invariable DE (4b)which are not available in Old Spanish. (3) Modern Italian (Cardinaletti &  Contrasting with Old Spanish (1), Modern Francoprovençal (2), and Modern Italian (3a), many other Romance varieties display a smaller range of interpretational possibilities of the definite article. In Modern Spanish, Modern Portuguese, Modern Romanian, and Modern French, for instance, a definite articleboth in the singular and in the plural (but, for the sake of parsimony, illustrated here for the singular only) is generally associated with a straightforwardly maximal and/or familiar reading and cannot be interpreted indefinitely (5a-8a; but see [11]). For an indefinite reading to emerge, these languages necessarily require bare nouns (5b-7b) except for French, where bare nouns are ruled out. In this latter language, the indefinite reading requires a partitive article (8b). 1 All Romance languages discussed in this paper also feature indefinite articles for the encoding of semantic indefiniteness with singular count nouns, a fact we will not explicitly deal with in what follows. In view of the puzzle sketched above and based on the assumption that morphological elements such as non-maximal el and l in (1) and (2) are not void of semantic content, the first research question of the present contribution is as follows: RQ 1: What is the semantics of the definite article in Romance data of type (1) -(2)?
In a nutshell, our answer to this first RQ will be that both (1) and (2) feature definite articles that come with representative object interpretations, available in what has been referred to as kind-oriented mode of talk (Krifka et al. 1995). Such nominals belong to the class of weak referentials (Aguilar-Guevara, Le Bruyn & Zwarts 2014) and yield type readings by means of what may be considered 'deferred' kind reference: they denote inherently non-specific, semantically number neutral regular objects whose only co(n)textual relevance is their being typical instantiations of their corresponding kind. As such, these clearly non-maximal definites are nominals with representative object interpretations and are not kind-denoting themselves.
In a second step, we zoom in on the concrete use of nominals with representative object interpretations and ask the following, second research question: RQ 2: Are there syntactic and semantic/pragmatic contexts that (dis)favor definite nominals with representative object interpretations of type (1) and (2) in Romance?
Our data, stemming both from corpora (written modality) as well as fieldwork studies (questionnaires with translation tasks, spoken modality), show that contexts favoring an 'habitual' reading of the predicate (recipes, treatises, narratives about what people used to do in former times etc.) strongly favor the use of non-maximal definites, in the Francoprovençal data especially in the scope of non-perfective verb tenses in prepositional or presentational complements, sometimes direct objects, in the Old Spanish data often in the scope of non-assertive mood (imperative, due to the genre of recipes), when introducing important discourse referents, and when being primed by preceding non-maximal definites. We use habitual in what follows to refer to predicates that denote routine situations or actions for the speaker and/or the subject of the respective sentence or which occur regularly or usually, without however meaning the actual habitual aspect (cf. e.g. Comrie 1985: 27).
The organization of the remainder of the paper is as follows: firstly, Section 2 provides a state of the art on non-maximal definite articles in Romance and beyond. Secondly, Section 3 presents and analyzes the Old Spanish and Modern Francoprovençal data, answering both RQ 1 and RQ 2. Section 4 then discusses these findings from a wider Romance perspective. Section 5 concludes.

SOA: Non-maximal Definite Articles in Romance and Beyond
As foreshadowed in Section 1, there are two major approaches to semantic definiteness. The first one, in line with the Russellian tradition (Russell 1905;Hawkins 1978), is known as the uniqueness approach, a classical label that is, however, inferior to the more general and nowadays well-established label maximality approach as, from a set theoretical perspective, only the latter does full justice to pluralities (Sharvy 1980). Under such a view, a definite description is felicitous iff it refers to a co(n)textually maximal set satisfying the description of the lexical noun. The competing familiarity approach, in turn, which goes back to Christophersen (1939) and was formalized in terms of File Change Semantics by Heim (1982), argues that the underlying semantic core of definiteness is familiarity. Under such an alternative viewand simplifying considerablya definite is felicitous iff it refers to a referent for which a file card has been created prior to the moment of utterance. Definite descriptions are, therefore, taken to always refer to entities that have previously been introduced into the universe of discourse. Even though there has been (and continues to be) long-standing scholarly discussion in favor of and against both approaches, it can be shown that the familiarity approach has severe problems in explaining a number of Hawkins' (1978) nine usage types of definite articles that can, in most cases, elegantly be dealt with, however, by assuming maximality as the semantic primitive underlying definiteness (Lyons 1999, 2-15). 2 The remainder of this Section presents two types of uses of the definite article that can neither be explained by maximality nor familiarity. We globally refer to such definites, which have increasingly come to the attention of linguists since Poesio (1994;cf. also Birner & Ward 1994), as weakly referential definites. Following the terminology introduced by Espinal & Cyrino (2017a,b), we will refer to the first subtype of weakly referential definites discussed as short weak definites (SWDs; Section 2.1). The second subtypehardly taken into consideration by previous literature at allis that of definites with representative object interpretations (ROIdefinites; Section 2.2). 3 Section 2.3 resumes the main differences between SWDs and ROI-definites. In Section 3, these difference will serve to show that the Romance nonmaximal definites of type (1) and (2) are ROI-definites and not SWDs.
