Exploring Optionality: the case of null objects in the medieval Romance languages

:


Introduction
Latin allowed null objects in specific syntactic and pragmatic contexts.With pragmatically conditioned null objects (see (1a)), "the reference of the object is nonfocal and highly topical, and can be readily inferred from the discourse" (Ledgeway, 2012: 73).When syntactically conditioned (see (1b)), the null object occurs in specific syntactic contexts such as coordination, answers to polar questions, and non-finite clauses (cf.Luraghi, 2004;Ledgeway, 2012).
(  ' (Cic. Cat. 1.2, in Luraghi, 1997: 239, glossing mine) By contrast, very few modern Romance languages have been said to have null objects.The most well-known example is that of Portuguese (European and Brazilian), but in these languages the null objects seem to be innovations, being mostly semantically licenced (Pescarini, 2021: 98-100).Standard French and Italian generally do not allow null objects: a modern translation of (1a) or (1b) with a null object in the second clause would be ungrammatical in both languages.(Given that Standard French is a non-null subject language, the null subject is not maintained in the translation.)In (2), the null object is put in the canonical object position because it is the opening sentence of the law: the antecedent is not yet fixed in the discourse, and the overt object, in this case, would be a full DP.The null object in (3) is put in clitic position in analogy with the overt object.
( The senate understands those things, the consul sees (them)' Contrary to their modern counterparts, old Romance languages such as Old French (OF) and Old Tuscan (OT), could have null objects in a similar way to Latin.Example (4a) provides an instance of a null direct object in a finite embedded clause in Old French, and example (4b) evinces a null indirect object in coordination in Old Tuscan.In this article, cases of both null direct and null indirect objects will be considered as their omission occurs in similar contexts and according to similar triggers.the.2PL eyes and cut.PST.3PL∅ la lingua the.SG tongue 'and they blinded him and cut (to him) the tongue' (Nuova Cronica, II, 15, a123, adapted from example (73) in Egerland, 2003. Glossing and translation PP) In the article, I consider the possibility that anaphoric null objects (i.e., null objects with an overt antecedent in the discourse) are optional in Old French and Old Tuscan.In this case, optionality corresponds to the idea that overt and null objects stand in free variation, with both options having the same interpretation.The article is based on a corpus study of 13 th century prose texts from Old French and Old Tuscan, which is later expanded to some 14 th century texts, in order to account for potential diachronic evolutions.
Section 2 defines more in detail the notion of optionality in the current theoretical framework, highlighting some limitations to the study of optionality in medieval languages.Section 3 outlines the state of the art on null objects in Old French and Old Tuscan.Section 4 provides quantitative data for null and overt objects in the corpora by considering some potential restrictions to their distributions.Section 5 then explores the potential for the optionality of objects by studying minimal pairs in the corpora.Sections 6 provides results for the distance of object with regards to their antecedents and section 7 presents a preliminary diachronic study with data from two 14 th century texts.Section 8 concludes the paper.

Optionality in Minimalism
The notion of optionality is a debated topic in Minimalism as it violates the principle of economy in the derivation: in Minimalism, a change in the derivation should yield a change in interpretation.

What is optionality?
Syntactic optionality is the result of two or more alternative forms which "use the same lexical resources" and "express the same meaning" coexisting in a single grammar (Sorace, 2000: 93).In other words, using one form over its alternative in a sentence does not lead to a variation in the interpretation of the sentence itself.
True optionality is a problematic topic in Minimalism, as it seems to violate the economy principles at the heart of the theory.Minimalism posits that any changes made in the derivation must have an effect on the output, which should be incompatible with the notion of a "semantically vacuous optionality" (Richards and Biberauer, 2006).For instance, changing the word order of a sentence must yield a difference in the interpretation of that sentence (e.g., scrambling the word order in German sentences yields a difference in the reading of the sentence, between weak and strong).Nonetheless, true optionality is attested in natural languages.Richards and Biberauer (2006) observe that such an optionality is possible in Afrikaans, where optional verb movement in embedded clauses and optional expletives in impersonal passives can be observed.In the first case, the two options are: 1) keep the verb in the final position of the clause, or 2) move the verb to the second position (exemplified in (5)).
I know that she has often Chopin played 'I know that she has often played Chopin.' According to Richards and Biberauer (2006), (5a) and (5b) do not have a different interpretation (in the absence of any special emphatic pronunciation).In other words, (5a) and (5b) are semantically equivalent, but their word order is different.Their proposed analysis is that these two structures differ in the way they satisfy the EPP requirement: the first structure satisfies it through spec-pied piping while the second structure satisfies the EPP(T) via DP-raising.True optionality in syntax arises when the economy principles underdetermine a derivation: the two options can both satisfy the requirements of the principles, and the choice of an option over the other is then up to the speaker (a similar approach is also adopted by Raynaud, 2018 for embedded wh-questions).Adli (2006) also considers the case of wh-word placement in modern spoken French as a case of optionality.In French, it is possible to have four different word order variants to construct questions, as exemplified in (6).Adli (2006) focuses on the first two variants, (6a) with wh-extraction and (6b) with wh-in situ, and observes that grammatical tests give the different variants the same level of grammaticality and are cognitively processed the same way.He also observes that there are no semantic differences between the wh-in situ and the wh-movement questions: they pattern alike and have the same presupposition requirements.The two options therefore alternate with no difference in their interpretations.
Up until now, I have been talking about syntactic optionality, where the exact same lexical resources are being used, but they occur in a different order, yielding a single interpretation.This type of optionality is thus focused on variations of word orders, rather than variations in the expression of a lexical element (e.g., the object).In the case of null and overt objects, this variation in word order is not relevant, and the type of optionality to be looked at is rather lexical or morphological.The difference between the coexisting options does not lie in a different organisation of the lexical resources (i.e., a different word order) but in the fact that there are two potential realisations of the same element (here the object): overt or null.Still, the variation does not lead to a difference in interpretation.
This type of optionality has also been considered for other phenomena such as the optional verbal agreement with the nominative object in Icelandic (Sigurðsson and Holmberg, 2008).In Icelandic, when the third person object is in the nominative, the verb can either agree with the nominative object, or it can receive the default agreement (3 rd person singular).
(7) (adapted from Ussery, 2009: 1) a. Sumum stelpunum líka peningarnir.Some girls.DAT like.3PLmoney-the.NOM.PL b.Sumum stelpunum líkar peningarnir.some girls.DAT like.3SGmoney-the.NOM.PL 'Some girls like the money.' In a similar way, in early French and Tuscan, two morphological realisations (overt pronoun or ∅) are possible for objects.The question, then, is whether the variation between the two morphological options yields a change in the interpretation of the sentence.

