Hirschbühler 1995 «Null Subjects in V1 Embedded Clauses in Philippe de Vigneulles Cent Nouvelles Nouvelles», in Adrian Battye & Ian Roberts (eds.) Clause Structure and Language

Null Subjects in V1 Embedded Clauses in Philippe de Vigneulles' Cent Nouvelles Nouvelles

Paul Hirschbühler

Université d'Ottawa

"C'est un système bien dangereux, monsieur Fred, bien dangereux, que celui qui consiste à partir de l'idée qu'on se fait de l'assassin pour arriver aux preuves dont on a besoin!" (Rouletabille)

O. Introduction The focus of the present study concerns the licensing of null subjects in embedded V1 constructions in a late Middle French text, theCent Nouvelles Nouvelles (from now on, the CNNV) by Philippes de Vigneulles'.1 More precisely, our goal is to characterize as precisely as possible the conditions under which these null subjects are found, hoping that some of the ideas discussed will find their way in a more principled account than the one that is offered here. One point that should be clear from the start is that it is not claimed that all the variables that appear to be required in characterizing the data examined here are necessarily relevant for other Middle French texts. Some of the facts are unexpected given previous literature. It may therefore be tempting to consider them as accidental patterns rather than considering them as significant. The second position will be adopted here, in the hope that close study of other texts will reveal similar patterns. This paper is organized as follows: we first very briefly review the main points of the grammar of Middle French with respect to main clauses that are relevant for the discussion of embedded clauses; then we present the main results of our analysis of null subjects in V1 embedded clauses; finally we turn to the data and discuss it in detail. 1. Main points of the analysis 1.1. (S)VO in main clauses In Middle French, assertive root clauses are of two main types. V2 sentences, which are analyzed as CPs, and SVO clauses which, following Vance (1988), we take to be IPs 1 Philippe the Vigneulles is a draper from Metz. He wrote his CNNs between 1505 and 1515. He has been described as a semi-literary author.

2

or, in a more ramified view of clausal structure, AGRPs.2 Among these SVO root clauses, there is a large enough number of examples where the subject is null to support the idea that for all persons, AGR, perhaps in combination with Tense, can license both the position and the content of pro subject to its left, i.e. under Spec-Head agreement.3 Examples are given in (1) below. Generally, these sentences do not strictly start with the verb (except for 'advint que' i.e. '[it] happened that']), but the initial element does not normally induce V2 effects in Vigneulles; this initial element may be a coordinating conjunction, the negative element 'ne' or phrases that are analyzed as adjoined to IP (see Vance (1988, sections 4.1.3, 4.2.1) and Roberts (to appear, section 2.3.1) for detailed arguments in favor of adjunction to IP in cases of that sort). (1) a "Tu jeuneras demain et aprés et aprez. Et ø1 te ordonne que ainsi se fasse ... 002042 'You will fast tomorrow and after and after. And [I] order you that [it] so be

done…4 b Et aultre responce n'en peust avoir ladicte femme. ø31 Ne demeura guiere aprés qu'elle fust morte. ø1 Ne sçay s'elle en alla … 062039 And another answer could not receive the mentioned woman. [It] didn't last long

before she was dead. [I] don't know if she went … c … pour tant que l'une des fois tu met la cornette d'ung costez, … aucunes fois tu

la fais aller devant, aucunes fois ø2 la fais aller derrier, et ø3 va et vient de tous

costez ... 044043 … as sometimes you put [your hat's] horn on one side, … sometimes you put it in

front, sometimes [you] make it go behind, and [it] goes in all directions. 2 For the sake of clarity, we will tend to use IP to refer to the highest maximal projection immediately dominated by CP. 3 There is no example of null subjects of the 2nd person singular in a V1 construction in main clauses, but we take this not as the reflection of a prohibition against that construction with a 2nd person sing., but rather as being related to the fact that there are only a handful examples (i.e. 12) of null subjects of the 2nd person sing. in assertive main clauses in this text. 4 The interlinear translations are literal translation for the most relevant parts, and more idiomatic English glosses for the rest. The first three numbers after an example correspond to the number of the short story, the next three to the line in that story. So, '002042' means story number 2, line 42. A null subject will be indicated by 'ø'; when present, a numeric subscript indicates the person.