(ii) What's wrong with Bill? -The woman he went out with last night was nasty to him. (Hawkins 1978: 130, 148) We are fully aware that there are more approaches to semantic definiteness than the maximality and the familiarity approach. Roberts (2003), for instance, argues for a blending of uniqueness and (weak) familiarity; von Heusinger (2013) assumes semantic definiteness to be about picking the most salient referent satisfying the description. Finally, Löbner (1985Löbner ( , 2011 distinguishes between relational ("semantic") and anaphoric or deictic ("pragmatic") definiteness. 3 There are at least two further subtypes of weakly referential definites not addressed in this paper. The first onealso primarily discussed by Poesio (1994) are long weak definites (Espinal & Cyrino 2017a), which correspond to relational nouns followed by a PP-modifier of the type the corner of the intersection or the student of a linguist. The second one are expletive definites, which rely on inalienable possession, as in Spanish Pedro se lavó la cara 'Pedro washed his face', or on extended inalienable possession, as in Spanish Pedro perdió el móvil 'Pedro lost his cell phone' (Espinal & Cyrino 2017a: 2, their ex. [3]).
El tren 'the train' in (9) is semantically ambiguous. On the one hand, we get a reading under which the respective nominal is maximal at regular object level, that is, refers to a specific train exemplar that is the only (relevant) one in the co(n)text. On the other hand, (9) can also mean 'María travelled from Sevilla to Madrid by means of train riding'. The second reading is the SWD-reading we are interested in. This SWDreading comes with some fuzzy 'generic flavor' (Carlson et al. 2006;Aguilar-Guevara & Zwarts 2013;Carlson et al. 2013;Schwarz 2014: 215ff.) and el tren 'the train' is semantically number-neutral (Carlson et al. 2006): (9) is felicitous even if María on her journey boarded several trains at different train stations with numerous train exemplars involved. SWDs such as (9) have a restricted syntactic distribution: they are most typical as complements of V° and P° (although the marginal existence of subject-SWDs has been argued for; cf. Leonetti 2019: 5). In addition, SWDs license sloppy readings with VP-stripping (cf. Carlson et al. 2006;Aguilar-Guevara & Zwarts 2013 To date, there exists no consensual semantic analysis of SWDs and it is, in fact, not even clear if they can be analyzed compositionally. This is because SWDs display some properties of idioms (cf., e.g., Heim 2011Heim : 1009Klein et al. 2013: 188f.). The lowest common denominator seems to be that if SWDs are amenable to a compositional analysis, then they represent cases of complex predicate formation (but cf. Corblin 2013). Under an analysis of SWDs as 'quasi kind-denoting' that capitalizes on their 'generic flavor' (see above) such nominals would be morphologically definite because kinds, qua rigid designation, are inherently unique (cf., e.g., Aguilar-Guevara & Zwarts 2013 for details). The main competitor to this analysis of SWDs, which, as far as we see all rely on the notion of type reference (cf. Beyssade 2013; Leonetti 2019), is one in terms of semantic (pseudo-)incorporation along the lines of, for instance, van Geenhoven (1998), Stvan (2009), andDayal (2011), among many others. Such analyses of SWDs, differently defended in Carlson et al. (2006Carlson et al. ( , 2013, Schwarz (2014), and Espinal & Cyrino (2017b), cash in on the fact that SWDs behave like pseudo-incorporated bare nouns in most respects (Carlson & Sussman 2005: 73ff.). This casts some doubt on whether SWDs are at all referring expressions. In fact, some pseudo-incorporational approaches hold it that SWDs form part of complex constructions that denote familiar event kinds (cf. Carlson et al. 2006Carlson et al. , 2013Espinal & Cyrino 2017b) and that it is such event kind familiarity that explains the morphological definiteness of SWDs.
(11) Spanish (Leonetti 1999: 873) Aquella mañana tuve mi primer encuentro con that morning I.had my first encounter with el gorila the gorilla 'That morning I had my first encounter with the gorilla' Just like SWDs (Section 2.1), el gorila 'the gorilla' in (11) is semantically ambiguous. One interpretation is that the definite denotes a maximal, regular gorilla referent. There is, however, a second, marked reading, under which the definite does not meet maximality at regular object level. Under this reading, which is truthconditionally equivalent to an indefinite (Oosterhof 2006: 67), el gorila now seems to act as a sort of 'place holder' for the entire species GORILLA, paraphrasable in (11) as 'the animal type gorilla'. This is the ROI-reading, which, following Krifka et al. (1995), introduces non-specific regular objects that only serve as prototypical instantiations of their corresponding kind. Just like SWDs, ROIs are often complements of V° and P°. They are mostly restricted to morphologically definite singular count nominals but are, according to Oosterhof (2006: 67f.) and at least for some speakers, also possible with Dutch bare plurals such as we stonden oog in oog met blauwe vinvissen 'we came face to face with blue whales'. Again similarly to SWDs, ROIs come with some sort of 'generic flavor' and it is at the kind level only that such definites fulfill maximality. Nevertheless, a genuine kind-denoting analysis is not an option: firstly, ROI-selecting predicates, such as, e.g., 'had my first encounter with X' in (11), are stage-level in the sense of Carlson (1977), that is, they only apply to individuals at a given point of time (roughly: 'tensed slices, i.e., stages of individuals') and not to individuals on a whole (among which are kinds, which are stable over time; cf. also Gerards 2020, Chapter 3.2.1.2). Secondly, ROI-definites refer to regular, non-specific objects modelable in join semilattices (cf. Krifka et al. 1995: 83ff.;von Koss Torkildsen 2002: 82f.; Mueller-Reichau 2013: 93): ROI-definites such as el gorila 'the gorilla' in (11)which are semantically number neutral (see [12] below)can refer to any of the atoms and sums of atoms in Figure 1. This is not the case of kinds as, ontologically, these latter are indivisible wholes (see also fn. 10). 4

Figure 1. Join semilattices and regular object reference
Source: Champollion (2017: 16) 4 That kinds are not just maximal sums over all possible worlds is shown by the fact that sentences with kind-denoting nominals, such as, e.g., Swans are white or The polar bear is dangerous, remain true even if we see a black swan or a polar bear that happens to be completely harmless, for instance, due to its being injured and in captivity.