The requirements of optionality
Optionality has two main requirements: 1) that there be two (or more) available options in the grammar, and 2) that these options be semantically vacuous, i.e., yield the same interpretation.In the case of Old French and Old Tuscan, there are two options to encode the object: overtly realizing it in the form of a clitic pronoun, or having a null pronominal object.In the second case, given that the minimal pairs found in the corpus make them alternate with clitics, the assumption in this article is that these null objects are in fact null clitic pronouns 1 .Some other aspects could play a role in the presence or absence of object pronouns in the different texts.One such aspect could be the register or style of the author.It has been argued that some aspects of optionality could be linked to the choice of registers.One example for optionality in Romance languages has to be clitic climbing (CC): in Italian, Spanish and Portuguese, the clitic in an infinitive clause can remain enclitic with the infinitive (as in (8a)) or it can climb to attach to the main verb (as in (8b)).( 8 In Spanish and Portuguese, register plays a role in whether the clitic will climb: in both languages, clitic climbing is more frequent in informal / spoken registers than in formal / written registers (see Davies, 1995 andDe Andrade, 2010) 2 .For both languages, the syntactic triggers of clitic climbing are still present in both registers, but the strength of these triggers varies across the registers, yielding differences in the frequency of CC.
In the case of the early Romance languages in study, it could be argued that overt and null objects are linked to a specific register (e.g., formal vs. informal) which could be represented by the fact that null objects occur either in narration or in direct speech.In this case, direct speech would represent a more informal register while narration would represent a more formal language.In the corpora, the question of a register-based division between null and overt objects does not seem to hold: null objects can be observed in both direct speech contexts and narration contexts in all the texts studied.
The second aspect which could influence the presence of two options for the encoding of the object is information structure.It could be the case that the presence, or absence, of the object in a sentence yields a difference in information structure, with some element of the sentence being marked for emphasis.In the Romance languages, this is the case for subjects: in languages like Italian, Spanish, and Portuguese, the nonexpression of the subject pronoun in a sentence is generally the default, unemphatic 1 A weak pronoun analysis could also be considered for null objects (as in Roberts (2014)), but such a consideration is beyond the scope of the present research.Thanks to one anonymous reviewer for pointing this out.option.When the subject pronoun is overtly realized, the emphasis is put on that pronoun, yielding a difference in interpretation between sentences with a null subject and sentences with an overt subject pronoun.In a similar way, it could be the case that null objects in Old French and Old Tuscan are the default option, and that the overt objects are emphatic.The empirical data, however, does not support this hypothesis, first because overt objects are more frequent in the texts than null objects, and second because of the nature of the overt objects themselves: as mentioned previously, the overt objects which coexist with null objects are clitics.By nature, clitics cannot be coordinated nor can they be focalized, meaning that they cannot bear the emphasis of the object alone (Pescarini, 2021: 11-12).
A second hypothesis linked to a variation in information structure could be that in the null object variant, emphasis could be put on the action denoted by the verb.This hypothesis might be tricky to argue for or against but some characteristics of minimal pairs can help.The presence in the corpus of minimal pairs occurring in the same contexts due to a parallelism in the structure (as shown in section 5.1.)seems to argue against such a hypothesis.If a null and an overt object can occur in the exact same syntactic contexts, with the null and the overt objects being as topical as each other and with the same relation to their antecedent and their referent, then it is unlikely that the null objects lead to a difference in interpretation compared to the overt objects.

