THE THIRD WAY: Prepositional Adverbials in the Diachrony of Romance (Part One)

This paper analyzes the role of adverbials including a preposition and an adjective (PApattern) in the diachrony of Romance, from Latin to present-day varieties. In view of present-day standardized Romance languages, PA-patterns (e. g., Sp. de seguro) appear to be lexicalized »adverbial locutions«. Empirical data show, however, that they are (i) productive and (ii) systematically used as alternatives to adverbial adjectives (e. g., Sp. viene seguro ›s / he comes sure‹) and adverbs ending in -mente (e. g., Sp. viene seguramente ›s / he surely comes‹), until the 17th century. Consequently, PA-patterns are likely to have systematically appeared as a third way of forming adverbials in terms of mainstream analyticity in the Latin-Romance transition. Regarding functional differentiation, PA-patterns display affinity to circumstantial modification, which secondarily favors the development of discourse functions. Two case studies provide a fine-graded diachronic analysis that checks seven claims formulated in the Introduction. The first case concerns Sp. fijo, a learned word borrowed from Classical Latin which undergoes colloquialization (> de fijo, a la fija, por fijo, etc.). The second deals with Fr. vrai, which belongs to the genuine oral tradition. The paper confirms the systematic role of the PA-pattern in the diachrony of Romance. 1


Introduction
Modifiers developing discourse functions have been analyzed intensively during the past decades. In Romance, these studies generally chose an adverb ending in -mente such as Sp. naturalmente, Fr. apparemment, It. sicuramente, etc. in order to observe their diachrony regarding polysemy, polyfunctionality, and grammaticalization / pragmaticalization. Diachronic studies on adjectival discourse markers are less frequent, probably because the parsing of units such as Sp. bueno, claro provides huge amounts of data, with adverbial functions being only a small part. Two studies by Ocampo 1 Paper financially supported by the Austrian Science Fund (FWF), project number P 30751-G30: »The third way: prepositional adverbials from Latin to Romance«. This fascicule of Romanische Forschungen contains chapters 1-6, the following fascicule chapters 7-10 and annex 1 (cf. references in 5.1, 5.2, 7, 8.1, 8.5).
This article is distributed under the terms of the Creative Commons Attribution 4.0 license (http://creativecommons.org/licenses/by/4.0/) which permits adaptation, alteration, reproduction and distribution without further permission provided the original work is attributed. The derivative works do not need to be licensed on the same terms.

Martin Hummel
Romanische Forschungen 131 (2019) (2006a, b), who indeed tried to single out the diachrony of Sp. bueno and claro used as discourse markers, show another difficulty. Bueno and claro are discourse markers typically used in informal spoken conversation. Consequently, they are under-represented in written sources. Now, only written sources are available for diachronic research. Hence, the occurrences Ocampo digs out are unlikely to match the real diachrony of spoken language. Finally, all these studies share a more general methodological problem. The semasiological analysis of single units does not allow the shedding of light on one of the most relevant processes in diachrony: variation and selection. While speaking, speakers choose from a range of alternative expressions. In long-term diachrony, such choices may lead to the replacement of one unit by another. For the study of competition and selection, an onomasiological approach is mandatory. Onomasiological analyses tackle units that compete for conveying a shared concept or function. The method requires the reading of the corpora because we cannot be sure to previously know all the alternatives. It goes without saying that commodity favors semasiological approaches in a sort of metalinguistic invisible-hand process that biases research. However, certain compromises allow for a more manageable approach to onomasiological analyses. It is indeed methodologically legitimate to read a relatively small corpus in order to retrieve the available alternatives, which can then be digitally parsed in bigger corpora. This is basically the method adopted here.
The present paper deals with the frequent or even systematic coexistence of alternatives for adverbial functions. While reading the 1 million word corpus used for the Sintaxis histórica de la lengua española (Hummel 2014b), I observed that three types of adverbials including the same adjective compete in the same conceptual and functional domains, e. g., the series Sp. seguro, seguramente, de seguro, which may be analogously translated to English: sure, surely, for sure. Although the three units are not fully equivalent, e. g., because seguro includes adjectival and adverbial functions, while adjectival functions are excluded by seguramente and de seguro, the three morphosyntactic alternatives overlap in the same epistemic adverbial function, at least until the 17th century. In the corpus used for Hummel (2014b), a similar situation can be observed for 56 triplets with analogous morphosyntactic structure. This means that there is a systematic coexistence of three variants including the same adjective. This fact has not been thoroughly studied in linguistic research. This means that traditional diachronic studies of oral discourse markers may not fail simply because the available texts are written, but also because the role of alternatives such as Sp. de fijo, en fijo, a la fija, a punto fijo, fijamente, etc. is not duly taken into account.
The 56 triplets will be analyzed in Section 3. Before this, Section 2 briefly develops the typological background and provides a cross-linguistically valid terminology for the study. The typological background is useful, not only because other languages show similar facts, as in the above mentioned English example sure, surely, for sure, but also for the comparison of Romance languages. While the development of discourse functions by short adjectival adverbs (e. g., Sp. seguro) and long adverbs (e. g., Sp. seguramente) has been intensively analyzed by research (see Hummel 2013, 2014a,b, Hummel / Valera 2017, Company Company 2012, 2014a,b, Bauer 2001Bauer , 2003Bauer , 2010, the role of prepositional adverbials has been neglected. This paper therefore concentrates on the latter. Given the lack of research in this domain, Section 4 sketches the state of research on adverbials joining a propositional head and an adjective in Romance. Section 5 tests the availability of equivalents used in Romance for the Spanish prepositional adverbials singled out in Section 3. After this overview of Romance in general, two case studies, on Sp. fija, fijamente, de fijo, a la fija, en fija (Section 6) and Fr. vrai, pour vrai, pour de vrai, pour le vrai, à vrai (Section 7), provide a first in depth analysis of the following claims that are substantiated in the general Sections 2-5: In Romance, prepositional adverbials with the structure »preposition + ADJ« are systematic alternatives for short adverbs (adverbial adjectives) and long adverbs ending in -ment(e), at least until the modern language period, the 17th century offering the scenario of divorce. Claim B: Prepositional adverbials are relevant for the Latin-Romance tradition. They are scarcely attested in the available texts, but have probably been crucial for the genuine oral tradition that links Latin and Romance. Claim C: Prepositional adverbials are part of the major process that transforms Latin to Romance via analyticization. Claim D: Prepositional adverbials considerably expand in the period following the formation of Romance until the 16th century, as a consequence of linguistic elaboration transforming the languages from a rudimentary state into a richly developed diversity in speaking and writing. This means that a learned tradition of the educated overlays the genuine oral one (Claim B) Claim E: After the 16th century, prepositional adverbials were eliminated or marginalized during the processes of purification and standardization, which led to reduced productivity and left a treasure of lexicalized units, the so-called »adverbial locutions«. Claim F: Prepositional adverbials may either continue the genuine oral tradition of Late Latin or be a result of linguistic elaboration followed by colloquialization. Colloquialization happens when units created in the language of the educated are progressively used in informal oral registers. Claim G: Colloquialization is particularly relevant for the development and spread of discourse functions. Prepositional adverbials may lay the grounds for discourse functions that are later taken over by adverbs in -mente.
It is noteworthy that Claims B and D can be either exclusive or complementary. Some authors claim indeed that prepositional adverbials with the structure »preposition + ADJ« did not exist in Latin. According to Eberenz (2005: 618-619), the proliferation of prepositional adverbials of all types (see below PX-patterns) until the 17th century reflects textlinguistic diversification. By contrast, the complementary Claim B argues that PA adverbials were common in spoken Latin, but surfaced only in the texts as written Late Latin approached everyday linguistic reality, thereby producing written Romance. This would mean that these adverbials belonged to a well-established spoken Latin tradition in terms of mainstream analyticity. The traditional distinction of analyticity vs. syntheticity may suffice here for a first descriptive approach, but I will critically discuss this point in Section 8 with insights from this paper. As for the case studies, they have been chosen with regard to the two scenarios in Claim F: Fr. vrai represents the genuine oral tradition (Section 7), while Sp. fijo stands for the colloquialization of an item borrowed from Classical Latin (Section 6). This paper thereby aims to determine the impact of morphosyntactically related units on the diachrony of discourse functions, combining this down-stream diachronic analysis with the up-stream reconstruction of oral traditions tracing back to Latin.