That ROI-definites are not just regular non-specific definites is evinced by two facts. Firstly, and again on a par with SWDs, ROI-definites license sloppy readings with VP-stripping (Pires de Oliveira 2013: 28), whereas regular non-specific strong definites do not (cf. Gerards 2020, Ch. 3.2.4.1). Secondly, ROI-definites, again just like SWDs, are semantically number neutral. This is shown by the fact that (12) is a perfectly felicitous continuation of (11) even without the interpretation that the antecedent el gorila in (11) only refers to a subset of una manada de veinte espaldas plateadas 'a band of twenty silverbacks'.  Krifka et al. (1995) propose that ROI-definites involve a switch from the default modality of semantic evaluation -which they call "object-oriented mode of talk"to the marked kind-oriented mode of talk. The latter is said to allow the evocation of kinds via regular objects acting as prototypical representatives of this kind, that is, to yield what may be considered a sort of 'deferred kind denotation': "the object in the situation described is only relevant as a representative of the whole kind [and] a property can be projected from the object to the kind" (Krifka et al. 1995: 83; cf. similarly Mueller-Reichau 2013: 93). Formally speaking, Krifka et al. (1995: 85f.) put forward the idea that ROI-nominals involve an IS-relation IS(x,y) ⟺ df (x=y ∨ R(x,y)), by means of which regular objects can representatively realize (R) their corresponding kind. It is, however, unclear what triggers the activation of the kindoriented mode of talk and, hence, that of ROIs. Krifka et al. (1995: 83), stating that "[l]anguages, and maybe registers within a language, seem to differ in their readiness to employ [the kind-oriented mode of talk]", favor a pragmatic explanation: [I]t seems futile to look for hard grammatical criteria that determine when we can use [the kind-oriented mode of talk]. Instead, the criteria we expect to find will have a more pragmatic flavor. For example, when we do not care about the object-level identity of the objects, as in the sentence We filmed the grizzly in Alaska, we can (and often will) choose to talk in the kind-oriented mode. (Krifka et al. 1995: 87)

Four Differences between SWDs and ROIs
There are at least four differences that justify to treat SWDs and ROIs as two separate types of weakly referential definites (cf. also Gerards & Stark 2020; Gerards 2020, Ch.

3.2.4.3). 5;6 5
This view is potentially against Leonetti (2019), who proposes a unified analysis of several types of weakly referential definites. Note, however, that Leonetti does not explicitly consider ROI-definites. 6 Pires de Oliveira (2013: 28) mentions a fifth difference between SWDs and ROIs, namely that only the former but not the latter come with semantic meaning enrichment. For The first difference between SWDs and ROIs is that the latter have been assumed to be a pragmatic, register-conditioned phenomenon (Section 2.2), whereas SWDs clearly are not. SWDs can occur out of the blue in any communicative register or situation (Zwarts 2014: 267).
The second difference lies in the different ability of SWDs and ROIs to introduce discourse referents. It is well known that SWDspossibly linked to their being pseudo-incorporations (Section 2.1)are considerably worse in this regard (Carlson et al. 2006;Aguilar-Guevara & Zwarts 2013;but cf. Brocher et al. 2020, who show that SWDs are not completely unable to do so). ROIs, in turn, do not display such defectiveness. This is shown by the fact that (13b) is a perfect continuation of (13a; [see also Stark & Gerards 2022]): (13) a.
When I was little, I often took pictures of itj in the savanna.
The third difference between SWDs and ROIs is that SWDs are only licensed with a very reduced number of lexical items with even (near) synonyms often not yielding an SWD-reading. Such lexical restrictions can be of nominal, verbal, or prepositional nature (for examples, see Carlson & Sussman 2005: 76;Carlson et al. 2006;Aguilar-Guevara & Zwarts 2013: 34;Klein et al. 2013: 189). ROIs, in turn, are, so it seems to us, in principle not lexically restricted, provided that the descriptive content of the noun can pragmatically be construed as possessing a corresponding, well-established kind. By way of example, imagine that glass bottles do not exist on Mars but are a precious, frequently discussed object on this planet. Now, a Martian visits Planet Earth, where he sees a shelf of ordinary, yet to him precious and special glass bottles in a store. Given this scenario, the Martian, upon his return home, canat least to our judgmentfelicitously utter (14) (14) can also be constructed for the verbal and the prepositional domain. Finally, the fourth difference between SWDs and ROIs is that SWDs are morphologically defective, being either morphologically singular or morphologically plural (Espinal & Cyrino 2017b: 130, as well as references therein). As noted in Section 2.2 ROIs, too, are mostly morphologically singular. Yet, reference was made by Oosterhof (2006) to Dutch, a language in which bare plurals may be amenable to ROIs, as well.
the reasons set out in Gerards (2020, Ch. 3.2.4.3), we do not consider this difference in the present paper.

ROI-definites in Old Spanish and Modern Francoprovençal
This Section answers the two RQs formulated in Section 1. Section 3.1 focuses on Old Spanish, Section 3.2 on Modern Francoprovençal.