Limitations of this study
The main limitation to considering optionality in early Romance languages is the reliance on corpus data.Generally, studies on optionality rely on grammatical judgement tasks and interviews with native speakers.The tasks and judgements revolve around minimal pairs and the way their interpretation might (or might not) differ.This is unfortunately impossible with historical versions of languages.Strict minimal pairs, i.e., pairs of sentences using the same lexical resources and syntactic structures, are quite rare in both corpora used for this study, but they enable a first foray into optionality.
The current article will explore the idea of optionality with relation to objects in two 13 th century Romance varieties: Old French and Old Tuscan.As null objects are quite rare compared to overt object clitics (see 3.1.),it will be interesting to consider the question of optionality through the lens of contexts in which null objects are attested.Indeed, even though they are quite rare, they appear frequently in certain contexts (e.g., in coordination), where they seem to coexist with overt objects until the 14 th century.The main research question is then: is there a case for optionality concerning the encoding of the pronominal object in Old French and Old Tuscan?

Null objects in the early Romance languages
The possibility of null objects has long been outlined in grammars of early Romance languages, with a reference to null objects already present in Tobler's grammar of Old French (Tobler, 1886: 406).However, there has not been extensive work carried out on the topic for either Old French or Old Tuscan.This section highlights the main works which have fully focused on null objects in these languages.

Null objects in Old French
Null objects in Old French were already observed in Tobler (1886: 406) with prepositional infinitives.Highlighting the fact that null objects are frequent in this context, he argues that the old language was closer to Latin than to Modern French. 3A century later, Jensen's (1990) grammar mentions that the non-repetition of an object pronoun was frequent in coordination structures.Jensen, quoting Tobler (1886), argues that 'generally speaking, the medieval language is averse to the use of pronouns that add nothing to the clarity of the sentence' (Jensen, 1990: 156).Even though mentions of null objects go back to the 19 th century, the literature on the topic in Old French is not as extensive as one might think.There are three main articles dedicated to studying the distribution of null objects in Old French (Arteaga, 1998;Donaldson, 2013;Schøsler, 1999).The phenomenon is also mentioned to varying degrees in grammars of the language (a.o.Jensen, 1990; Grande Grammaire Historique du Français (GGHF), 2020).Donaldson (2013) "Oder es kann das Pronomen beim zweiten Infinitiv ganz ausbleiben, wenn das Objekt sich aus dem Zusammenhang leicht ergibt […]; so auch mit einer Präposition: tendit ses deus mains por juer a la corone e por prendre, Serm.Poit.193.In solcher Weise das pronominale Objekt unausgesprochen zu lassen, ist ja auch sonst die alte Sprache sehr geneigt, die hierin der lateinischen Knappheit näher steht als der neufranzösischen Pedanterie." (Tobler, 1886: 405-406) [Or the pronoun can be omitted completely from the second infinitive if the object is easily recognisable from the context [...]; so also with a preposition […] To leave the pronominal object unpronounced in such a way is, after all, a tendency of the old language, which in this respect is closer to Latin brevity than to modern French pedantry.(translation DeepL)] iii. adjunct clauses: if the object of a non-finite clause is also the argument (pronominal or nominal) of the main clause the non-finite clause is adjoined to, the object in the non-finite clause can be omitted In this article, I focus on anaphoric null objects in Old French, meaning objects which have an explicit antecedent in the discourse.Contexts (iv.) to (vi.) will therefore not be considered, as they do not require an explicit antecedent and are generally retrieved from general world knowledge.
The focus is instead on five contexts: coordination, écrasement, adjunct clauses and left-dislocation without resumption.Additionally, null object can be found in main clauses, when the referent of the object has been fixed in a preceding clause: Old French (Eust., 30) Que qu' elle parla a lui, ele esgarda viselment, while that she spoke to him she watched visually si aperçut un sieng que ses mariz soloit avoir thus saw a sign that her husband used.tohave 'While she was speaking to him, she looked (at him), and saw a sign that her husband used to have' In order to have enough data on a phenomenon that is quite infrequent, I focus on five contexts which can trigger null objects (as well as overt objects).In all these contexts, having a null object is generally not the only option, and examples with overt object clitics can be found for most of these contexts in the data (with the exception of écrasement structures, which will be considered in 5.3.).The aim of the article is to make a case for the potential optionality of overt and null objects in the languages, and not to analyse the underlying phenomena that could trigger null objects 4 .The fact that all these contexts display both null and overt objects seems to point towards optionality.Schøsler (1999) observes that, in Old and Middle French, there is a very strong tendency to overtly realize the object instead of omitting it.Even in contexts favouring null objects, the objects tend to be overtly realised.Nevertheless, null and overt objects occur in similar contexts, in accordance with optionality.In addition, she highlights the fact that [+human] objects tend to not be null.This second observation does not hold for the corpus, as will be shown in section 4.2.