The systematic investigation of the seven claims requires a long-term project of a whole team, especially because each series of variants needs an individual analysis, which often includes complex Latin-Romance interfaces. The Austrian Science Fund (FWF) has recently granted a project that will allow us to tackle most of these questions in a systematic pan-Romanic way (»The Third Way: Prepositional Adverbials from Latin to Romance«, starting in June 2018). This paper is a first approach to the topic.

Typological background and terminology
In order to provide an operational terminological basis for cross-linguistic analyses, I distinguish Type A adverbs (adjective-adverbs: Fr. couper court, Engl. to cut short), Type B adverbs (derived adverbs: Sp. rápidamente, Engl. quickly), and Type C adverbials (Sp. de fijo, a la fija, Engl. in short). Type A constitutes the genuine Indo-European tradition, not only in Romance but also in English (to run quick) and German (schnell rennen). In this tradition, the less marked form of the adjective is used for adverbial functions (Hummel 2013, 2014a, Fortson IV 2011: 211). However, in English and most Romance languages, Type B has become the adverb preferred by standard (Fr. courir rapidement, Engl. to run quickly). Type C is not restricted to the pattern »preposition + ADJ« as in Fr. à la légère (henceforth PA-pattern); it also includes nouns, as in Fr. avec rapidité (PN-pattern), verbs, as in Fr. à reculons, It. a tastone / tastoni, Pt. aos tropeções, Rom. de-a binelea (PV-pattern), and adverbs, especially those of time or space (cf. Meyer-Lübke 1972: § § 468-473), as in late Latin a foras (PAdv-pattern). I use PX-pattern as a general term for these structures, with further subcategorizations as PA-pattern, PN-pattern, PV-pattern, PAdv-pattern, etc. This paper focuses on Type C adverbials following the PA-pattern and its coexistence with Types A and B. The PA-pattern itself consists of a cluster of secondary patterns according to the preposition used, the usage of the article, the role of inflection, and univerbation: Fr. d'ordinaire, en bref, à la dure, acertes, etc. (more details in 4.1).
From a general linguistic point of view, Hengeveld (1992) classifies the languages of the world into three types of adverbs. Flexible languages use Type A, differentiated languages Type B, and rigid languages use other word-classes for adverbial functions. Although he basically accepts these categories, Salazar (2007) convincingly argues that the three types often coexist in the same language, as is indeed the case in Romance and English: happy, happily, with happiness; short, shortly, in short. Hence, a variationist approach is required rather than strict typological classification.
Moreover, a strictly word-class oriented approach would rule out syntactic PX-adverbials, which is problematic from a functional point of view. The use of other word-classes such as nouns for adverbial functions, which is the structure Hengeveld aims at for Type C, are rare in Romance (e. g., Sp. pasarlo bomba ›lit. to enjoy oneself bomb‹, ›to have a good time‹, Fr. boire nature ›lit. to drink sth. nature‹, ›to drink a product without additional ingredients‹), but typological studies show that the conversion of nouns or verbs to adverbial functions may prevail in other languages of the world (Hengeveld 1992). In this paper, PX-patterns are considered as full members of Type C, not only as »paraphrases«, as is often the case. »Paraphrase« implicitly means the replacement of an allegedly more basic unit, generally a member of a word-class. The paper argues against such word-class biased approaches. »Word-class obstinacy« (Maas 2010: 82) is one of the reasons why Type C has not been acknowledged as a diachronic mainstream that constitutes a relevant third way in the Latin-Romance transition in the domain of adverbial functions.  (2017), the Old Spanish core of the PA-pattern had indeed directional semantics (e. g., a derechas), which was later extended to manner semantics (e. g. a solas, a secas). 2 If this is correct, PA-patterns with manner interpretation are Romance innovations that developed by analogy from time and space PX adverbials tracing back to Latin. Despite these isolated observations, no systematic account of the role the PA-pattern plays in the diachrony of Latin is available at present. Consequently, the detailed documentation of this pattern in Latin is a major desideratum for research. Section 5.1 provides the results of a first spot check.
In Romance linguistics, prepositional phrases with adjectives such as Sp. a las claras, a medias, a solas are more neglected by grammars than the also underexposed Type A (see state of the art for Spanish in Rodríguez Molina 2014: 742-743). This is especially the case for modern grammars and manuals, while older authors dealing with diachronic aspects mention at least some PA adverbials ( Nomen« 3 (Meyer-Lübke 1972, vol. 2: § 418). In principle, Meyer-Lübke's term Nomen includes adjectives, but the discussion of the data concerns nouns in the narrow sense.
The reasons for disregarding PA-patterns are twofold. Linguists specialized in grammar and syntax tend to consider paraphrases as secondary solutions or underlying structures that may be genetically relevant. The very term paraphrase mirrors the conception of replacing a primary item. Thus, the main reason for neglecting PX-patterns is word-class obstinacy. In line with this, I have criticized the exclusion of prepositional phrases from Type C in typological approaches (Section 2). At school, standard questions ask if slow in to drive slow is an adjective or an adverb, that is, we become used to analyzing such phrases in terms of word-class and not of simple syntactic polyfunctionality. Discourse markers are often considered strange units, normally excluded by traditional grammars, or a class apart because they do not permit answering questions for word-class membership, while their function in syntax and discourse can easily be identified.
The second reason for their being disregarded is that PX-patterns are perceived as lexicalized locutions, that is, units of the lexicon to be studied by phraseology, not by syntax or morphology. PA adverbials such as Sp. a las buenas are generally analyzed as lexicalized adverbials. Lexicologists use terms such as locuciones adverbiales in Spanish (García-Page 2008: 120-126, with additional bibliography). Consequently, these locutions are well documented in the DLE, the dictionary of the Spanish Academy (Real Academia Española 2014). From this point of view, the PA-pattern could be considered almost unproductive today except for the productive »in an ADJ way / manner«-patterns (see, among others, García-Page 2008: 120-129; 2014; Penadés 2005, 2012; bibliography in Aguilar 2010). This may be justified from a present-day perspective, but not when considering the formation period up until the 17th century. Even for present-day Spanish, García-Page (2008: 120, 126) points out that it is almost impossible to provide a complete inventory of adverbial locutions because some patterns are still productive.
Grammars of the old languages also tend to restrict their analysis to locutions and univerbations, possibly because the present-day usage affects their perception of the past. Rheinfelder's Altfranzösische Grammatik, for example, does not mention Type C in the chapter on adverbs (1967: 70-73). Type C is indeed mentioned in the chapter on prepositions (vol. 2: Romanische Forschungen 131 (2019) 316-346), but the PA-pattern is lacking. Rheinfelder's focus lies on lexicalized locutions, not on productivity. A spot check of the Altfranzösisches Wörterbuch (Tobler / Lommatzsch 1969) -which is a dictionary, not a grammar! -shows that PA-patterns were indeed abundantly used in Old French. To provide just one example, in the entry for petit we find analytic variants such as a (bien) petit (today still: petit à petit), de petit, par un petit, por un petit, which are completely out of use in modern standard. Examples such as Fr. à la vite, al isnel, en brave, can be added. It is noteworthy that PA-variants not accepted in standard can still be found in informal present-day texts, e. g., Fr. »car en réel elles préfèrent les bad boys gros cons« (my Internet corpus).