Old Spanish
The data discussed in this Section come from a set of 1,439 nominals with ingredientdenoting nouns manually extracted from a total of 20 Old Spanish technical recipe treatises from the 13 th -16 th century and are a subset of the data analyzed in . 7 The text type of technical recipe treatises as well as the time range were chosen because a previous corpus study based on another 2,911 nominals from the large-scale corpus CORDE, 8 on which more information can be found in Gerards (2020, Ch. 4.2.1), had revealed that in Spanish it is this textual genre of this particular period of time that displays a large number of non-maximal definites of type (1-2), especially in 'habitual' contexts in the scope of non-assertive mood (imperative/subjunctive). All of the 1,439 nominals analyzed in the present paper stem from semantically indefinite contexts with stage-level predicates, i.e., do not fulfil maximality at regular object level and cannot be genuinely kind-denoting (cf. Section 2.1). In particular, none of the ingredient-denoting nouns analysed were contained in any kind of recipe ingredient list (in any case only very sporadically) included at the beginning of medieval recipe treatises. What is important for the purpose of this paper is that non-maximal definites (15)  In a first step, we will use the diagnostics elaborated in Section 2.3 to show that in cases of (15) [and (1)] we are not dealing with SWDs (Section 2.1) but could be dealing with ROI-definites (2.2). In a second step, we provide further positive evidence for this ROI-analysis. Finally, we present a binomial mixed-effects logistic regression model to determine if there are syntactic, semantic, and pragmatic contexts (dis)favoring Old Spanish ROI-definites as compared to bare nominals (16). 7 In addition to the types of nominals discussed in this Section (bare nominals and nominals headed by non-maximal definite articles), this data set also contains partitives and nominals headed by quantifiers such as 'some', 'a bit', etc., which are not relevant to the present discussion. 8 CORDE (http://corpus.rae.es/cordenet.html) contains data from the beginning of documentation until 1974 (236,709,914 tokens; 34,155 texts) from Spain, Latin America, and the Philippines.
The first criterion established in Section 2.3 to distinguish between SWDs and ROI-definites is that the latter but not the former are a pragmatic, register-conditioned phenomenon. This criterion cannot be assessed in the present paper due to the exclusive consideration of data from technical recipe treatises (where non-maximal definites are, however, highly frequent; see also Section 4 for a hypothesis in this regard).
The second criterion differentiating SWDs and ROI-definites presented in Section 2.3 is that the latter are unproblematic as antecedents of anaphoric expressions while the former are clearly defective in this regard. In this regard, the Old Spanish data behave like ROI-definites: out of the 423 attestations of non-maximal definites in the data set, the majority of 269 (= 63.6 %) are picked up anaphorically. The third property in which SWDs and ROIs behave differently is that SWDs are only licensed with a very reduced number of lexical items, while this is not true of the latter. Again, our Old Spanish data clearly align with ROIs: firstly, the 423 nonmaximal definites in the data set are distributed over 169 nominal lexemes. Secondly, the 286 non-maximal definites that are not governed by a preposition but are an argument of V° are distributed over 30 different verbal lexemes. The 137 attestations of non-maximal definites governed by P° are so by con 'with' (n = 96), de 'of' (n = 10), en 'in' (n = 29), and sobre 'over' (n = 2).
Finally, the morphologically singular example (18), forming a minimal pair with morphologically plural (17), shows that our Old Spanish non-maximal definites can occur both in the singular and in the plural, again aligning them with ROI-definites but not with SWDs, which are morphologically defective. In what follows, we will adduce four further pieces of positive evidence that Old Spanish non-maximal definites of the type under discussion are ROI-definites in the sense of Section 2.2.
Firstly, note that Old Spanish non-maximal definites are compatible with modifiers that are unequivocally kind-level. This is shown in (19) containing the modifying relative clause quel dizen eya 'that is called eya'. The compatibility of such modifiers is expected under an ROI-analysis, given that ROI-definites yield type readings by means of what may be considered 'deferred' kind reference (Section 2.2). 9 9 In this spirit, we will from now on represent such type readings in the English glosses by <(type of)>. Secondly, Old Spanish non-maximal definitesas expected under an ROIanalysisintroduce regular objects that are non-specific and which only serve as prototypical instantiations of their corresponding kind (Section 2.2). Non-specificity is evinced by (20), which contains a modifying relative clause in subjunctive mood, an unequivocal trigger of non-specificity of the antecedent of the relative clause in Spanish (Leonetti 1999 boil.PTCP la camomjlla the chamomile 'And he shall often wash himself with (the type of) vinegar in which has been cooked (the substance type) chamomile.' Thirdly, Old Spanish non-maximal definitesagain as expected under an ROIanalysis (Section 2.2)are semantically number neutral. This sets them apart not from SWDs (Section 2.1), but from regular non-specific definites. As proof of this, witness (21), commented on below. The crucial point about (21) is the verb ovieren 'they will have', which is 3 rd person plural. It's [pro]-subject, in turn, is coreferential with the singular non-maximal definite el carnero 'the (type) wether', embedded into de la carne del carnero 'of the (type) wether meat'.