Null objects in Old Tuscan
In Old Tuscan, objects were usually overtly realized in a sentence, and, if the object was null, it usually received a generic interpretation (Salvi and Renzi, 2010: 128)."Only exceptionally can an unexpressed direct object with a finite verb have a definite interpretation […] The non-expression of a definite direct object is on the other hand more frequent with non-finite verb-forms.' (Salvi and Renzi, 2010: 128-129, translation mine) 5 .In other words, anaphoric null objects are not that frequent in Old Tuscan, but they are attested in specific contexts such as in non-finite clauses.A 4 As was rightly pointed out by one anonymous reviewer, these contexts could also be analysed as resulting from different syntactic phenomena: ellipsis for coordination, Person-Case-Constraint for écrasement, parasitic gaps for adjunct clauses, and an empty object position due to movement in LD.
context of omission involving finite verbs mentioned in the Grammatica dell 'Italiano Antico (GIA, Salvi and Renzi, 2010: 463)  According to the existing literature on the topic (see Luraghi, 1998, andEgerland, 2003), coordination is the most productive context of object drop, especially in the 13 th century (as these mostly involve finite verbs, this observation seems to go against Salvi and Renzi, 2010).As a matter of fact, most of the examples of null objects found in the corpus occur in such structures.However, null objects are not restricted only to coordination in Old Tuscan: Egerland (2003) also provides examples of null objects in embedded clauses, as well as adjunct non-finite clauses (as mentioned in the GIA).Similarly, non-anaphoric null objects such as arbitrary human objects or null cognate objects are also attested.
One context of omission which is observed in Old French but not in Old Tuscan is that of écrasement.Old Tuscan does not systematically reduce 3 rd person object clusters to one clitic but allows both clitics to co-occur.This does not mean that objects are never null in these contexts, but they do not follow the rule observed in Old French.This will be a point of comparison which will be considered in section 5.3.
These facts prompt Salvi and Renzi to write that "thus one has to admit that in Old Italian, contrary to modern Italian, the expression of the definite direct object was not obligatory" (Salvi and Renzi, 2010: 129, translation and emphasis PP) 6 .The aim of this article is to test this assumption.

Methodology
I collected data from two corpora, one for Old French and one for Old Tuscan.For both corpora, the texts chosen were narrative prose texts from the 13 th century.Prose was chosen over verse to counter a potential bias in the distribution of null objects due 6 "cosicché bisogna ammettere che in italiano antico, differentemente che in italiano moderno, l'espressione dell'oggetto diretto definite non fosse obbligatoria."(Salvi and Renzi, 2010: 129) to the internal rules of verse (and the metric).Two potential examples of verseconditioned null objects are provided in ( 12) and ( 13 In both cases, the meters of the verse influence the absence (or presence) of the object.In ( 12), the null object is in the first conjunct, which never happens in the prose corpus.In (13), the 3 rd person clitic cluster is fully realised in the text, a case which is not observed in any of the French prose texts studied.This difference in behaviour can be linked to verse: both texts are written in octosyllables with a cesura after the fourth syllable.In (12), having the overt object in the first conjunct would add a syllable to the verse, making it irregular.In (13), the direct object clitic is necessary in order to respect the meter (eight syllables).
Similarly, preference was given to narrative texts over other types of texts in the corpus, such as legal texts, in an attempt to avoid potential biases due to the influence of Latin and formulaic turns of phrases.Overall, the Old French corpus is about 100,000 words long and yielded 175 examples of null objects, while the Old Tuscan corpus is about 47,000 words long and yielded 61 instances of null objects.For both corpora, the texts were manually parsed and annotated for null objects 7 .The metadata for the two corpora is summarised in the following table.

7
For both corpora, full texts were analysed.In the case of Old French, the texts were obtained from the Base de Français Médiéval (database of medieval French).For Old Tuscan, pre-existing corpora such as OVI and TLIO did not provide access to the full texts so, instead, the data were gathered from different editions of the texts.