3

d ..Puis, cela faict, ø3 print de l'ancre … Cela faict, ø3 s'en retourna avec les autres, et quant temps fut de partir, ø6 prindrent congié et ø6 s'en allèrent … 001019 Then, this done, [he] took some ink.… This done, [he] went back with the others,

and when time was to leave, [they] took their leave and left … e ø31 Advint un temps que l'on se doubtoit à Mets pour aucuns gens d'armes…

061018 [It] came a time when people were afraid in Mets because of soldiers … f Je la te venderés une somme qui sera dicte, et ø4 metterons les vins à bon pris et ainsi ø4 buverons de la marchandise. 091048 I will sell it to you for an amount that will be said, and [we] will put the wines at a

good price and so [we] will drink from the trade. g ... , j'en prendra la chose sur moy, et ø5 direz que je le vous ay fait faire. 015089 I will take it upon myself, and [you] will say that I had you do it. We now turn to our general claims concerning null subjects in V1 embedded clauses in the CNNV. 1.2. Null subjects in V1 embedded clauses In this section, we summarize the results of our examination of null subjects in V1 embedded clauses in the CNNV and sketch an analysis of the facts. In the licensing of null subjects, two factors must be taken into consideration: formal licensing and identification (Rizzi 1986, Adams 1988, Vance 1988, Roberts to appear). Formal licensing of SpecIP by INFL or an element in COMP and identification of the content of the null subject by INFL/AGR are discussed in the next four subsections. 1.2.1. Subordinate clause introducers In the CNNV, it appears that for the purpose of accounting for V1 embedded clauses, either a three way distinction is required among the embedded clause introducers, or, if a two way distinction is maintained, as in Hirschbühler and Junker (1988), Adams (1988) and Roberts (to appear), it differs to varying degrees from the division advocated in these works. As we will see in section 3, it is necessary to make the following descriptive distinctions:

4

1. Non licensing subordinators: the conjonction 'que' and perhaps all of the (simple and complex) conjunctions: they never allow V1, whether with null subjects (except when the verb is in the 2 p.pl.) or with postverbal nominal subjects.

2. Partially licensing subordinators: the 'que' that introduces relative and comparative clauses as well as indirect questions (WH-'que'). 'Que' here does not allow V1 with null subjects, but allows V1 with postverbal nominal subjects.

3. Licensing subordinators: lexical WH-phrases (excluding those instances WH-'que' that are a real WH word), that allow V1 with null and postverbal nominal subjects.

In order to account for some aspects of this situation, recent suggestions by Roberts (to appear, setion 2.3.5) will be adopted. Although in Vigneulles evidence from main clauses shows that INFL (or AGR) has the capacity to formally license a null subject by Spec-Head agreement, we consider that in embedded clauses, formal licensing (or not) of an empty position adjacent to an element dominated by C or a projection of C depends on that element and not on INFL. This might be looked at as the resolution of a conflict between potential licensers: if among two competing potential licensers (e.g. one in INFL and one in C or SpecCP), one is in a position to formally license under government while the other is in a position enabling it to formally license under Spec-Head agreement, the government variant mode of formal licensing prevails.5 5 Santorini (1990) develops an analysis of Yiddish based on Platzack and Holmberg (1990). She suggests that a finiteness operator [+F] is located in COMP (for V2 languages) or INFL (for non-V2 languages). [+F] licenses nominative case under government (strict c-command) and/or agreement configurations, depending on the language. In addition, she defends the idea that empty expletives must be (i) licensed by being head-governed by a case assigner, and (ii) identified by being coindexed with AGR. The analysis sketched in the text could be reformulated by saying that in wh-clauses not introduced by 'que', C0 contains the [+F] feature, while this feature would be absent from the other embedded clauses. It could then be said that licensing in Vigneulles is restricted to head government by a case-assigner when there is a configuration of head-government at hand. The "when-clause" is to ensure that licensing under Spec-Head agreement is possible in Vigneulles when there is no potential head governor, i.e. in V1 assertive main clauses of the SVO type (whether they be analyzed as CPs or IPs). Considering embedded clauses, an empty SpecIP could not be licensed by a [+F] feature in INFL given that there is a higher position (COMP) where a [+F] feature might have appeared. One attractive aspect of the analysis is that it identifies precisely than in the approach adopted here the particular grammatical property responsible for the licensing of SpecIP under government. I have not yet commited myself to this analysis because I don't see why in embedded clauses the finiteness feature would be located in C0 only in those wh-clauses not introduced by 'que'