Univerbation (e. g., Fr. acertes) is a particularly salient process of grammaticalization and lexicalization. It necessarily means that such synthetic forms originated from productive PA-structures. According to Echenique (2008b: 392; cf. Vicente Llvata 2011), in the domain of adverbial locutions, grammaticalization combines with lexicalization. Now, the univerbation process presupposes frequent usage and, in most cases, usage in spoken language. The orality-literacy interface is therefore a crucial issue. In sum, PA-patterns require a more differentiated diachronic analysis than paraphrases of the »an ADJ way / manner« type.
The way grammars analyze adverbs is also revealing. Subsequent to word-class, most modern manuals, grammars, and linguistic studies prioritize the semantic and functional classification of adverbs into groups named adverbs of manner, point of view, intensification, etc. The third priority for classification is the analysis according to the distributional syntactic feature »modifier of X«. Since grammars traditionally refer to the unit »sentence«, sentence adverbs modify the largest unit. Discourse organization (e. g., »connectors«) is not in focus, but new grammars sometimes dedicate a separate chapter to this domain, generally under the label of »discourse markers / particles«. Since the definition of discourse markers is not restricted to discourse organization but includes all kinds of subjective presentation of a given proposition, such chapters overlap with the traditional sections on »sentence adverbs« and »intensification«. The separation thus follows more the demarcation of oral data (»discourse markers«) and written data (»sentence adverbs«).
It is striking that morphosyntactic aspects of adverbs play only a minor role in grammars, if any. Generally, morphosyntactic features are briefly mentioned, but the further analysis is illustrated only with Type B adverbs in -mente. Morphosyntax is also completely lacking in the long monography on adverbs by Sonntag (2005). The grammars' silence on morphosyntax has probably to do with (i) the dogma of the invariability of adverbs, which is questionable (Ledgeway 2017, Hummel 2015, 2017b), (ii) the concentration on one type of adverb, the one preferred by standard (-mente) (but adverbial suffixes are often analyzed as inflectional morphemes, see, e. g., Giegerich 2012, for English -ly), and (iii) the analysis of PA-patterns as simple paraphrases for prioritized word-classes.
Studies on the Romanian adverb are more sensitive to morphology and morphosyntax. Consequently, PX-patterns are mentioned, albeit not systematically analyzed (Chircu 2008, Vasile 2013). Chircu (2008: 68-73) quotes examples for all PX-patterns in Romance except for Old French. However, Type C in general, and the PA-pattern in particular, is not seen as a relevant third path to express adverbial functions.
As for Italian, according to personal readings of Italian literature and the observation of spoken Italian, the PA-pattern is still frequently used, if not productive (da solo, d'improvviso, di solito, di continuo, a lungo, in secco, in sordina, alla larga, al più presto, sul serio, di nuovo, di recente, di sicuro, di molto, di nascosto, per prima, in grande, di certo, etc.). Rohlfs As for Spanish, the Nueva gramática de la lengua española (RAE-ASALE 2009: § § 30.15 to 30.17; cf. Pavón Lucero 1999) rightly insists on variation and stresses the continuum between adverbial locutions and productive prepositional phrases. The differentiated analysis in this grammar reflects the substantial work on Spanish adverbial locutions and phraseologisms in general during the past decades. However, nearly all publications concern present-day synchrony, without clearly working out synchronic diasystematic variation (see, however, the publications of Echenique and her group, who seriously tackle this aspect). Spanish informants consider PA adverbials as spoken substandard (see, e. g., Fuentes Rodríguez: »variante poco culta« ( To conclude, this brief and certainly incomplete bibliographical overview shows that there is some rather unsystematic work on PX-patterns, but almost nothing on the PA-pattern. The main reasons for this state of the art are word-class obstinacy, the assumption of secondariness (»paraphrase«), and the assumption of lexicalization. Historically, lexicalization went hand in hand with standardization, which excluded productive Type C adverbials, accepting only fully established units. But this does not necessarily hold for informal spoken language and dialects. Until the 17th century, the usage and formation of PA adverbials is a systematic and productive process, as we will see now. In total, 56 different adjectives are used in the corpus as the lexical base of triplets constituted by Type A, B, and C adverbials. In 14 series, the adjective undergoes gender and / or number agreement.
The following examples illustrate uninflected (1) and inflected (2) cases (often with a definite article), as well as the usage of PA adverbials with a copula verb (3): (1) (a) tan de ligero avías de conjecturar de lo passado nocibles sospechas (13th c.) The coexistence of Type A and Type C (PA) is firmly established until the 17th century, that is, the epoch where purism and standardization start to intervene. 4 It seems that alternatives tend to drift away from initial synonymy in diachrony, at least in some cases. While saber seguro / seguramente / de seguro could be used as synonyms meaning ›to know with certainty‹ until the 17th century, saber seguramente means ›to probably know‹ in present-day Spanish, and de seguro is very marginal. Type C often falls out of use in standard, except for some lexicalized adverbial locutions that survived the selectivity of purism and standard. In the case of saber cierto / ciertamente / de cierto / por cierto, which initially overlapped 4 In further research, a study on those cases where Type A is not paralleled by Type C, and vice versa, could provide complementary insights on tendencies of functional differentiation. as epistemic adverbs of manner meaning ›to know with certainty‹, they all turned out to be used as discourse markers outside VP. The tradition of productively using Type C declines in Spain, but it is conserved in America, at least in rural varieties. All the examples from the 17th century and later (1b, 2a, 2c, 3b) stem from America. In Europe, the tradition stops abruptly, with a set of lexicalized adverbial locutions remaining in the lexicon (e. g., por cierto, a secas). In this sense, the PA-pattern undergoes a marginalization process similar to that of Type A in European standard(s) (Hummel 2013).

Morphosyntactic alternatives from an onomasiological perspective
Whereas in 4.1 the adverbials have been analyzed in terms of morphosyntactic alternatives, this section develops an onomasiological view of their common functions in sentence and discourse / text. In the following, all the 56 TypeABC-triplets of the corpus are listed according to functional paradigms. 5 The first group integrates units conveying epistemic or evidential functions in discourse: The Type C adverbials of this group do not simply focus on the way an event is realized, but on circumstances such as a las malas ›in bad terms‹, a tuerto ›clumsily‹, en secreto / a oscuras ›secretly‹, etc.
Romanische Forschungen 131 (2019) In the same vein, in the fourth group the result of an event is not referred to as a property of an entity, but as a circumstance, e. g., to do something completo ›completely‹, medio ›half‹: The last group conveys a local circumstantial perspective on the event: The functional classification of the triplets provides evidence for their paradigmatization according to crucial functions in sentence and discourse. Taken together, these functional domains are relevant for foregrounding the circumstances propositions are subject to: time, place, perspective, truth, source of knowledge, focus, specification, intensification, etc. This means that the members of the triplets compete for crucial functions of discourse presentation and organization. The fact that Type C adverbials do not simply focus on the core function of manner, but tend to add a circumstantial aspect, is a consequence of the fact that the head of the phrase, the preposition, creates a distance in the relation between the modifier and the modified, if compared to Type A and B. Further research will be needed to define how each of the prepositions semantically defines this distance. It is important to see that, in a certain sense, discourse functions also refer to circumstantial aspects of propositions and smaller units. It comes as no surprise, then, that discourse functions, in the broad sense of ›subjectively presenting and organizing information in discourse‹, play a prominent role the deployment of these triplets. Text, speaker attitudes, communicative goals, and discourse organization are circumstances that help situating the propositions in the stream of discourse.
The fact that circumstantial concepts appear as the common denominator of Type A, B, and C follows from the fact that Type C is the most restricted type in terms of functions, since the broadest category, Type A, includes adjectival and adverbial functions, and Type B adverbs in -mente figure in all grammars as an extremely polyfunctional group. Consequently, the functional overlapping tends to follow the functional restrictions of Type C to circumstances. However, the range of functions realized by Type C adverbials also exceeds the domain of intersection.