Fourthly and finally, Old Spanish non-maximal definites discussed cannot be accounted for as camouflaged strong definites licensed bywhatever kind offramebased semantics under which the substance would implicitly be given due to the fact that we are dealing with recipes. If it were, we would not be able to explain frequent intratextual minimal pairs of type (22) and (23). These data are from two different recipes in one collection of recipes and contain one and the same lexical noun olio rosado 'rose-infused oil' with one and the same syntactic function (complement of the P° con 'with'). Yet, (22) is headed by a non-maximal definite article while (23) [priming]. After checking whether all levels of the independent variables are attested, as well as for complete separation (Levshina 2015: 273), the original data set of 1,439 nominals had to be reduced to 1,422 nominals. We then performed backward stepwise selection by means of the function drop1 from the lme4 package (Bates et al. 2015) in order to determine which fixed-effects independent variables are worth keeping in the model (Levshina 2015: 266f.). This showed that only [syntactic function], [+/-anaphoric uptake], and [priming] are so. Finally, the data were submitted to a binomial mixed-effects logistic regression analysis with R by means of the glmer-function from lme4. Table 1 shows the results of this inferential statistical analysis and, except for the intercept, only includes statistically significant effects. 11

11
The results in Table 2 are rounded to three decimal places. Abbreviations: RL = reference level; significance codes: '***' = p < 0.001, '**' = p < 0.01, '*' = p < 0.05. Multicollinearity of the model was assessed by means of the variance inflection factor vif. The vif-score, calculable in R by means of the function vif from the package car (Fox & Weisberg As represented in Table 1, Old Spanish non-maximal ROI-definites are favored if the nominal is taken up anaphorically (1.630 times more likely than bare nouns) and by priming by other non-maximal ROI-definites preceding them (2.195 times more likely than bare nouns). They are disfavored, in turn, as complements of P° (only 0.408 times as likely as bare nouns) and by priming by morphological indefinites preceding them (only 0.179 times as likely as bare nouns). The implications of these results will be taken up in Section 4. Before, however, we will turn to Modern Francoprovençal non-maximal definites.

Modern Francoprovençal non-maximal definite articles
In this Section, we present fieldwork data from Francoprovençal that we consider ROIuses of the definite article, just as those presented in the preceding Section for Old Spanish. The origin and number of these data are very different from that of the Old Spanish ones, so that our argumentation cannot be fully parallel to the one in Section 3.1. We nevertheless consider it useful to highlight potential parallels in order to raise awareness for ROI-definites, hitherto neglected in the literature on nominal determination. Francoprovençal is a non-standardized and highly endangered Gallo-Romance language spoken today by about 110'000 to 160'000 speakers in France, Switzerland and Italy (Zulato, Kasstan & Nagy 2018: 13). It was never the official language of an administrative entity and never saw standardization. Nowadays, all speakers of Francoprovençal are bilingual and, for Swiss varieties, mostly older than 70 years, with the notable exception of Evolène in the canton of Valais in Switzerland, where the local variety of Francoprovençal is still transmitted to the next generation 2019), should be as low as possible but, in any case, not above 10 or 5, according to more conservative sources (cf. Levshina 2015: 160). All vif-scores ranged between 1 and 1.2.
(cf. Kristol 2016: 351). We also find competent speakers in the Aosta Valley, in Italy, due to active language policy. In what follows, we will discuss data from two different regions, mainly based on questionnaires (translation task), some on guided interviews. Firstly, a fieldwork trip in May 2017 to the Aosta valley under the direction of the much missed Federica Diémoz (2019 †) resulted in the elicitation of 627 nominals (not all included in what follows) produced by 17 native speakers from the localities Saint Nicolas, Saint Vincent, Fénis, Pontey, plus different native speakers at BREL (Bureau Régional pour l'ethnologie et la linguistique; for further description of the fieldwork trip and all results, see . Second, some data stem from the ALAVAL, an audiovisual database (Atlas linguistique audiovisuel des dialectes francoprovençaux du Valais romand; http://alaval.unine.ch), and were collected between 1994 and 2001, elicited by means of a questionnaire (translations and open questions for free speech) and transcribed in IPA. They come from 21 communalities in Valais (Switzerland; 42 speakers), 2 communalities in eastern Haute-Savoie (France; 4 speakers), and 2 communalities in the Aosta Valley (Italy; 4 speakers). For the present study, we analyzed 28 input sentences containing noun phrases with a partitive article from the 2017 data (only Aosta Valley; 7 fem.sg., 6 masc.sg., 8 fem.pl., 7 masc.pl.) and 46 input sentences from the ALAVAL data containing noun phrases with a partitive article (7 fem.sg., 9 masc.sg., 17 fem.pl., 13 masc.pl). In the vast majority of them (69 input sentences, see Tables 3, 4 and 5), non-maximal definite articles occurred as variants of the fully-fledged partitive article (PA) and of invariable de (DE) in the input.
The following Table 2 gives a general overview of the absolute occurrences of definite articles as equivalents of partitive articles in our data (both fieldwork form 2017 and ALAVAL): Definite articles seem to be the second most frequent option for Francoprovençal speakers to clearly convey an indefinite meaning. They occur most in three main contexts: in presentational complements (24), in prepositional complements (25) and with direct objects (26), all three positions under the scope of imperfective tenses and in utterances with a so-called 'habitual flavor'. week 'In spring, we picked ø mint for a week.' All in all, 11 input sentences contain PAs in presentational complements after a verb in the present tense or simple past with imperfective aspect. The following Table  3 illustrates the options speakers chose for their Francoprovençal translations, categorized for gender and number of the respective lexical noun: Table 3. Equivalents for PA nominals as presentational complements (11 input sentences) Again, definite articles are the second most frequent option for the speakers after the unmarked option DE (for most Francoprovençal varieties), with an outlier in the feminine singular, where the definite article outnumbers DE by far. Please recall that examples like (24) or (25) above clearly have an 'habitual flavor', and that example (25) additionally evokes a culinary context ('recipe-like').