Quantitative findings: The profile of objects
Syntactically, null objects and overt objects seem to be able to occur in the same contexts.Nonetheless, it is possible that only objects with a specific profile can be null (e.g. in Brazilian Portuguese, null objects seem bound to an animacy restriction: inanimate objects are almost always null whereas animate objects tend to be overtly realized (Schwenter, 2006)). 8 One anonymous reviewer suggested including the number of object clitics in each text, in order to provide quantitative data to compare the distribution of null and overt objects in both varieties.While this is possible for French, as the corpus is syntactically annotated for personal pronouns, the same is not the case for the Old Tuscan corpus.A sample of five pages for both text yielded: 134 object clitics for 7 null objects in Il Novellino and 78 clitics for 4 null objects in Il libro dei Vizî e delle Virtudi.
Annotating the data according to their internal syntactic and semantic properties yielded some interesting overlaps for null objects.In Figure 1, the semantic properties taken into account are [+/-animate], [+/-specific], [+/-human] and any combination thereof.In Figure 2, the syntactic properties taken into account are person (1 st , 2 nd , 3 rd ), number (singular / plural), and gender (masculine / feminine) and any combination of the three.(Absolute numbers given in addition to percentages in both figures.) One observation which can be made from the data is that null objects in both Old French and Old Tuscan are overwhelmingly [+human], which is in stark contrast with the previous literature's observation that, in OF at least, null objects tend to be nonhuman (see Schøsler, 1999). 9 The figures seem to show very strong tendencies in the profile of null objects which may explain why some objects are null and not others.From figure 1 and 2, it appears that 3 rd person singular objects which are human and specific are more easily null than any other type of object.
However, as can be seen in figures 3 and 4, comparing the profile of null objects to that of overt objects in the same texts does not showcase huge differences in distribution.Both overt and null objects in the texts tend to be 3 rd person masculine, human and specific.Due to the length of the texts and the number of object clitics present in each text, it was not possible to annotate all of the object clitics for the different syntactic and semantic properties.Instead, a sample of overt object clitics were randomly selected from each text of the corpus, mirroring the number of objects in the different texts.For example, for the Queste del Saint Graal, given that 103 null 9 One anonymous reviewer also suggested that it was surprising that null forms were predominantly animate, as stronger forms are preferably associated with animate objects (see Moignet, 1970), but may be due to a more general preference for [+animate] and [+human] objects in the texts.Annotating all the object clitics in the texts in relation to animacy would be too much for the present article, however, a sample of 10 pages from Aucassin et Nicolette seems to show that 3 rd person overt objects in the corpora are generally animate (80 animate (mostly human) objects vs 24 inanimate clitics).The topic of the texts (mostly focused on characters rather than story) may thus influence the animacy of the objects.
OF dropped objects OT dropped objects OF dropped objects OT dropped objects Exploring Optionality: Null objects in medieval Romance Isogloss 2024, 10(3)/4 15 objects were annotated, 103 object clitics were randomly selected from the excerpt studied and annotated for the semantic and syntactic properties.(Absolute numbers given in addition to percentages in both figures.) When looking at the two graphs, we can see minute differences in the distribution of the semantic and syntactic properties of overt and null objects in both Old French and Old Tuscan.For example, while some overt objects in Old Tuscan and Old French are 2 nd person plural feminine, none of the null objects in the texts have these properties.
Given the amount of data we are dealing with, it is not necessarily the case that such a difference is relevant.In fact, post hoc tests (such as the equivalence test from the Parameter package in R (Lüdecke et al., 2020)) 10 show good evidence to accept the null hypothesis for both the semantic and syntactic properties of the objects. 10 Results of the posthoc texts for each set are provided in an Appendix at the end of the article.Thus, while at first glance null objects may appear to be restricted to a very specific type of objects, further analysis shows that it is not the case: both null and overt objects in the corpora follow the same tendencies, which reinforces the idea that they are two lexical options to express the same element.

Finding minimal pairs to account for optionality
Optionality has two main requirements: 1) two or more available options in the grammar and 2) the options have the same interpretation.Studies on optionality generally rely on minimal pairs to account for it, as they show the different options available in a language in order to encode the same meaning.This section will provide minimal pairs for four contexts: coordination, prepositional infinitives, adjuncts clauses, and clitic clusters.Null objects in these contexts are generally no longer grammatical in modern French or modern Italian, with some exceptions.One caveat mentioned in section 2.3. is that strict minimal pairs are not very frequent in the corpus.In order to have enough qualitative data to explore the notion of optionality with regards to null and overt objects, a more lenient definition of minimal pair is sometimes adopted: rather than focusing on lexical minimal pairs, the focus is on structural pairs (e.g., two adjunct participles with an object referring back to the object in the main clause, or two coordination structures of the type: V DP-object & (Object-pronoun) V), with one part of the pair having an overt object and the other a null object.

Minimal pairs in coordination
Coordination is the context in which null objects are found most frequently in both OF and OT.
Let's consider first the possibility to have two options (an overt and a null object) in coordination by looking at minimal pairs.Only Old Tuscan evinces strict minimal pairs in this context, with an example given in ( 14): the lexical resources and the structure are, at least superficially, the same.The only difference is that the object in the second conjunct in (14b) is null.( 14 In both cases, the pairs occur very close together in the text and seem to be obtained from a parallelism between structures.The examples in ( 15) are juxtaposed, and therefore occur in immediate vicinity, while (14a) and (14b) occur further apart (in two successive paragraphs), but in a very similar context: both conjunctions are preceded by the clause come nella detta gente / come in quella gente 'like in the said people / like in these people', which indicates a parallelism of structures.Interestingly, in both cases, the option with the overt object occurs first in the text, followed by the one with the null object.It therefore seems that, once the referent of the object is fixed and the expression is "acquired" by the reader, it may become possible to drop it in a similar expression in the immediate vicinity.
A more lenient definition of minimal pairs can yield more examples of similar structures yielding an overt and a null object.Example ( 16) provides an example of an almost minimal pair in Old French with a full DP-antecedent being either resumed in the second conjunct with an object pronoun (16a) or with a null object (16b).( 16b 17b) are taken to be minimal pairs as the coordinated verbs are the same and occur with a similar structure, even though the relationship with the antecedent is different in the two sentences.
All the examples given until now for both Old French and Old Tuscan show that the option competing with null objects is overt clitic pronouns.This is not a very surprising observation as this article deals with anaphoric null objects, which require their antecedents to be fixed in the surrounding discourse.
I now turn to the question of the interpretation of the two options.As mentioned previously, a difference in interpretation would argue against optionality.The strongest minimal pair available for this is the one in example ( 14).Considering the larger context in which it occurs could give us clues as to whether the interpretation changes between overt and null objects.
( In the case of ( 18), including the larger context makes it quite clear that the two coordinations are linked in a parallelism of structure (between the first battle and the second battle which involved exactly same participants).In both examples, the clitics and the null object have the same referent (the virtues), the clitic in the second conjunct in (18a) cannot, by definition, be focalized.Similarly, emphasis in both examples is on come nella detta gente / come in quella gente rather than on the coordination structure (with and without null object).In terms of topicality, the clitic in the second conjunct of (18a) is as topical as the null object in (18b), and the clitic and null object in (18b) are also just as topical as each other.There thus seems to be no difference in reference selection or information-structure between the two examples which could explain the distribution of the null object compared to overt objects.
In coordination, it is therefore possible for objects to be either overtly realised or null, without either option being marked for information-structure, making null/overt objects optional in this context.