The following examples show the use of »de + ADJ« to refer to source-cause (see more cases and variants in Hummel 2014b: 702-703): (4) (a) no pudo venir a verte de turbada [Celestina] (b) de modo que aunque algunos lloran, es sólo de cumplimiento y la risa de contento [Rey gallo] It is not excluded that adverbial saber de cierto ›to know with certainty‹ was initially motivated by the idea of source ›to know from certainty‹. Future research will have to separately analyze the patterns »de + ADJ«, »a la + ADJ«, »por + ADJ«, etc. in order to better understand the relation the preposition establishes between the modifier and the modified. In the present paper, I look only at the domain of functional intersection with Type A and B. From this point of view, the circumstantial conditioning of propositions becomes salient. Regarding my initial enthusiasm, claiming that PA-patterns might have been a systematic third way for all adverbial functions (Hummel 2014b), the analysis of the data in this section points rather to a more restricted competition in the domain of circumstantial modification. However, these restrictions could also be the result of progressive functional differentia- tion from initial synonymy, as in the case of seguro, seguramente, de seguro. Be this as it may, there undoubtedly existed a systematic coexistence and competition of the three morphosyntactic variants, at least in the domain of circumstantial adverbial modification. This includes functions in discourse, text, sentence, information structure, and emphasis. Given that the number of discourse markers and sentence adverbs is never exceedingly high, the 56 triplets detected in Spanish certainly cover a significant part of these functions in Spanish until the 17th century. Consequently, the long-term role of PA-patterns might have been important or even crucial in such domains. They should not be overlooked when we tackle the diachrony of these functions.
In terms of hypothesis, this logically includes the possibility of discourse functions being first developed by Type C, as we will indeed see in the case study on Sp. fijo (Claim G). Consequently, the rise of discourse functions assumed by Type B adverbs in modern Spanish might in some cases be related to the disappearance of PA adverbials in modern standard Spanish. Thus Type B might have taken over discourse functions that were previously expressed by Type C. The substitutive relation of the three morphosyntactic adverbials is interesting in this sense. As shown above, Type A and Type B adverbs may assume, amongst many other functions, those realized by Type C. Now, upcoming standardization clearly favored Type B (as also in English, Hummel 2014a). Standardization thus could have exerted a pressure, creating a »sponge-effect« whereby Type B adverbials assumed functions realized by other adverbials. This is very clear for English adverbs ending in -ly replacing older short forms in standard English. As shown in Hummel (2014a), the replacement of Type A adverbials by Type B is a general tendency in the modern Western linguistic culture, in the standards of both Germanic and Romance. We are used to deciding if we should use first or firstly, more important or more importantly, slow or slowly, etc., the situation in Romance being very much the same. This does not mean that this »sponge-effect« was the main reason for Type B adverbs developing discourse function, but it may have played a role in some cases. Studies in diachrony should consider this possibility.

The PA-pattern in Latin and Romance
In the previous sections, I have repeatedly insisted on the impact of purism and standardization on the coexistence of and partial competition between Types A, B and C. This impact is obviously restricted to the standards of writing and to formal oral discourse. To express it differently, this does not mean that the impact operated with the same strength on the informal oral traditions. In fact, informants classify many of the PA-patterns as »spoken«, »informal«, or »colloquial«, and authors such as Moreno de Alba (2009: 638) allude to their being used in rural varieties of American Spanish, which perfectly matches with the rich documentation in American dialects by Kany (1970). In Sections 6 and 7, I will come back to the survival of PA-patterns in rural varieties with uninterrupted oral tradition. Section 5 instead deals with the origins of the PA-patterns. Subsection 5.1 therefore takes a closer look at Latin, while 5.2 adopts a reconstructional diachronic perspective.

Latin and Medieval Spanish
Written Late Latin used -iter as the canonical suffix for the derivation of adverbs from adjectives (e. g., realiter). Short adverbs were also abundantly used, much more so than in Classical Latin (Löfstedt 1967, Hummel 2013), as well as paraphrases with the ablative mente, especially in Christian texts (Queirazza 1970). By contrast, according to the bibliography (see Section 3), the PA-pattern did not -or almost not -exist in Classical Latin. Since further research is lacking, this was implicitly extended to Latin in general. A spot check in the Thesaurus Linguae Latinae (TLL) shows, however, that the PA-pattern can be traced back to the first literary texts of Latin in the 3rd century B. C., displaying continuous use until the high-medieval period. One such unit with continuous usage -possibly the only one -is the adverb denuo. 7 It figures several times in Plautus' comedies (also ab integro, ex integro (de integro in Cicero), ex improviso, de improviso, de repente (see also Torrego 2002)). Old Latin denuo reappears with remotivated analytic orthography as de novo in the 7th century. The fact that denuo shows univerbation in the first documents provides evidence for its antiquity and previous grammaticalization with the metaphorical meaning ›again, repeatedly‹. PA adverbials can also be retrieved in classical poetry (Lucretius, Virgile): de plano ›clearly‹, ex longo ›for a long time‹, as well as in technical and legal texts: de propio habere ›to have as one's own‹. In the imperial period, Late Latin, Christian texts, and in Medieval Latin we 7 The spot check has been conducted by Jairo García Sánchez, who is a member of the FWF project, with the support of Benjamín García-Hernández. have found: de praeterito ›in the past‹, ex / de longinquo longinquo ›from a distance‹, de proximo ›closely‹, etc.
These data suggest some interesting conclusions. Although PA-patterns are not completely lacking in written Latin, as it is assumed in the bibliography, they are very marginal in the spot check data. This could reflect a situation where typically spoken devices were almost completely rejected by writing, leaving some traces in written texts, which is exactly the situation we have in present-day Romance. The few examples provide evidence for a tradition that splits into an oral branch (e. g., de novo) and a written educated one, as in the last examples. In written Latin, examples from the oral tradition are more marginal than others, which slightly progress in Late Latin.
In Medieval Spanish, PA-patterns are frequently used. In her study on the formation of »prepositional locutions« (PX-patterns) in highly formal legal and historiographic Medieval Spanish texts, Codita (2016: 13-17) observes that, in contrast to what is generally assumed, these locutions were very productive and highly frequent at that time, to the point that she considers them the most relevant resource in the process of elaboration of the new Spanish language. In her inventory figure items such as a lo largo de, a longe de, a mal de, a medio de, en lo alto de, en luengo de. It seems that the formal texts under scrutiny testify to a discourse tradition with prepositional locutions (PX-patterns in general) used as connectors (see also Kabatek 2005: 156-159). For reasons that are not overtly mentioned, the PA adverbials mentioned by García Valle (2010a, b), in the same type of text, are almost absent in the list of locutions provided by Codita (2016: 135-136; 219-225). García Valle (2010a,b) quotes examples such as a tuerto, por firme, de conocido, en doble, en alto, en ancho, en luengo, en gordo, a medias, de nuevo, a sueltas, a osadas. This might be a reflex of productive elaboration going hand in hand with individual variation at a time when standards did not exert a uniforming constraint on the writers. This is plausible insofar as the organization of written texts always differs from the organization of informal spoken discourse. Consequently, written Medieval Spanish had to innovate in this domain. Interestingly, the PA-patterns found in formal texts are rather informal if compared to other PX-patterns used as connectors.
As far as the development of discourse functions is concerned, Late Latin adverbs ending in -iter provide abundant evidence for the expression of such functions. If we look at St. Augustine's Late Latin sermons, we find a significant number of adverbs with circumstantial and discoursive functions: aeternaliter, contemptibiliter, incomparabiliter, inevitabiliter, . Given the fact that -iter disappeared in the Latin-Romance transition, PA-patterns, and PX-patterns in general, could have been one way to functionally and conceptually replace them (see vero in Annex 1), although -mente was certainly the best candidate for this purpose.