5

Considering then the three classes of subordinators and the case of null subjects, we would say that the first two types of subordinators mentioned are not formal licensers (don't participate in formal licensing) while those in the last group are. The grouping of partially licensing subordinators (on which we come back in section 1.2.4) with non licensing subordinators rather than with licensing subordinators is important because in previous studies that I am aware of, WH-'que' was grouped with the latter type, either because the facts required this or because they were not looked at in enough detail (as, e.g., in Hirschbühler and Junker 1988). 1.2.2. AGR features As for the identification of the content of the empty category by features of AGR, it appears that, here again, a three way distinction is necessary. Following Roberts (to appear, section 2.2.3-4, 2.3.5) we consider that while Middle French has a formally rich inflectional paradigm (i.e. the paradigm is morphologically uniform, with a slot for agreement affixes available for each person), this paradigm is functionally weak (in that there is more than one syncretism). In principle, either type of richness is enough to identify the content of pro subject in the appropriate structural configuration. We will see that in addition to this, in the CNNV, strength of individual person affixes need to be added to the variables playing a role in the distribution of null subjects when the content of pro in the configuration of Spec-head agreement with INFL. Thus we distinguish between: 1. Strong AGR: 2nd. p.pl. (i.e. person 5). Allows V1 with null subjects in embedded

clauses of all types. That is, the 2nd p.pl. is able to formally license SpecIP (as well as identify its content) under Spec-Head configuration in embedded clauses despite a filled C0 or SpecCP. One would expect 1st. p.pl. (i.e. person 4) to belong to this class, if 'rich' is directly related to transparency of morphological indication of person and number. As the data will suggest, it is not clear that the relation is so direct.

2. Healthy AGR: 3rd singular (expletive, i.e. person 31, and non-expletive, i.e. person 3) and 3rd p. pl. (person 6). These are able to identify the content of pro in a configuration of Spec-Head agreement in V1 embedded clauses provided that the position is formally licensed by a licensing subordinator.

(as I consider that the interrogative 'que' is a WH-word in spec CP, it couldn't be that the [+F] feature is present in C only when C is devoid of lexical material); I also have difficulty conceiving a finiteness operator distinct of INFL (or Tense).

6

3. Weak AGR: 1st and 2nd singular. With some exceptions that will be discussed, they are not found in a V1 construction in any of the embedded clause types. One way to describe their behaviour would be to say that their ability to identify the content of pro in a configuration of Spec-head agreement is parasitic on whether or not they also actually formally license the subject pro position. Given that in embedded clauses any element in C or SpecCP, when adjacent to SpecIP, is a potential formal licenser, weak AGR is never in a position to license pro in that position.

1.2.3. Non licensing subordinators and licenced null subjects Support for the analysis proposed thus far is provided by XP (S)VO subordinate clauses, which are discussed in section 4. There are 'XP (S)VO' subordinate clauses just as there are main clauses of that type. When the subject is null, they resemble V2 constructions, but there are clear indications that these are constructions with XP adjoined to IP. In this situation the subject may be null under Spec-Head agreement, not only in main clauses, but in embedded clauses too, even when the subordinator is of the non-licensing type. The intervening XP prevents C from governing SpecIP, i.e. the complementizer system does not provide a potential licenser. As a result, INFL (or AGR) is the only potential formal licenser for SpecIP, and formally licenses it. And, as might be expected, given that INFL formally licenses SpecIP, it (or AGR) is again able to identify its content, and null subjects are found for persons 1 and 2. 1.2.4. Stylistic Inversion vs null subjects Recently, it has been suggested that the so-called Stylistic Inversion construction should be analyzed, whether for contemporary French (e.g. Deprez' 1988 interesting analysis of Triggered Inversion) or previous periods of the language (Vance 1988, Roberts to appear), by appealing to the presence of an expletive pro in the highest SpecX" node subjacent to C. In Vigneulles, there are 85 cases of V1 with a postverbal nominal subject in embedded clauses. All of the examples are WH constructions (or operator constructions) of one type or another. What is of interest to us is that the subordinate clause introducer may be 'que', as shown in (2), contrary to what we found was the case for null subjects, even for impersonal (uses of) verbs. The examples with 'que' are divided as follows: 10 comparatives, 5 indirect questions, 18 relatives, and 1 topicalization. (2) a car il estoit mieulx l'homme pour le despendre que n'estoit ledit curé, … 011051

7

because he was more the man to spend it than was the priest (comparative) b Ores maintenant je vous dirés que fist ung aultre fol ... 099135 Now I will tell you what did another fool … (indirect question) c Et encor n'est rien au regard des peines et tourmens qu'a endurez nostre doulx Sauveur Jhesu Crist pour nous. (relative) 059028 And [it] still is nothing compared to the pains and torture that suffered our kind

Savior Jesus Christ for us. d ... je ne m'y suis pas fain; aussi croy je que n'a pas fait le Ribauldez, mon compaignon." 058090 … I did not deprive myself of it; neither I do believe that did the bad man, my

friend. (topicalization out of complement clause). ['aussi' is complement of 'faire'; it has the meaning of «non plus» (cf. Martin et Wilmet, 1980, §12)]