In prepositional complements, again in the scope of verbs in the present tense or imperfective past (imparfait) and in utterances with an habitual flavor (see example [25] above), we find even more definite articles as translations of PAs. This is shown in the following Table 4. Strikingly, in the feminine singular (55 %) and the feminine plural (62.8 %), the majority of speakers chose the definite article, which is also the most frequent option for the masculine singular.
Finally, as represented in Table 5, the definite article also occurs as a translation for PAs in direct objects, again mostly, though not exclusively, in imperfective, habitual contexts: Table 5. Equivalents for PA nominals in direct object position (48 input sentences) One input sentence differs from the other contexts in which we find the definite article in our data in that the passé composé is considered a perfective context, see the following example. In how far gender (the feminine singular triggers, again, most definite articles, with 46.7%) is indirectly favoring the choice of the nominal determiner in this example cannot be determined.

Discussion
This Section discusses the findings of Section 3 in light of the general question about the correct semantic characterization of the non-maximal definites in our data and in Romance in general.
Section 3.1 showed that the non-maximal definites in the Old Spanish corpus data share six properties indicating their status as ROI-definites, i.e., as morphologically definite nominals whose only co(n)textual relevance is their being typical instantiations of their corresponding kind. These properties were: their availability as antecedents of anaphoric expressions, the lack of lexical restrictions, the lack of morphological number restrictions, their compatibility with kind-level modifiers, their non-specificity and semantic number neutrality, as well as the lack of strong readings because of contextual equivalence with bare nominals/indefinites. The respective nominals were evinced to be favored over bare nominals when introducing important discourse referents to be taken up anaphorically as well as by priming by other, preceding non-maximal definites. Disfavoring factors, in turn, were shown to be the syntactic function of complement of P° and priming by an indefinite, preceding nominal. Section 3.2 showed for Francoprovençal fieldwork data (mainly translation task) that non-maximal definites are quite frequent with presentational complements (which reminds of Italian, see Leonetti 2019), but also with prepositional complements (this is a difference to the Old Spanish data), all preferring contexts with an 'habitual flavor' (e.g. non perfective tenses such as imparfait or present tense). Feminine (singular) seems to increase the relative number of non-maximal definite articles (instead of the unmarked invariable DE), a morphological observation which we will not discuss any further in this paper (but see Strebel 2022 for a similar effect in nonstandard uses of the French partitive article).
The Old Spanish findings, which also find a perfect parallel in Old Portuguese (cf. Gerards 2020: Ch. 6), are highly interesting for several reasons. Firstly, they are yet another piece of evidence for the fact that, speaking in probabilistic terms (and always, of course, within the boundaries of grammatical systems), structural priming plays an important role in predicting the choice of linguistic variants, in the sense that "speakers [tend] to reuse recently experienced structures" as a cognitive mechanism of adaptation in language production (Dell & Jacobs 2016: 212; see also Branigan & Pickering 2017, Gries & Kootstra 2017, among many others). Recall from Section 3.1 that rigorous assessment of the question whether Old Spanish non-maximal definites are favored by or even exclusive to the text genre of technical recipe prose could not be provided in the present paper. Yet, a positive answer in this regard does seem likely to us. First of all, Gerards & Kabatek (2018) demonstrate the general possibility of genre biases of (certain readings of) grammatical elements in Old Romance. Secondly, Old Spanish non-maximal uses of the definite article have hitherto not even been noted by specialist scholars (cf., e.g., Ortiz Ciscomani 2009), which would be highly surprising if no genre bias were at stake with them. If this hypothesis of ours is confirmed by future research, then such a genre bias/restriction would not only be one more argument in favor of our ROI-analysis (cf. Section 3.1) but also potentially strengthen the previously formulated view that "less common, and hence more surprising, prime structures lead to more priming than common ones" (Dell & Jacobs 2016: 213, citing Jaeger et al. 2013. The second important finding for Old Spanish, namely that non-maximal ROIdefinites are significantly favored over bare nominals when introducing important, i.e., anaphorically taken up discourse referents, is highly relevant both from a synchronic and a diachronic perspective. Synchronically speaking, it further underscores that the choice between the two truth-conditionally equivalent variants (cf. Section 2.2) of nonmaximal definites vs. bare nominals is not entirely free and, in a sense, follows the well-known principle of iconicity (cf., e.g., Velupillai 2012: 35): antecedents of anaphoric expressions need to be accessed in the working memory of speakers at the moment of uttering the anaphoric expression, while this is not the case of referents mentioned only once upon their introduction. Arguably, thus, the overall processing load associated to referents taken up anaphorically in a discourse is higher than that of referents mentioned only once; the former can be considered 'more costly' from a cognitive point of view. In the case of the Old Spanish data, this enhanced cognitive load associated to a referent is, so we would like to suggest, prospectively signaled upon its introduction, namely by a morphosyntactically more complex encoding option that mirrors this enhanced load. This morphosyntactically more complex encoding option is the non-maximal definite article. Textually peripheric referents not taken up in the further discourse, in turn, imply less cognitive effort, whichfollowing the idea of iconicity sketchedwould be why they are significantly more often introduced by way of the simplest systemically possible encoding option, i.e., bare nouns. Such a 'pragmatic exploitation' of the Old Spanish definite article can also be accounted for nicely from a wider Romance as well as from a diachronic point of view. First of all, Epstein (2001: 79, his ex. [4]) convincingly and independently of our Old Spanish data shows that pragmatic exploitation of the definite article in terms of weakly referential definites introducing important new discourse referents and possibly serving as "attention getters" (Carlier & De Mulder 2010: 265) was also common in Old French. Secondly, it is well known that already during the first phases of its grammaticalization into a definite article in Late Latin, ille could not only signal topic continuity via anaphoric chains (cf. Carlier & De Mulder 2010: 259-263) but instead also introduce new referents that are to only become important and, hence, topical and/or prominent in the further discourse (Epstein 1993;Carlier & De Mulder 2010: 248f., 251-255;Danckaert et al. 2021: 18f.). Especially in this latter use, identified by Selig (1992: 165) as the primordial functional domain in Late Latin, ille, qua attention-getter (cf. above), thus constituted an important pragmatic means to create a textual relief (German: Textrelief). For the argumentation of the present paper it is furthermore crucial that such new, first-mention uses of Late Latin ille couldlike our Old Spanish non-maximal ROI-definitesrefer non-specifically (Selig 1992: 165ff.). In the following paragraph, we will, however, add some qualifying remarks with regard to the notion of newness of the important discourse referents introduced by Old Spanish non-maximal ROI-definites.