Minimal pairs in non-coordination contexts
Null objects are not only restricted to coordination and can be found in other contexts.This section will focus on two contexts in which null objects are encountered: finite embedded-main clause pairs and non-finite adjunct clauses (including prepositional infinitives and participial clauses), as they are two contexts in which minimal pairs (according to their structural properties) could be found in the corpora.
Null objects are attested in structures such as (19), where the antecedent is in an embedded clause in the left periphery of the sentence and the null object is in the main clause.Examples (19a) and (19b) occur at very different points of the text (as noted by the distance of 157 paragraphs in the text), and in different contexts which do not include the same referents.The minimal pairs are therefore not instances of parallelism in structures, contrary to example (18).
Another context which has been highlighted as a productive context of null objects in both Old French and Old Tuscan is that of adjunct non-finite clauses: these include participial clauses (as in ( 21)) as well as prepositional infinitives (exemplified in ( 20) and ( 22)).Once again, strict minimal pairs being difficult to come across in the data, the focus was on a minimal pair in terms of structure rather than in lexical resources.
(20) a.Old Tuscan (Nov.,20)One anonymous reviewer pointed out that (20a) was still natural in modern Romance languages without null objects.In the case of French and Italian, it may also be the case.As rightly mentioned by one anonymous reviewer.In modern French and Italian as well, such an example would be very natural in certain contexts.'and punishing (him), he lashes and torments him thus '. b. Old Tuscan (Nov.,39)22c), where the object of a prepositional infinitive appears to be a strong pronoun (lui vs le).It would then seem that, with prepositional infinitives, a third option might be possible, in addition to null 15 One anonymous reviewer rightfully pointed out that the clitic in (21b) may not be optional as it is not coreferent with an argument in the main clause: it is therefore in a context of topic-shift, or topic-discontinuity (Frascarelli, 2007;Frascarelli & Hinterhölzl (2007)).While topic continuity seems to be unproblematic for optionality, topic shift requires further study.
objects and clitics: using a strong pronoun.Olivier (2022) argues that this third option is due to a transition period between enclisis (as in (22a)) and proclisis, which is not yet fixed as the clitic place in prepositional infinitives.The transition period is marked by a frequent use of these apparently strong pronouns in preverbal position.These are only apparently strong because they do not behave as freely as strong pronouns: e.g.these pronouns never occur postverbally, and nothing can occur between the preposition and the verb (Olivier, 2022: 218).Old Tuscan does not display this third option with the same restrictions, and instead uses truly strong pronouns.Olivier (2022: 219) argues for 'weak' pronouns in the sense of Cardinaletti and Starke (1999), which may have been used emphatically, due to their apparently strong morphology.This would mean that this third option leads to a difference in interpretation and therefore does not fit into our notion of optionality.If these pronouns are not emphatic, it would provide a third option for the encoding of objects in this particular context.Thus, while in Old Tuscan, null objects compete only with overt object clitics, in Old French, null objects mostly competed with object clitics but, in one specific context, may have also competed with weak pronouns.
With prepositional infinitives and adjuncts clauses as well, there are at least two options to encode the object: null objects and object clitics.Both options occur in similar structures, and do not trigger a difference of interpretation in the sentences in which they occur.