Given that Stylistic Inversion is found in V1 'que' clauses with a WH gap, while impersonal null subjects are not, it would seem that the expletive of Stylistic Inversion constructions, if there is one, should be treated differently from the null expletive of impersonal constructions. Impersonal null subjects have a content that needs to be identified by AGR, the content being the features of person and number. In the CNNV, it is apparent that WH-'que' does not formally license the position of an impersonal null subject. This indicates that WH-'que' is not strong enough to license the SpecIP position when INFL (or AGR) is grammaticaly active w.r.t. SpecIP (which is the case here, in the sense that identification of the content of SpecIP can only come from AGR). This is so even when AGR is itself rendered inoperative as a formal licenser for SpecIP by the presence in C or SpecCP of a potential licenser governing it. So, the presence of 'que' in C or SpecCP prevents AGR from acting as an formal licenser, and the relation of AGR to SpecIP, required for content identification, contributes to WH-'que' not being an actual formal licenser for that same position. The situation is quite different in the case of Stylistic Inversion. Here, SpecIP can be thought of as radically empty (i.e. as not containing an impersonal pro) or as associated via some type of coindexation to the postverbal subject. In either case AGR is not active w.r.t. SpecIP; it is active w.r.t. the postverbal subject, assigning it nominative case. Not being active w.r.t. SpecIP, AGR is not a potential formal licenser for that position, resulting in WH-'que' fulfilling that role. As a more detailed examination of Stylistic Inversion is not required for the goals of this paper, it will not be discussed in more details below.

8

These are in a nutshell the main claims of this paper. Keeping them in mind will allow us to have an easier time looking at the data in the CNNV in more detail, to which we now turn. In section 2, we give a general overview of the data. In section 3, we turn to a detailed discussion of null subjects in embedded V1 constructions. In section 4, we examine the cases where a constituent intervenes between the embedded clause introducer and the verb.

9

2. Data Following a standard view, the null subject (Spo) is analyzed as pro. In what follows, its distribution is compared to that of the lexical pronominal subject (Spr). Both types of pronominal subjects together are designated by Sp. Spr and Spo have been divided in several groups to determine the possible role of various factors in the licensing of Spo. The following potential factors have been examined:

- the type of subordinate clause as well as the specific subordinator used - the verb position: does it immediately follow the subordinator or not? - the person and number of the verb - the particular verb used

The distribution of Spo in subordinate clauses of the type 'Sub (XP) V…' is summarized in Table 2a and Table 14 below. Table 2b and Table 13 give us the same information for subordinate clauses with a preverbal lexical pronominal subject Spr. The tables are divided in three parts, the two main ones corresponding in part to the divisions discussed in previous studies: Hirschbühler et Junker (1988), Vance (1988a, 1988b), Adams (1987a, 19887b, 1988) and Dupuis (1988, 1989). In part A, we have grouped together the conjunctive clauses introduced by 'que' or by a conjunctive phrase ending with 'que', as well as WH-clauses introduced by 'que': cleft sentences, restrictive relative, comparative and superlative clauses, and indirect questions. Part B, contains temporal and conditional clauses, as there were reasons to expect that they might pattern with category C as well as A, given some of the existing literature (Hirschbühler et Junker (1988), Adams (1988)). Finally, part C of the tables contains WH-clauses (indirect questions, relatives, comparatives), except for those introduced by 'que'. Before turning to these Tables, we shall give a brief overview of the distribution of null and pronominal subjects in the CNNV. The total number of null subjects in the text is 2403. This number excludes (1) direct questions and (2) the second (third, etc.) of series of coordinate clauses whose null subject is interpreted as identical to the subject of the preceding coordinated clause. The null subjects can be further subdivided as in Table 1. As can be seen, the 513 examples of null subjects in subordinate clauses represent 21.35% of the total number of null subjects in the text.

10

Table 1: Distribution of null subjects (Spo): Person 1 2 3 316 ce 4 5 6 Tot % Main clauses 127 12 927 415 28 28 53 300 1890 78, 61 Subord. cl. 7 6 124 207 4 7 127 31 514 21,38 134 18 1051 622 32 35 180 331 2404 100 The total number of embedded lexical pronominal subjects (Spr) is 3981. This excludes 'on' (one), for which there is no null variant. It includes: - 3884 preverbal Spr immediately the subordinate clause introducer (cf. Table 2b), - 72 preverbal Spr separated from the subordinate clause introducer by some phrase