Recall from above (this Section) that there is sound reason to believe that Old Spanish non-maximal ROI-definites are favored by or even exclusive to the text genre of technical recipe prose, that is, of highly specialized text genres. Now, a reasonable way to make sense of thisdeveloped in more detail in Gerards (2020: Ch. 6.2.3)would be to argue that such non-maximal definites are typical of 'expert talk' because it is precisely in highly specialized text genres (and only in these) that they can be used felicitously: Both the author and the expert reader can pragmatically be presupposed to have previous knowledge of the existence of certain types of ingredients or species (NB.: not, however, of particularized, maximal sets of portions or sets of sets of these ingredients; there are no maximal referents in this regard in the co(n)text!). Obviously, this utterance-previous knowledge of existence at the kind level or, if one wishes, familiarity, to some extent relativizes the newness condition at the portion/set level discussed above. Nevertheless, it is still consistent with other, independent findings on the grammaticalization of the Romance definite article. As argued for by Carlier & De Mulder (2010: 241, our highlighting cf. also p. 249), in Late Latin the use of "ille signals that the referent cannot wholly be identified by elements in the context of utterance [and] thus suggests that previous knowledge should be activated in order to retrieve the referent". Now, this previous knowledge (i.e., familiarity), we would like to suggest, is that of the general existence of certain types of ingredients or species at the kind level that can felicitously be presupposed only for expert authors and readers of technical recipe prose. By virtue of (a severely reduced form of) the original deictic potential of its distal demonstrative source (for more details, see Gerards 20202: Ch. 6.2.3.2), Old Spanish non-maximal definite articles with ROI-readings point to shared knowledge at the kind level. 12 Turning now to the Francoprovençal findings, note, first of all, that these are interesting in two straightforward areal respects: Firstly, they contradict former observations of this area of Italoromance. Cardinaletti & Giusti (2018), in their synthesis of 100 year-old AIS data 13 of the Aosta Valley, show that no non-maximal ROI-definites were attested there. Taking into account this (and other) findings, they postulate a correlation, according to which regions with partitive articles (cf. [4a]) or invariable obligatory DE (cf. [4b]) only have one of these two options to express indefinite objects. This is mostly true of Modern Francoprovençal in the Aosta Valley (but see Stark & Davatz 2021), but our Modern Francoprovençal non-maximal ROIdefinites (cf. [2]) come into play and change the overall picture: if the older AIS data are correct, then in our fieldwork data from the beginning of the 21 st century we observe an important linguistic change of the system, namely that non-maximal ROIdefinites co-exist as (new?) variants of either partitive articles or invariable DE. Secondly, new fieldwork data for Northern Italian dialects adduced by Pinzin & Poletto (2022b) display non-maximal definite articles as 'background noise' all across their data points, i.e., feature them as options in answers to translation tasks from Emilia Romagna, Friuli and Liguria. Quite interesting for us, in this context, is the nature of their input sentences in the questionnaires used: very frequently, the sentences evoke 'cooking contexts', i.e., match the text genre resorted to for investigating Old Spanish non-maximal definites in Section 3.1.
(28) Northern Italian (Pinzin &  Such uses of early definite articles are, in fact, reminiscent of the recognitional use of distal demonstratives (e.g., that person… you know, Ms. Miller), by means of which "the referent is not mentioned in the preceding context or present in the current discourse situation [at the portion/set level and with which, instead t]he demonstrative points to 'specific, presumably shared' knowledge" (Himmelmann 2001: 833;cf. also Carlier & De Mulder 2010: 263f.). As observed by Epstein (2001: 186), such uses of distal demonstratives, which come with a strong component of speaker-hearer "intersubjectivity" (Carlier & De Mulder 2010: 266), are not necessary for the identification of the referent and are, instead, "expressive". This is why such recognitional uses of distal demonstratives could constitute a major catalyst for the grammaticalization of demonstratives into definite articles (cf. also Gerards 2020: Ch. 6.2.3.2). 13 Cf. Jaberg, Karl & Jakob Jud. 1928-1940  As for their semantics, the non-maximal Italoromance definites discussed in Cardinaletti & Giusti (2018) and Pinzin & Poletto (2022b) pattern with bare nominals rather than with the Italian partitive article in that they only appear in atelic contexts, have a non-specific interpretation and appear in contexts with an 'habitual flavor', all of which is, again, true of or at least strongly suggested for Old Spanish (cf. Section 3.1). At this pointand even though the data we discuss here are extremely different in nature (corpus data vs. fieldwork data, spanning over 8 centuries and several Romance languages and varieties)we would like to take stock and underline the following recurrent features of our non-maximal ROI-definites: -They are all clearly indefinite; strong readings are excluded (by experimental setting or contextual equivalence to indefinites). -They are non-specific. -They prefer 'cooking contexts' or in general non-perfective contexts.