Minimal pairs with clitic clusters in Old French
The last point in this section concerns the context of écrasement of 3 rd person object clitic clusters in Old French (mostly appearing with ditransitive verbs).As mentioned in 3.1.,écrasement is frequently observed in Old French, independently of the syntactic contexts in which it occurs.The direct object in these reduced structures is still syntactically active and can trigger participle agreement, as exemplified in (23).
(23) Old French (Graal,§113,179d) et tolue ∅ li eust il and taken.F.SG him.DAT had.3SG he 'and he would have taken (her) from him' In (23), the participle is marked for feminine but the overt arguments (subject and indirect object) are both masculine, so a feminine element which could trigger the agreement is clearly missing here: the direct object.
In Old Tuscan, such a reduction is not attested: 3 rd person object clitic clusters are generally overtly realised, with only some instances of an object pronoun being null (cf. (24c) where, in this case, the missing object is not the direct but the indirect object, which is unexpected in a context of écrasement).However, these null objects seem to be more due to the overall context of occurrence rather than the specificities of the clitic cluster itself.
(25) him' (Donaldson, 2013: 69, adapted  The text this example comes from is in verse which, as I have highlighted in section 4.1.,can bias the distribution of null objects.In (25), the realisation of the full cluster may not be due to the cluster itself but to the requirements of the verse it occurs in.
In all the Old French prose texts studied, a clitic cluster of 3 rd person pronouns in direct and indirect object positions is never realized as two pronouns, but always with only the indirect object overtly realised.This seems to show that, compared to the other contexts of omission, écrasement may not be a context where null objects are optional, but instead, where it is required.It has been argued that écrasement is a process of phonetic reduction of the pronouns rather than a syntactic context for null objects (GGHF, 2020): instead of a null element in the syntax, the direct object is simply not realised at PF. This, in turn, would mean that the non-expression of the object in these contexts is different from the null objects observed in other contexts highlighted in sections 5.1.and 5.2.

The distance from the antecedents
An additional aspect which could be relevant as a trigger for null or overt objects is their distance with relation to their antecedent.It could be the case that null objects occur closer to their antecedents than overt objects.
Figures 5 and 6 represent the distance of null and overt objects from their antecedents in Old French (figure 5) and Old Tuscan (figure 6).For both OF and OT, the overt objects are the overt clitics previously annotated for semantic and syntactic properties.The distance is calculated in terms of number of words between the closest realisation of an antecedent and the objects.Here, closest realisation meant that the antecedent does not have to be a full DP that was introduced for the first time, but it could also be the closest pronoun in a pronoun chain, a possessive article referring back to the antecedent, etc.In example (26), the antecedent of the first pronoun is the DP la cetera 'the zither'.For the second pronoun, even though the clitic still refers to 'the zither', the antecedent is taken to be la 'it' in the second conjunct rather than the full DP in the first, as it is the closest element which makes the referent more easily recoverable in the discourse.Both figures show that, on average, overt objects occur further away from their antecedents, compared to null objects.This difference is however minute, as this is a difference of 2 words between null and overt objects.In Old French, null objects are on average 7 words away from their closest antecedent while overt objects are on average 9 words away.Similarly, in Old Tuscan, null objects are on average 5 words away from their closest antecedent while overt objects are 7 words away.In addition, null objects are not necessarily extremely close to their antecedents, with one example of a null object in Old French occurring almost 50 words away from its antecedent.Similarly, overt objects, like null objects, can also occur 1 or 2 words away from their antecedents.It is not because an object occurs close to its antecedent that it will automatically be null.The distance from the antecedent may nonetheless influence the distribution of null objects: the closer to its antecedent an object is, the more likely it is to be null.

Preliminary diachronic outlook
As has been shown in section 5., in all the contexts in which they occur, null objects seem to compete with overt object clitics, with no differences in interpretation.The following section provides a preliminary foray into diachrony with a study of 14 th  Starting with Old French, Clari's Conquête de Constantinople yielded 56 null objects, out of a 34,000-word text.To compare, in the 13 th century corpus of Old French, Aucassin et Nicolette, Vie de Saint Benoit, Vie de Saint Eustache (all together 34,500 words) yielded 72 instances of null objects.There thus seems to be a decrease in the number of null objects in the 14 th century.Out of these 56 examples, 17 occur in contexts of écrasement.By comparison, only 3 examples occur in écrasement contexts in the three-text-corpus.Adding the Queste del Saint Graal, 26 examples out of 103 occur in écrasement.Putting it in proportions, 30% of the null objects in Constantinople occur in clitic clusters, while only 16% do in the 13 th century corpus.Similarly, the null objects occur more frequently in coordination contexts in Clari's text than in the 13 th century texts studied, with 37.5% of the examples in the 14 th century compared to 25% in the comparable 13 th century subcorpus.This preliminary study therefore points towards a gradual loss of productivity in null objects in the 14 th century.
For Old Tuscan, Boccaccio's Decameron provides some very contrasting data: in an excerpt of 24,500 words, only 7 examples of null objects were found, compared to 26 examples in the shorter excerpt from il Novellino (16,500 words).Nonetheless, the tendency for null objects to occur mostly in coordination, observed in the 13 th century texts, is still present in the Decameron, with 5 out of 7 instances occurring in coordination.
This apparent reduction in productivity in both Old French and Old Tuscan diachrony seems to point towards two things: first, null objects are being lost in both languages the closer they get to their modern counterparts, second, coordination is the most persistent context of object drop in both languages, potentially indicating that the contexts inherited from Latin where null objects are the rule are the most persistent.
This decrease in null objects could be due to the development of object clitics in OF and OT.Contrary to Latin which allowed null objects but did not have a clitic system, OF and OT have clitics.As mentioned previously, object clitics are a lot more frequent in the texts than null objects: in terms of acquisition, an acquirer would more frequently be confronted with an overt clitic in the object position than a null object, which would lead to the generalisation of the use of clitics in all contexts (including the ones yielding null objects) 16 .As a result, the optionality with regards to the encoding of the object disappears and only overt clitics remain as an option to encode the object.The OF and OT periods would thus be transition periods between a system allowing null objects and without object clitics (Latin) and a system disallowing null objects and with object clitics (modern Romance).During this transition period, both null objects and overt clitics coexist, but the contexts in which null objects appear get gradually restricted.The last contexts where null objects disappear are the contexts in which they occurred most frequently (e.g., in coordination).