(cf. Table 13), (to which 9 examples of 'on' might be added), and - 25 postverbal Spr. The 514 embedded clauses with Spo represent 11.43% of embedded Sp, i.e. of the total of Spr + Spo (4495). As mentioned above, this excludes coordinations that could be interpreted as coordinations of VPs, as in (3a) below, but it does not exclude conjoined clauses whose implicit subject is distinct from that of the preceding clause, as in (3b): (3) a …, advint qu'il print ung bachet et ø le rua auprés d'un mullon de foin... 055058 …, [it] happened that he took a pike and [that he] threw it on a stack of hay b … l'enfant luy confessa toute la maniere et comment son pere avoit tuez sa vaiche

et l'avoit sallez et ø en avoient plussieurs fois mangez. 008066 … the child confessed to him how (= that) his father had killed his cow and had

pickled it and [how they] had several time eaten of it. Contrary to null subjects, lexical pronominal subjects appearing after a coordinating conjunction have been included in the counts. Thus, the proportion of Spo vs Spr is bent towards Spr. 3. Sub ø V We first examine the cases where no constituent intervenes between the subordinator and SpecIP, leaving to section 4 the study of examples with an intervening XP. Table 2a summarizes the facts for null subjects in embedded V1 constructions. There are 179 such Spo (second column). The column 'Total ø' gives the sum of the embedded

6 '31' represents impersonal subjects.

11

Spo in V1 (i.e. Sub ø VX, Table 2a) with those in V2 (i.e. Sub XP VX, Table 14) constructions.

12

Table 2a : Subordinator ø V X (179/514 = 34.82% of embedded Spo) A. Que Total ø Sub ø V 1 2 ce 3 31 4 5 6 purpose 7 5 1 4 cause 16 6 6 subject/object 114 22 22 misc. 13 5 5 concessive 7 3 3 result 47 2 1 1 optative 2 0 0 comp. 'que' 6 2 2 R. Rel.: 'que' 28 19 2 17 I.Q.: WH que 0 0 total 240 64 1 3 60 B. Spec. cases Total ø Sub ø V 1 2 ce 3 31 4 5 6 'quant' ('when') 13 5 1 4 Temporal phrase 4 3 1 2 conditional 'si' 63 34 4 1 0 29 total 80 42 1 1 4 3 4 29 C. Wh ≠ 'que' Total ø Sub ø V 1 2 ce 3 31 4 5 6 I.Q.: si 6 3 2 1 I.Q.: wh≠ 'que' 5 3 2 1 NP Appos. Rel. 42 36 11 22 1 2 CP Appos. Rel. 26 11 5 4 2 R.Rel. ≠ 'que' 7 3 2 1 'comme' ('as') 108 17 2 15 0 total 194 73 20 31 1 16 5 Total A, B, C 514 179 1 1 4 24 35 4 105 5 The 179 Spo examples in Table 2a are to be compared to the 3884 Spr examples of Table 2b; they represent 4.40% of the total Sp in V1 constructions (4063). Null subjects are excluded in category A (except for the Strong AGR fifth person and, to some extent the fourth). If only categories B and C are considered, Spo represents 7.41 % of the total of Spo + Spr (6.51% for category B and 8.06% for category C). The embedded V1 construction with Spo is therefore not that frequent compared to the embedded V2 with preverbal Spr, but it can't be ignored, since its distribution appears to be subject to restrictions related to the type of subordinate clause, the person of the verb, and the personal vs impersonal distinction, as is discussed below.

13

Table 2b: Subordinator Spr V X (3884) A. Que Sub Spr V 1 2 ce 3 31 4 5 6 purpose 30 2 1 0 20 1 0 1 5 cause 143 14 4 3 79 10 10 3 20 subject/object 1027 86 32 108 554 103 16 38 90 misc. 62 1 3 2 40 4 1 2 9 concessive 57 6 2 3 34 5 0 4 3 result 379 11 9 10 246 42 3 3 55 optative 10 2 0 1 5 0 1 0 1 comparative / superlative: que 106 6 4 1 67 7 3 8 10 Rest. Rel. : que 506 72 12 3 283 29 14 21 72 I.Q.: WH que 129 12 2 24 37 27 5 7 15 total A 2449 212 69 155 1365 228 53 87 280 B. Spec. cases Sub Spr V 1 2 ce 3 31 4 5 6 'quant' ('when') 267 8 8 24 166 9 4 1 47 Temporal phrase 115 10 2 2 60 8 4 2 27 conditional 'si' 221 26 19 8 97 17 4 33 17 total B 603 44 29 34 323 34 12 36 91 C. Wh ≠ 'que' Sub Spr V 1 2 ce 3 31 4 5 6 I.Q.: 'si' 66 2 2 2 42 6 0 3 9 I.Q.: 'wh' ≠ que 122 17 2 11 67 0 1 7 17 NP Appos. Rel. 87 15 1 0 55 4 3 4 5 CP Appos. Rel. 99 12 0 4 63 8 0 0 12 Rest.Rel. ≠ 'que' 183 38 1 2 98 11 14 5 14 'comme' ('as') 275 38 7 0 78 39 1 90 22 total C 832 122 13 19 403 68 19 109 79 total A, B, C 3884 378 111 208 2091 330 84 232 450 3.1. Category A: 'que' clauses Part A of Table 2a contains the type of subordinate clauses which exclude null subjects in a V1 construction, except for person 5 and perhaps person 4, where person-number morphological marking is transparent. The presence of null subjects of person 5 is striking, as they are 60 Spo subjects against 87 Spr subjects, i.e. 40.81% of Spo subjects. The licensing of Spo subjects is often related to the transparency of person-number Inflectional marking on the verb; in this respect, one would not be surprised if person 4 were to behave very much like person 5. But there are only 3 examples of a Spo subject against 53 cases of a Spr subject for person 4, i.e. only 5.35% of Spo subjects. If a total of