Corresponding PA-patterns in Romance
The fact that the PA-patterns appear in early texts written in Romance, being abundantly used in the Medieval language period, supports the hypothesis of a long oral tradition in Latin that surfaces with Romance, that is, as the written tradition approached the spoken one (Claim B). Since the simple analysis of written texts is biased by the implications of writing, a complementary reconstructional approach is required in order to better understand the underlying oral tradition.
The oldest attested PA-pattern, Old Latin denuo and its later variants, has already been mentioned in 5.1. This adverbial reappears in Romance: The Old Latin orthographies of denuo and some of its direct followers, reflects inherited pronunciation (Pt. novo is the canonical example of the regional persistence of final /-u/ in Vulgar Latin, which is held responsible for the closed stressed vowel /-o-/ in the first syllable). The fact that its use is continuously attested from Old Latin to Romance also supports its belonging to the oral tradition. In addition, the remotivated orthography Lat. de novo (7th century) provides evidence for how canonical writing handles underlying oral models, while some Romance variants come close the Old Latin spelling. Hence, the Romance variants may be more telling as attempts to reconstruct the diachrony of spoken Latin than some of the attestations in written Latin texts.
For historical comparative linguistics, the occurrence of the same phenomenon in all Romance languages provides evidence for (i) a common Latin origin and (ii) its use in spoken Latin. If we only look at French, Italian, Portuguese, and Spanish, this methodology may be misleading Romanische Forschungen 131 (2019) because these languages share not only a common oral tradition but also a closely related tradition of written Latin, which secondarily became the model for written Romance or »Neo-Latin«. Consequently, it is important to include languages and dialects such as Sardinian, Romanian, or Friulian, which did not undergo standardization (or if so, then only late).
Annex 1 provides a synopsis of equivalents for the 56 Spanish PA-patterns in Catalan, French, Friulian, Italian, Portuguese, Sardinian, and Spanish. The data have been retrieved from corpora and native speakers. While the parsing of corpora was restricted to morphosyntactic equivalents, the native speakers also provided equivalent PA-patterns that contain other adjectives than in Spanish.
A thorough corpus analysis of such data requires years of research, which is indeed intended in the framework of the FWF project. The present paper only aims to provide an overview. This overview clearly shows that the PA-pattern is used in all varieties under scrutiny in the same or similar functional domains. My case studies on Sp. fijo and Fr. vrai (Sections 6 and 7) will provide further evidence for clear structural analogies. The data in Annex 1 include triplets that reflect a shared oral tradition (e. g., the triplets starting with Sp. derecho, largo, grueso), and triplets tracing back to shared intellectual traditions of writing and speaking of the educated (e. g., continuo, especial). However, the two traditions are not fully separated but linked by interfaces permitting the osmosis of units in both directions. The example of rewriting (and possibly repronouncing) Lat. denuo as de novo shows how the educated tradition integrated spoken language, while cases such as Sp. a la continua are »colloquializations« of the language of the educated (Claim F). I will come back to the latter in the case study on Sp. fijo. At any rate, the data in Annex 1 provide clear evidence for the existence of related pan-Romanic traditions of using analogous PA-patterns. Fijo is used at the left periphery of the sentence (6), as a member of VP (7, 8) and as an affirmative answer (9). In all the examples, fijo conveys the speaker's certainty about a propositional content. Commenting on Albelda (2002), Hummel (2012: 167) observes that it is almost impossible to retrieve traces of this epistemic discourse marker in corpora, even in large corpora such as CREA. Since the epistemic functions of fijo were in use in the recent language period covered by the CREA corpus, the only plausible explanation is that this corpus does not objectively reflect informal spoken Spanish. Formality and code seem to operate as a filter.
Having discovered the systematic usage of the PA-pattern in diachrony, Hummel later (2014b: 700) suggests reconsidering the diachrony of the discourse functions of fijo by including the PA-pattern, especially de fijo. Kany (1970: 332) provides further evidence for the use of the variants a la fija and en fija in Latin American dialects. This was the starting point for a diachronic case study on fijo, which I presented at the Congreso Internacional de Lingüística Hispánica (Leuven, 20-22/9/2017). At that time, two papers by Lavale-Ortiz (in print a, b) including the PA-pattern in the diachronic analysis were already in print. These two fine-grained semantic and syntactic analyses describe the diachrony of fijo from a cognitive linguistic point of view, dealing with the conceptual semantics of fijo. This allows me to focus here on crucial aspects regarding the claims mentioned in the Introduction. While Lavale-Ortiz mainly uses the diachronic CORDE corpus, my study is based on the three subcorpora of the CDH corpus (CDH nuclear, S. XII-1975XII- , 1975XII- -2000  fijo remains close to its adjectival nature. 8 Combinations with full verbs are mainly instances of mirar fijo (177 cases). Epistemic fijo with full verbs turns out to be marginal (see below). Until the eve of the 18th century, fijo and its older variant fixo were primarily used in the following domains and contexts. The first domains of usage are astronomy (10) and religion (11, 12), together with first abstractions targeting the domain of thinking and permanence in time, especially regarding rules and law (13): (10) Et sepas que el grado con que la Estrella se para en medio del cielo; es fixo & de una manera en todos los lugares (1277, Alfonxo X, Libro del Alorca) (11) ya dixe en el prohemio que cada vno deue ser fixo en los fundamentos e artículos de su fe (1422-1433, Guadalfajara, Traducción y glossas de la Biblia de Alba) (12) el que fixo quisiere estar en la catholica fe (ibid.). (13) E porque tuviesen fixo en la memoria este cuento, Puso en los diez mandamientos de las tablas seyscientas e treze letras (1417, Villena, Tratado de lepra) These domains prevail until the end of the 15th century. We also find the first occurrences of mirar fijo, in literary texts (14): (14) mirandolo fixo, retroçederya (1424-1520, Anonymous, Cancionero de Juan Fernández de Íxar) In the 16th century, depending on the type of text, most of the occurrences concern physical and spatial fixedness, as in example (15), with some extensions to temporal permanence (16) and moral constancy (17) temporal fixedness and permanence were available at this stage in the semantics of fixo / fijo. During the centuries, uses referring to fix contracts and jobs naturally follow this cline.
The first epistemic functions in discourse appear as early as the 16th century (18), keeping close to the basic meaning ›to be fixed / definite‹ (18,19,20), but already with parenthetical marking with y or comma: (18)  Although the complete conceptual semantics of fijo was in use at that time, which by the way is still the case in present-day Spanish, the epistemic occurrences undergo further abstraction, coming closer to the meaning ›to be certain / true‹ from the speaker's point of view: The analysis of »verb + fixo / fijo« thereby seems to corroborate the hypothesis of diachronic syntax deletion being the first step on the way to discourse markers (Company Company 2004): es fijo que > fijo que > fijo (see present-day examples (6) to (9)). As far as the hypothesis of syntax deletion is concerned, we have to be cautious insofar as syntax often changes when oral items pass to written language, e. g., when we simply say »OK« or »Fine!« but write It is ok / fine / true that … (see Hummel 2012: 331-357). In other words, using fully explicit copular constructions in written language does not mean that the same occurs in spoken language. In the case under scrutiny, the low frequency of es fijo que, with 22 occurrences over time according to the data in Lavale-Ortiz (in print b), calls into question this grammaticalization path, in particular because the first occurrence of possibly reduced fijo que stems from 1970, according to the same author. What we can say is that copular constructions formed one path where epistemic functions were promoted. The examples above (18-22) show further that this process happened in a type of written discourse that comes close to spoken colloquial language. This supports the colloquialization claims (F, G).