-They possess anaphoric potential and may even be favored if taken up anaphorically (only Old Spanish attested). -They have no morphological number restrictions (singular and plural attested).
-They have no lexical restrictions, as far as we can see.
These properties quite clearly indicate the wide availability of ROI-definites (next to classical short weak definites like take the train or else) in manyyet not all -Romance languages and varieties (and maybe beyond Romance; cf. Section 5). Some Romance languagese.g., Modern Francoprovençal (cf. [2a]), Modern Italian (cf.
[3a]), and Old Spanish (cf. [1])make wide use of non-maximal ROI-definites, also with mass and plural nouns, i.e., under the conditions mostly discussed in this paper. In other Romance languages, in turn (e.g., Modern Spanish [5a], Modern Portuguese [6a], Modern Romanian [7a], and Modern French [8a]), non-maximal ROI-definites are, as far as we can see, restricted to singular count nouns (cf. [11] for Modern Spanish). Interestingly, but except for Old Spanish (cf. , widely available Romance ROI-definites go hand in hand with the availability of invariable DE or partitive articles, i.e., with that of another indefinite determiner besides the general indefinite article based on the Latin numeral UNUS. On the contrary, but now at least with the exception of Modern French, restricted use of Romance ROI-definites seems to rather be typical of languages possessing neither invariable DE nor partitive articles. Future work into this potential correlation may shed light on why Romance languages differ as to the availability of ROI-definites. Yet, what we do want to clarify is that in any case ROI-definites cannot be considered equivalents of invariable DE or partitive articles beyond pure truth-conditionality: ROI-definites but neither invariable DE nor partitive articles are, after all, definite articles fulfilling the maximality condition of the definite article at the kind level (Section 2.1). Invariable DE and partitive articles, in turn, do not come with any kind of maximality and partitive articles only etymologically but not synchronically contain definite articles . A further open issue to us at this point, especially in view of the contradictory findings for Old Spanish and Modern Francoprovençal, is the role of P° for (dis)favoring non-maximal ROI-definites.

Conclusions
This paper investigated a highly underresearched non-maximal, non-familiar, and weakly referential use of the definite article in Romance, namely, definites with a socalled representative object interpretation (ROI) licensed by the kind-oriented mode of talk. Such nominals were shown to denote inherently non-specific, semantically number neutral regular objects whose only co(n)textual relevance is their being typical instantiations of their corresponding kind, without, however, being kind-denoting themselves. The data discussed in the paper mainly came from Old Spanish and Modern Francoprovençal, though Modern Italian non-maximal definites were also taken into consideration, whenever possible. We demonstrated that Romance ROIdefinites are favored by 'habitual' readings of the predicate (recipes, treatises, narratives about what people used to do in former times, etc., i.e., on a more general level, by 'expert talk'). As for Francoprovençal, this was evinced to be the case especially in the scope of non-perfective verb tenses in prepositional or presentational complements and sometimes in direct objects. In Old Spanish, in turn, non-maximal definites often occur in the scope of non-assertive mood (imperative/subjunctive), probably due to the genre of recipes investigated, while, at the same time, such nominals preferably introduce important discourse referents. In addition, the Old Spanish data also served to show that Romance ROI-definites are positively susceptible to priming by preceding non-maximal definites. The vast majority of these findings was made sense of in the discussion Section 4, although we did not provide a final answer as to why some Romance languages seem to make wider use of ROIdefinites than others, i.e., not only with singular count nouns but also with mass and plural nouns. Yet, a tentative hypothesis in terms of a positive correlation with the availability of partitive articles and invariable DEto be investigated in future researchwas put forward.
As a closing remark, we would like to point to the fact that puzzling nonmaximal uses of the definite article not analyzable in terms of short weak definites have been mentioned across the literature for several non-Romance languages, as well. Among these languages are some Modern Northern Swedish dialects (35), Modern Basque (36), and Modern Moroccan Arabic ([37], where "DET" is the glossing chosen by the authors for definite articles with non-maximal readings).
(35) Northern Swedish (Dahl 2015: 51) Hä finns vättne däri hinken. it exists water.DEF there.in bucket.DEF 'There is ø water in the bucket.' (36) Basque (Etxeberria 2014: 340) Ane-k ardo-a edan zuen. Ane-ERG wine-DEF.SG.ABS drink AUX 'Ane drank (the) wine.' (37) Moroccan Arabic (Maas & Procházka 2022: 17) ka-n-xdːəm l-ma u ᵵ-ᵵin IND-1SG-work.IPFV DET-water and DET-clay 'I work with ø water and ø clay.' A fruitful line of investigation could be to cross-linguistically re-assess the semantics of the definite articles in (35) -(37) in light of the findings on Romance ROI-definites presented in the present paper. We sincerely hope that our contribution will be seen as an invitation to tackle these and other research questions linked to (different) weakly referential readings of the definite article in Romance and beyond.