14

56 cases is enough to feel confident that these 5.35% represent a regular syntactic phenomenon, but of low frequency, factors distinct from morphological distinctness will have to be found to explain the difference between person 4 and person 5. I have nothing to contribute on this aspect of the data at the present time. A comparison of Table 2a with the parallel section of Table 2b shows that the absence of null subjects at a person distinct from 5 or 4 is significant beyond any doubt, as there is only one null subject in that case against 2309 examples with a Spr subject. This unique example is of person 3 (against 1365 Spr examples of person 3). Thus the subordinators of Category A do not formally license null subjects in SpecIP and prevent Infl from doing it. (3) … et se print à luitter encontre celle tendre bergiere par tel maniere et par tel force que la ruyst au bas…. 069015 …and [he] started to fight with this shepherdess in such a way and with such force

that he threw her down … One more comment needs to be made regarding Category A of Table 2a. Given their behaviour w.r.t. to null subject, comparatives-superlatives, restrictive relatives and clefts (grouped under R.Rel. 'que'), as well as indirect questions headed by 'que' (absent from Table 2a from lack of examples) have been grouped with conjunctive clauses introduced by 'que' rather than with the WH-constructions found under Category C. When we take into account only persons 3, 31 (expletive) and 6 (excluding persons 1 and 2, given that they are never represented by null subjects in Category C, as well as person 4 and 5 because they could allow a null subject in SpecIP whatever the type of the embedded clause), there are 486 examples of WH clauses introduced by 'que' with a pronominal subject against none with a null subject. By comparison, the percentage of Spo as opposed to Spr is 34.54% for person 5, and 8.33% for person 4 (but again, there are only two examples of a null subject in this last case). Thus, these wh-clauses headed by 'que' behave just as regular non-wh 'que' clauses. This is important since in other cases they have been assumed or shown to behave like wh-clauses in general (see for example Adams 1988, Hirschühler and Junker (1988), Vance (1988), Roberts (to appear)), the distinction between the two types of clauses headed by 'que' being attributed then to the presence of a +WH feature in the 'que' relatives. In the present case, one is led to conclude that in the CNNV, a WH-feature associated with 'que' is not enough to formally license a SpecIP position that

15

dominates pro. As we suggested in 1.2.4, the situation just described could be described by saying that when INFL/AGR is grammaticaly active w.r.t. SpecIP as the identifier of its content, it counts as a potential formal licenser for it, but not as an actual one, because of the filled embedded C or SpecCP. Competition with another potential formal licenser prevents +WH-'que'from formally identifying SpecIP. This concludes our survey of category A with V1 embedded clauses. We now turn to categories B and C which will be discussed at some length. 3.2. Categories B and C. A glance at Categories B and especially C of Table 2a reveals that numberwise null subjects are nearly absent for person 1, 2, and 4. This leads to the tentative hypothesis that in the context of embedded clause introducers which (at least for a number of them) don't exclude the possibility of null subjects in the V1 construction, certain persons of the conjugation are unable to identify null subjects in SpecIP. This hypothesis is based on a comparison between Spo and Spr for each person; in Table 3 below, 'ce' has been omitted:

Table 3: Proportion of null subjects w.r.t. the total of Sp

Category B Sub Sp(o)V 1 2 ce 3 31 4 5 6 total (T 2a) Spo 42 1 1 4 3 4 0 29 0 total (T 2b) Spr 603 44 29 34 323 34 12 36 91 % Spo 6.96 2.22 3.33 10.52 0.92 10.52 0 44.61 0 Category C Sub Sp(o)V 1 2 ce 3 31 4 5 6 total (T 2a) Spo 73 0 0 0 20 31 1 16 5 total (T 2b) Sp 832 122 13 19 403 68 19 109 79 % Spo 8.06 0 0 0 4.72 31.31 5 12.8 5.95 total B+C Sub Sp(o)V 1 2 ce 3 31 4 5 6 Spo 115 1 1 4 23 35 1 45 5 Spr 1435 166 42 53 726 102 31 145 170 % Spo 7.35 0.59 2.32 7.01 3.16 25.54 3.12 23.68 2.85 The percentages of Spo for each person w.r.t. the sum of Spo (Table 2a) and Spr (Table 2b) show that only persons 31 (the impersonal (uses of) verbs) and 5 have a high percentage of null subjects in the B and C types embedded clause types. As for person 3, the substantial number of examples of that person, in particular in Category C, allows one to conclude that the percentage of null subjects of person 3 reflects a regular syntactic phenomenon, simply

16

of low, but not insignificant, frequency.7 'Ce' will be commented upon later. As there is only one example of a null subject of person 4, the higher percentage of this person compared to person 3 may not be taken as reflecting a regular phenomenon. Here, as in category A, person 4 differs from person 5 in that the clear morphological marking of person and number on the verb is not correlated with a high percentage of null subjects. Considering categories A, B, and C, null subjects represent only 4.54% for person 4, while they represent 31.15% for person 5! This difference cannot be imputed to a formal difference in the morphological distinctness, as both these persons are equally morphologically distinct from all the others. 3.2.1. Category B: temporal and conditional clauses Three types of clauses have been considered here: clauses introduced by 'quant' ('when'), other temporal clauses, and conditional clauses. They will be examined in that order. 3.2.1.1. 'Quant' A priori, one may be tempted to analyze 'quant' as an item of category C, that is as a WH phrase. Altough null subjects represent quite a small percentage (1,83%), one might want to rely on the mere existence of such examples as supporting the status of 'quant' as a formal licenser. However, an examination of the data shows that 'quant' does not appear to regularly license an empty element in SpecIP, and that it should therefore be grouped with elements of category A. A first argument comes from a comparison with some OF data. Old French texts like La Mort le Roi Artu and La Queste del Saint Graal exceptionally allowed cases of V1 constructions with null subject in embedded clauses introduced by 'quant' (as well as in conditional clauses, to which we come in 3.2.1.3), contrary to what was the case for other types of subordinate clauses. But, upon examination (cf. Hirschbühler 1990), the only unequivocal examples of null subject V1 clauses introduced by 'quant' noted in these texts were all variants of the same impersonal expression, "quant vint ..." ('when [it] came to ...],

7 According to Santorini (1990, fn 10), 1% is the level at which a phenomenon does not reflect a regular feature of the language: "On the basis of detailed quantitative work of my own and others, it appears that it is common for well-established generalizations in a language to be violated in naturally-occurring usage at a low, relatively constant rate of about 1%. For instance, the relative frequency of resumptive pronouns in English in non-island environments is around 1% (Anthony Kroch, pers. comm.)"

17

which alternates with "quant ce vint ..." (cf. Zink, 1987, Martin et Wilmet 1980, § 325, 2). In the CNNV, the parallel expression is always realized with an expressed pronominal subject ("quant se/ce vint..."), which is unexpected if 'quant' participated in formal licensing of SpecIP. Consider then the 4 examples of a null subject with an impersonal verb. They consist of two expressions: "quant est de moy" ('as far as [it] is of me', i.e. 'as far as I am concerned'; 026016; 098206), "quant est à nous" ('as far as we are concerned'; 090277), and "quant luy en souvint" ('when [it] came back to him', i.e. 'when he remembered'; 062017). In this last case, 'luy' could actually be a strong form, i.e. it could start an embedded V2 construction, or a clitic form, but there is no example with 'moi' (strong) or 'me' (clitic) that would allow us to decide. These type of examples could represent set expressions, relics of an older stage of the language. Importantly, there is no example with a null subject involving the type of impersonal expression that is found in the 9 examples with an Spr subject (one of these cases, 091279, omitted below, is an example where 'quant' is taken up by 'que', and so one might want to classify it with 'que'): (4) a ...car, quant il fait beau, je le vueil bien, et quant il pleust, aussi fais je. Quant il faict froit ou quant il faict chault, … 034033/34 …since, when it is nice weather, I am happy with it, and when it rains, also am I.