Emphasizing the role of individual pioneering obviously includes the question of how this innovation became socially accepted. According to the corpus, social acceptance only occurred towards the end of the 18th century, if we take use by different authors and frequency as criteria. While the example (23)  Only 42 epistemic occurrences with copular verbs can be detected in total, 22 of them in the 19th century and only 4 in the 20th, which thus marks a drastic decline. However, epistemic use with copular verbs clearly prevails over combinations with full verbs. Combinations of epistemic fijo with full verbs are rare and sporadic (8 cases in total). The first stems from Mexico (Mx) (27), the next from Spain, separated by more than 200 years: In (27) and (28), fijo means ›to know with certainty‹, that is, it has no discourse function expressing an epistemic evaluation by the speaker. It is noteworthy that it is rather the collocation with the epistemic verb saber than fijo itself which creates the epistemic meaning. Fijo simply adapts to the context: to fixly know. As a consequence, although combinations with full verbs may convey epistemic meanings, only copular constructions clearly develop the epistemic point of view of the speaker. In my corpus, only one instance expressing the speaker's subjective view can be found in the context of a full verb: In sum, the data confirm the difficulties of retrieving diachronic traces of the epistemic discourse marker fijo in written corpora. The data certainly reflect social and regional expansion, but the low frequency of the »verb + fijo « pattern does not display evidence for any type of grammaticalization process. All epistemic discourse functions are marked by syntax or parenthesis, and all epistemic modifiers of the verb can be analyzed as contextual adaptations to an epistemic verb, which is a way to overtly mark this function, with the exception of (29).

De fijo
Let us now check if PA-patterns could have prepared the ground for the present-day discourse functions (see examples (6) to (9)) (Claim G). Before starting, it is important to bear in mind that fijo is not an inherited unit, that is, a unit transmitted by spoken Latin, but a word Romance directly borrowed from Classical Latin in the discourse of the educated (see details in 6.8). Consequently, fijo is a candidate for what I have called in the Introduction »colloquialization«. The hypothesis of colloquialization (Claim F) assumes that PA-patterns were added to Type A or B as the discourse of the educated penetrated informal spoken Romance. Consequently, the usage of »PREP + fijo« should preferentially occur in contexts marked by informal orality. According to the CDH corpus (all three subcorpora), de fijo first occurs in the 18th century, with an absolute frequency of five cases, which are cited in the following: These examples are very interesting from both the empirical (i) and the theoretical (ii) points of view.
(i) The most important finding is that de fijo systematically realizes epistemic functions from its first attestation in 1732. This also holds for the very first attestation cited by Lavale-Ortiz (in print b): (35) ke si lo ubiera otro kativado sino Ali, yo lo abría denegado de fijo (1600, Anónimo) The uncommon orthography in (35) indicates that de fijo was indeed first used in the informal spoken language of speakers with a low level of education. This fact is supported by the systematic usage of the folk orthography de fijo, with only two instances of etymological de fixo. As shown by the metalinguistic commentary in (83), this probably means that the (educated) etymological pronunciation [fikso] was completely unusual in this case, while it was still widely used in the structure »verb + fixo «. The further development of de fijo is impressive, in terms of frequency, with 400 instances in the 19th and 119 in the 20th century. All instances but two convey an epistemic meaning, the two exceptions expressing a local concept (residir / establecerse de fijo ›to dwell / settle permanently‹). As for regional distribution, the corpus contains 313 Peninsular and 211 American cases. While the structure »verb + fijo « analyzed in 6.2 remains at a more sporadic and productive level over time, de fijo becomes lexicalized as an epistemic adverb during the 19th century. Its frequent usage in the idiolect of some authors even points to fashionable usage in the last decades of the 19th century (see also Lavale-Ortiz, in print b). As a consequence (see Thompson / Mulac 1991), de fijo undergoes rich syntactic and pragmatic variation, as shown by the following examples: (1929, Es, González Anaya, La oración de la Tarde) (44) De fijo, de fijo, el Gobierno de Madrid sabía ya a tal hora que una heroica pitillera marinedina (1883, Es) (45) pero si la prensa coge este asunto por su cuenta, de fijo haremos del profesor de Graz el hombre más feliz de la tierra. (1884, RD) (46) -¿Este? ¡De fijo que sí! Le creía curado de espantos, como todos nosotros, pero estoy viendo que (1895, Es) (47) para practicar ese trabajo razonado hoy, ¡aunque tan bárbaro ayer! pues será de fijo la que le haya servido al autor para verificar los ensayos, que dieron por resultado (1889, Es) (48) Yorick (De fijo lo sabe ya, y viene buscando quimera.) (1867, Es) These examples anticipate all the variants of simple »verb + fijo « attested in present-day Spanish (see examples (6) to (9)). In the second half of the 20th century, the use of de fijo decays. There are only 16 cases attested after 1949. The decay of de fijo in the second half of the 20th century coincides with the rise of the discourse marker fijo.
It is noteworthy, in this context, that three of the first five occurrences, in the 18th century, contain saber de fijo (31, 32, 33), to which we could add the synonym constar de fijo ›to be known with certainty‹ in example (30). This seems to anticipate the present-day usage of saber fijo (example (28). There are 63 occurrences of saber de fijo in the 19th and 20th centuries, which could have prepared the ground for saber fijo. Structurally, the step from saber de fijo to saber fijo is a small one. In a historical context, where youth language and informal communication show a general preference for using so-called adverbialized adjectives (Type A), this transition seems natural. It can indeed be shown that the pattern »verb + fijo« becomes usual in the 20th century. The process starts with non-epistemic verbs, especially mirar ›to look‹. In the 20th century, 69 of 101 occurrences of »verb + fijo « are realized by the collocation mirar fijo ›to look strait in the eyes‹. Hence, seen the other way round, the Type A pattern »verb + fijo« was also paving the way for acceptance of epistemic fijo. Nevertheless, we still have to wonder why de fijo is richly attested in the corpus, while its putative successor, the discourse marker fijo, is not. Possibly, the latter was simply not discovered by literature (see the discovery of a la fija by regional literature in 6.4).
Looking back at the data, it is useful to separate the epistemic adverb in saber de fijo ›to know with certainty‹, which expresses certainty inside the proposition, from the discourse marker de fijo sabe que ›it's sure he knows that‹. The discourse marker function does not appear in the ex-Romanische Forschungen 131 (2019) amples from the 18th century (30-34), but example (36), the first clear case, in 1828, shows that the development was fast. As for the collocation with saber, the first clear case of a discourse marker function is quoted in (48). Consequently, diachrony follows the path saber de fijo ›to know for sure‹ > de fijo sabe ›he surely knows, for sure he knows‹. This means that the epistemic adverbial function of de fijo appears from its first attestations, while the extra-propositional discourse functions are a second step of development.
(ii) The development of de fijo is also very interesting for diachronic theory. De fijo has the epistemic meaning from its very first attestation. This means that we cannot even use the term of grammaticalization, understood as a diachronic process leading from lexical meaning to grammatical meaning. Further differentiation seems to be required. In Hummel (2012: 359-404), I have argued that grammaticalization theory needs a ground zero level called grammaticalization 0 , which would be more or less equivalent to what is generally called the (first) »local context« from where grammaticalization starts. My proposal was more radical insofar as I suggested considering grammaticalization 0 as all the cases where a unit becomes dominated by the grammatical force of another unit (word, phrase, sentence, discourse). In this sense, placing the modifier slow in NP (a slow car) is argued to realize a type of grammaticalization 0 different to that in VP (to drive slow), a difference that we would describe in terms of adjectival vs. adverbial function. This means that the range of phenomena included in the term at the productive level is larger than generally assumed. Now, in some cases, textual or pragmatic force may layer with grammaticalization 0 . Following this logic, pragmatic domination may lead to pragmaticalization, e. g., using Sp. obvio or fijo in the slot of a yes / no answer in discourse converts it into an affirmative particle by the force of pragmatics. Thus, the discourse functions developed by de fijo would be a consequence of pragmaticalization rather than grammaticalization.