Conclusions
This article explored the idea that (null) objects in medieval Romance could be considered optional: they coexisted in most syntactic contexts with overt object clitics and the choice of an overt or a null object did not lead to a change in interpretation.The only context in which null objects seem to be required is that of OF écrasement, which may show that the underlying trigger for this context is different from the other contexts of omission.The article also showed that the distribution of null objects compared to overt objects was not restricted in terms of semantic or syntactic properties (overt and null objects generally have the same profile).The only difference between the two types of objects is that null objects occur closer to their antecedents in both OF and OT.The closer an object is to its antecedent, the more likely it is to be null.The article then provided some preliminary diachronic observations based on texts from the 14 th century.Diachronically, it seems that null objects are less frequent in later centuries, and mostly occur in contexts of coordination (and écrasement for Old French).Further research is of course required in order to see whether these tendencies truly reflect the situation in diachrony.
Even though optionality is a debated topic in Generative Grammar and Minimalism, the article provided some interesting historical data which point to the presence of a morphological optionality in the encoding of objects in both OF and OT.True minimal pairs are rare in the corpus, nonetheless some examples can be found and they point towards a coexistence of null and overt objects in different syntactic contexts.

16
One anonymous reviewer also suggested that the loss of null objects may be part of a general typological shift whereby word order becomes more fixed in Romance and therefore 'clashes' with null objects, or that acquirers shifted towards the most used form (object clitics) in all environments, with acquisition drifting towards the most salient option, instead of maintaining the two options.

2
Thanks to one anonymous reviewer for pointing out the facts of clitic climbing.
them.ACC gave 'the court jester gave them to him'

Figure 1 .
Figure 1.Semantic properties of null Figure 2. Syntactic properties of null objects in OF (N=175) and OT (N=61) objects in OF (N=175) and OT (N=61)

(
14 the zither, broke it and threw it away'

Figure 5 .
Figure 5. Distance from antecedent in Figure 6.Distance from antecedent in OF OT century texts.The two 14 th century texts studied are La Conquête de Constantinople, by Robert de Clari (based on the manuscript date) and Il Decameron by Boccaccio.
(lit."crushing")in an object clitic cluster, if both the direct object and the indirect object are 3 rd person, only the indirect object is overtly realised.This process is not limited to a specific Arteaga's (1998)eaga's (1998)article, highlights 7 contexts of omission: i. coordination: in a coordination, if the object is the same for both conjoined verbs, the object in the second conjunct can be omitted.In Old French, this is possible even if the verbs assign different cases to their objects.(a) (Vie de Saint Eustache, XIII, 16) were bringing her back to the city to burn (her)' iv.null cognate objects: some transitive verbs, such as eat, drink, speak, can be used intransitively, with an understood object which does not have an antecedent in the discourse.In those cases, the null cognate object is 'something edible', 'something drinkable' and 'something speakable'.
v. formulaic or conventional expressions: in formulaic or conventional expressions, the object can be omitted even without an explicit antecedent.In the example, the understood object is "'Your land, who will defend (it)?' is that of coordination.
).And then he goes back to take his weapons, / that he found at the head of a table / where they had been brought to him.'

Table 1 .
Metadata for the Old French and Old Tuscan corpora.Already from this table, we can observe that null objects are not a very frequent phenomenon, with few examples overall in the different texts compared to the overt object clitics.Overt objects are generally the preferred option.
No such example of a strict minimal pair is observed in Old French.The closest example of such a pair is exemplified in (15), but the second alternative is negated, which may influence the possibility to omit the object 11 .
In (16), the two parts of the pair occur in completely different paragraphs of the text, and therefore are not cases of structural parallelism.Many coordination structures occur between the two examples, with and without null objects in the second conjunct.
You should also bear in mind the second battle, which began again from us to you in the descendants of Noah, like in these people we vanquished you and banished (you)' 12One caveat to underline for this example: the presence of the impersonal pronoun in (17a) may influence the choice of null/overt object.
boughtt a big baby goat and roasted it.And roasting him, he drew out the bowels and ate them.'In(21a)and(21c), the object in the gerund is either overt or null, but both occur in a case of topic continuity: both the null object in (21a) and the object clitic in (21c) refer to the object of the verb in the clause preceding the adjunct clause.

Table 2 .
Distribution of null objects (NO) in the OF and OT 14 th century texts, compared to the 13 th century corpora (absolute numbers)