It seems that this is exactly what happened with de fijo. Inside the prepositional phrase, fijo is dominated by the grammatical force of the preposition de, which activates a circumstantial interpretation, as shown in 4.2. This would correspond to the level of grammaticalization 0 . The preposition de itself is probably too weak and vague to activate a specific epistemic function. This functional specification is instead supported by the conceptual meaning of fijo, more concretely by the semantic development of the adjective described in 6.2. As shown in 6.2, the ground has indeed been previously prepared for fijo to assume abstract functions in the domains of temporal and spatial permanence as well as certainty. The first attestation of the combination »copular verb + fijo « with epistemic function, in 1726 (example 22), appears at the same time as the first attestation of de fijo, in 1732 (if we put aside the isolated first attestation in 1600 in (35)). The epistemic function seems to have been »in the air«. The most straightforward ways of expressing this function were (i) copular constructions (es fijo que) and (ii) de fijo, while the same function is rather marginal within the complex variety of functions and meanings realized with other combinations of »verb + fijo «.
Analogy also supports the epistemic meaning of de fijo, since the paradigm of adverbials with the structure (de + ADJ) was already available, as shown in 4.2 (e. g., de cierto, de seguro and de vero), in the epistemic domain. This means that a conceptually and functionally adequate unit could assume an epistemic function following this paradigm. All this supports the hypothesis of a network type of diachronic influence on fijo rather than a linear type. Linearity surfaces in this network, as de fijo > fijo becomes a mainstream cline for the development of the epistemic discourse marker in the 20th century.
As for the Claim B of the PA-pattern being related to the oral tradition, the data show that 346 out of 523 examples can be classified as »oral« according to the criteria »direct discourse, free indirect style, inner monologue, use of first and second person«. In addition, most of the other cases are written in an argumentative, often polemic and, at least virtually, dialogic style, as in example (47). Hence, there are clear traces of subjacent orality in the written texts that are available for diachronic analysis. In terms of reconstruction, this supports the hypothesis of PA adverbials belonging to the oral tradition of Romance. The fact that the inflected variant de fija (que) also occurs, according to Lavale-Ortiz (in print b; cf. RAE-ASALE 2009: 2377), provides further evidence for the belonging of de fijo to informal oral registers (Claim B).

A la fija
As with de fijo / de fija, the inflected variants of the PA-pattern such as a las malas, a las claras, a la continua belong to informal spoken Spanish, but they are still more familiar and clearly colloquial, even if standard accepts some of them as adverbial locutions, e. g., a secas. The variant a la fija is almost unknown in Spain, even if research on rural varieties may hide surprises. The CDH corpus contains 39 instances of a la fija, the first one Romanische Forschungen 131 (2019) from 1853. All of them are regionally confined to the Rio de la Plata and to the period from 1850 to 1920, the only exception being an example from Colombia dating from 1985. The Rioplatense examples all belong to literary texts which aim to reflect rural language, preferentially using dialectal expressions, including the close synonym de siguro (the spelling siguro vs. canonical seguro reflects the rural pronunciation): (49) Ya que va a soltar su número 4, lárguelo a la fija, patroncito, como nosotros, velay ahora se lo hemos atracao a los Rosines de Echagüe (1853, Ar) (50) Deje estar: que luego hemos de acabar con toda esa sabandija, de siguro, y a la fija (1853, Ar) (51) ¿Quiénes son ellos? A la fija, ésta es la primera pregunta que en sus adentros se hará (1853, Ar) (52) -Nada digo, sino que algún »matrero« trujo la miel, y a la fija se llevó algún cordero a la vuelta -contestó el capataz, recomenzando su tarea (1890, Ur) Almost all examples stem from only two authors, from Argentina (Ar) and Uruguay (Ur). The fact that these authors try to document rural communication and the absence of a la fija in other types of text identify this pattern as rural dialect. The regional limitation to the Rio de la Plata could be an objective fact in dialectological terms, but it might also depend on the choices made by authors sharing the same literary movement. Kany (1970: 332) quotes examples from Argentina, Uruguay, Chile, and Colombia. Lavale-Ortiz (in print b) finds many attestations in recent American Spanish varieties (Colombia, Nicaragua, Venezuela, Costa Rica, Uruguay, Argentina). At any rate, the data show that the pattern is productively used in dialects, that is, varieties which tend to conserve old oral traditions. As it is highly improbable that all these regional attestations in America trace back to a common source in the Rio de la Plata region, they are more likely to belong to the same dialectal tradition introduced from Europe during colonization (but see arguments for a differentiated analysis in 6.8). In terms of diachronic reconstruction, this confirms the hypothesis that PA-patterns are part of an old oral tradition (Claim B). At present, Argentinian speakers may not know a la fija, but they spontaneously mention expressions such as es la fija, es una fija, e. g., »Es una fija que en 2018 Argentina crece más del 3 %« (by the economist Miguel Kiguel, < www. infobae.com/economia >, 22. 11. 2017) (see also Lavale-Ortiz in print b). These expressions testify to a fashionable use of fijo in this region.
en fijo but a network of influences possibly including both de fijo and the PN-pattern. The attestations of en fija indicate its affinity to informal oral registers used in rural context. It is noteworthy that only feminine nouns appear in this pattern. Since the absence of masculine nouns is highly unlikely to happen (Why not use *en camino fijo, *en momento fijo ?) if the pattern emerges spontaneously, we should analyze this series as adaptations of spoken en fija to preferences of written discourse, which generally require a noun for adjectival agreement. Consequently, the traditional syntax deletion hypothesis (en forma fija > en fija) cannot be confirmed. The path seems to be en fija (speaking) > en forma fija (writing).

Por fijo
The oldest PA-pattern is por fijo. If we except the conceptual variant »caminar por lo fixo« ›on firm ground‹ (1573-1581, Aguado, Historia de Santa Marta y Nuevo Reino de Granada), all the occurrences of por fijo convey an epistemic meaning: (58) Mediodía era por fijo, Las doce daba el reló (1600, Es) (59) Y en lo que asienta por fijo de que nunca solicitó en el confesonario (1658, Ch) (60) me habia de embarcar poco más de la tercera parte del dinero que se me prometió por fijo en Madrid se me embarcaria en ellos (1668, Es) (61) teniendo todos por fixo aver llegado la hora de descubrirse el Dorado (1723, Oviedo y Baños, Historia de la conquista y población de la provincia de Venezuela) (62) Allí se lee, a propósito de la casa de Hércules: »Sentaremos por fijo que Túbal dió principio a la fábrica de la torre« (1880, Es) The fact that there are only nine attestations, between 1600 and 1880, shows that por fijo is a minor variant that disappears in Modern Spanish. Even if the first attestation (58) conveys an epistemic meaning, the other examples are bridging contexts where fijo still means ›fix, definitive‹. It is striking that the register of use is rather formal, recalling legal or administrative discourse. The more informal example (58) is the one where the epistemic meaning appears most clearly. A possible explanation for using por fijo in formal registers might reflect attempts to (re)latinize the Spanish language. The preposition por comes closest to the instrumental meaning of the Latin ablative, which was abundantly used to build adverbials (e. g., certo, etc.). Consequently, por fijo would have to be analyzed as a Romanische Forschungen 131 (2019) PA-pattern of the educated. This would confirm the hypothesis of coexisting oral and written traditions of using PA-patterns. A more substantiated explanation for this would have to include all »por + ADJ« patterns of Spanish.

A punto fijo
The fact that diachrony has to be situated in a network of horizontal relations between units bearing the same stem is again confirmed by the development of the PN adverbial a punto fijo (including the spelling fixo), which I will outline very briefly (more details in Lavale-Ortiz in print a, b In (66), the author still felt the necessity to reinforce the epistemic concept using seguro ›for sure‹ and the epistemic verbs averiguar ›to find out‹ and saber ›to know‹ in the context, while (67) and (68) mirror a more lexicalized stage. All the 303 occurrences of a punto fijo convey the epistemic adverbial function, with some bridging contexts such as determinar a punto fijo. As in the case of de fijo, most of these occurrences are concentrated in the last decades of the 19th century and the first decades of the 20th, before a punto fijo goes out of use around 1960. De fijo and a punto fijo follow the Romanische Forschungen 131 (2019) same frequency curve. This clearly supports their being related cognitively. The bulk of the occurrences of a punto fijo consist of almost lexicalized collocations with saber, ignorar, and decir.
The discourse function conveying the speaker's certainty about a proposition surfaces only late and slowly. In the first dislocations to the left, a punto fijo conserves its function as a sentence adverbial modifying saber ›to know with certainty‹ (69-74). (70) is a bridging context allowing for both interpretations. The first clear case is (73) in 1918, followed by the still ambiguous example (74) in 1920. The discourse marker function remains marginal over time.
(1871, Es) (72) -A punto fijo, no lo sé yo tampoco -responde el religioso (1908, Ar) (73) El, a punto fijo, sabía que a tales alturas quizá sus padres podrían excusar su conducta (1918, Ve) (74) A punto fijo no sabe qué diferencia puede existir entre la ciudad y la Alhambra (1920, Es) As with de fijo, a punto fijo is used on both sides of the Atlantic Ocean. While the generalization of de fijo is closely related to oral traditions and literary models, a punto fijo originates in administrative and military discourse. Its spread over America seems to follow the same path as the pronoun of address usted, which was part of the official use in the civil and military administration of the Spanish colonial empire. The path consists of progressively adopting the official use in the common local language. The fact that the first American occurrences of a punto fijo, in 1743 and 1762, stem from the viceroyalties of Mexico and Peru, which represented the king, and an isolated instance as far away as the Philippines (1764), in a war report, provide additional support for the relevance of this sociohistorical diachronic path. From the social-psychological point of view, it is interesting to note the speakers' inclination to easily adopt models provided by authorities in the colonial period.

Fijamente
As in English, Romance languages prefer the long adverb in standard, a process that diachronically culminates in hypercorrect forms such as primeramente for primero, paralleled by Engl. firstly and first, and the curi-ous (now archaic) usage of erstlich in German (see details Hummel 2014a, 2017a, 2018). This generates a sponge-effect on previously used Type A and Type C adverbs, starting with the fashionable usage of Type B in the 16th century. It comes as no surprise, then, that Sp. fijamente assumes all the above mentioned functions, especially in elaborated and narrative discourse. However, in the case of fijo, the assumption of a sponge-effect absorbing previously existing discourse functions must be relativized because fijamente is first to assume epistemic functions, only slightly preceded by infrequent por fijo. This might be related to the use of adverbs ending in -iter with discourse functions in Late Latin (see examples in 5.1). These adverbs could have been calqued by -mente.
(75) de luengo a luengo, puestas tablas muy gruesas, tan altas como otro estado, e muy fijamente atadas con muy buenos maderos, por fuera e por dentro (1535, Es) (76) Discurriendo en esto estuvo un rato, sin dar en lo cierto fijamente, y pidiendo recaudo de escribir, la respondió desta suerte (1625, Es) (77) Dícese que en una Ciudad de España, sin saberse qual es fijamente (unos cuentan que en Sevilla, otros que en Valladolid) (1639, Es) (78) El segundo es cuando no se puede comprehender fijamente el intento del toro, porque viniendo a embestir suele con presteza parar (1639, Es) (79) lo que en él se hace es sin haber visto las reses ni saber fijamente dónde están, y habiéndolas de seguir, a necesidad se usa de esta ballestería. (1644, Es) At this stage, we must bear in mind a historical fact that is often methodologically disregarded by national linguistics. One may be tempted to describe the diachrony of Sp. fijo in terms of abstraction and grammaticalization inside the Spanish language, a process that leads to epistemic concepts and functions. This is indeed possible, and the results seem to confirm the grammaticalization process (see Lavale-Ortiz in print a). Fixamente / fijamente would thus appear as a candidate for the first variant including the stem fixo / fijo to be grammaticalized as an epistemic adverbial. Yet appearances are deceiving. If we do not take into account the relevance of written and spoken Latin until the eve of modernity, the results are methodologically biased.
Fixus is abundantly attested in Latin. Interestingly, the motivated use in Lat. fixa mente necessarily entails an epistemic meaning since its first and certainly motivated formation, because mens means ›spirit, attitude, intention‹. From this point of view, it is rather fijamente ›to firmly (hold, Romanische Forschungen 131 (2019) etc.)‹ that should be described as developing from an epistemic to a physical concept, going hand in hand with the progressive semantic bleaching and grammaticalization of mente from ›mind‹ to ›grammatical morpheme marking a manner adverb‹. In the case of fijo, the Latin-Romance contact persisted over centuries in the language of the educated, as shown by the following citation from the end of the 15th century with fixa mente tenere ›to firmly keep in mind‹: (80) generis dicitur ab scrutando: vt fit quando magister scrutatur si discipuli eius fixa mente teneant quod ab eo didicerint. Scrutinium huius significationis fit in ómnibus (1490, Es) The first examples in Spanish (81, 82) date even earlier. Interestingly, the full understanding of their meaning requires a motivated interpretation of mente ›mind‹: (81) nos conuiene e con verdat seer inprenssionados e muy fixa mente tener que el saber de la diuinidat non lo cobro de otra substançia e esseer (1422, Es) (82) mismo benefiçio tiene el saber de las cosas auenideras que, asy sabidas, abraçanse fixa mente con el libre albitrio, e causase, e asy casadas, nasçe el fijo del librar (1422, Es) In both cases, fixa mente refers to a firm mental determination. The examples (80-82) show that motivation was a fact of both Medieval Latin and Medieval Spanish. Moreover, it is perfectly possible that fixa mente was first created in Spanish, as suggested by the examples, and then calqued to the Neo-Latin of the educated. This means that we cannot assume the above suggested bleaching process Lat. fixa mente ›with firm intention‹ > Sp. fixamente ›firmly‹, since fixa mente could have been formed for the first time in Spanish when mente was already grammaticalized. At this stage of grammaticalization, mente could be combined with both meanings, (re)motivated ›firmly believe‹ and bleached ›firmly attach‹. These facts show that the relation between Latin and Spanish was an intimate one. In fact, in the relevant discourse most speakers practiced both. More fine-grained studies on Medieval Latin would be necessary to deepen the analysis of this relation. Three more facts underline the belonging of fixo / fijo to the language of the educated where motivation by the Latin etymology was likely to play a role. In the first place, none of the inherited allomorphs of -mente, -miente or -mientre, is attested, which were highly frequent at that time (e. g., derechamientre). Second, the fact that the orthography fixo is used alongside fijo until the end of the 18th century has to be situated in the tradition of Latinization. Apparently, the memory of Latin was particularly long-living in the case of fixo / fijo, except for the more colloquial units de fijo, a la fija, en fijo, as shown above. The following citation underlines that some speakers insisted on pronouncing it according to the orthography: (83) De esta suerte, si se escribe fixo siguiendo la etimología, uno leerá ficso según la potestad de la x, otro fijo por la anticipada noticia que tiene de la pronunciación española. (Mayans y Siscar, 1723) Finally, one might argue that the low frequency of fixamente / fijamente is another hint for its originally belonging to the language of an intellectual elite. The complete CDH corpus includes 1950 occurrences, but only 26 of them belong to the period before 1700. All this suggests taking a closer look at Latin. Fixa mente is not attested in the TLL, but according to this source the verb figere, its past participle fixum, and the canonical adverb fixē had developed in Latin the whole range of concepts including physical fixing (84), mental fixing (eyes, brain, intention, decision, faith; witness (85,86)), as well as certainty, in variants such as speaker's certainty (87,88), also in collocation with verum (89) or the verb credere (90):