When it is cold or when it is warm, it is always all the same to me b … quant il vous plaira, (…when it it will please you) 048024 c … quant il tonneroit. (…when it it will thunder) 083102 d … quant il estoit question de paier taille au prince ... 091018 …when it was the moment to pay tax to the prince e … mais quant il failloit paier les feuz ou gabelle, 091065 … but when it was necessary to pay for each house or for the salt tax The final example of 'quant' is not impersonal. It is of person 2. (5) … quant ø n'a seullement la puissance de saillir une fois 039081 … when [you] don't even have the strenght to make love once This is the only example of null person 2 in a V1 subordinate clause in the whole text, and the only null personal subject with 'quant', against 258 pronominal personal (i.e. excluding the impersonal) subjects. Here the verb is preceded by the negative marker 'ne'. It has often been noted that 'ne' was present in the case of a null subject in contexts where a

18

null subject was otherwise normally excluded. It is therefore possible that here also, the null subject in SpecIP should not be specifically related to the presence of 'quant'. All this suggests that 'quant' should be assimilated to conjunctions of Category A as far as null subjects in an embedded V1 construction are concerned, despite the fact that it might be analyzed as a temporal free relative.

19

3.2.1.2. Other temporal clauses There are three cases of a null subject distinct from person 5 in a temporal clause introduced by a complex phrase (avant que, depuis que, devant que). Two are of person 3 and one is of person 1 (the only case of that person in an embedded V1 clause in the text). (6) a Mais avant que ø3 ce partist, ... 011008

But before that [he] refl-left, … b ... je me suis cuidiez lassez morir depuis que ø1 partys de ceans. 017056 … I though that I would nearly die since [I] left from here… c ... et dit qu'il courrerait une lance devant que ø3 entre au four ... 096092

… and he said that he would make love once before that [he] enter the oven … I am not sure what to say here, given the small number of Spo against the 115 corresponding examples with a preverbal Spr in Table 2b (2.54% Spo), and in particular against the 60 examples of Spr of person 3 against only two Spo (3.22% Spo). Perhaps, as the possibility was raised in the case of 'quant', temporal clauses could be analyzed as involving WH-movement (operator movement). The difference between 'quant' and these complex subordinate phrases could then be that 'quant' would be treated idiosyncratically in the same way as WH-'que', while the expressions with more phonetic weight would be treated in the same way as the WH subordinate clause introducers of Category C as far as formal licensing of SpecIP is concerned. This would however not be enough to account for (6c), with a null subject of person 1, which is a weak AGR (see 1.2.2 and below). At this point, I consider these examples as potentially problematic. 3.2.1.3. Conditional clauses Consider finally the case of conditional clauses. On a total of 34 examples with Spo, 29 are of person 5 and the 5 others represent 'ce' in 4 cases and person 3 in one case (against 118 examples with Spr for 'ce', person 3 and the impersonal, i.e. Spo represents 4.06% of all types of 3rd person singular). Given these 5 last examples, should conditional clauses be grouped together with the embedded constructions in category C, whose introducing element participates in licencing a null SpecIP? Probably not. It should be pointed out that in the Old French texts Mort Artu and Queste du Saint Graal, conditional clauses represented the second type of V1 embedded clauses with null subject (the other one being the sentences in 'quant', as noted above). But all the examples

20

were cases of the unique expression "se ne fust...", i.e. "si ce n'était pour ..." ('if it were not for') (Hirschbühler 1990).8 In the CNNV, three or four examples are of this type: (7) a -… et se ø n'eust esté qu'il tenoit sa chemise endroict le trou, … 081129 … and if [it ]weren't that he was holding his shirt against the ashole, … b -Et se ø ne fut cela, … 096027 And if [it] weren't for that, … c … comme se ø fut esté toutes les punasies du monde,… 028135 … as if [it] had been all the stench in the world, … Example (8) is possibly of that type, as the version with a pronominal subject would most probably 'ce' rather than 'il': (8) -... tous les gens des villaiges circonvoisin y venoient pour ouyr son sermon comme se fust estez Dieu, … 034005 … all the people from the surrounding villages were going there to listen to his

sermon, as if he had been God, … In all these cases, given that omission of 'ce' is otherwise exceptional in the text, the absence of 'ce' here might be favored in order to avoid sequences like "se ce … ", in a manner reminiscent deletion of partitive articles 'du', 'de la' des' after after preposition 'de' in examples like "je parle de chevaux" ('I am talking about horses'), from underlying "je parle de des chevaux". This interpretation is however weakened by the fact that sequences like 'se ce' are not categorically excluded: (9) … se ce ne fut de peur de gaster mon lit, … 041067 … if it was not for fear of damaging my bed, … Considering finally example (10), one notes that it is introduced by 'si' rather than the more usual 'se' 'Se', as opposed to 'si', introduces the conditional clause in 3/4 of the 8 The null subject is parallel to 'ce' rather than 'il', and the omission of 'ce' could therefore be due to avoidance of strings like "se ce", just as there is deletion of one 'de' in "je parle de (des) chevaux", given that omission of 'ce' is otherwise exceptional in the text. Sequences like "se ce' are however not categorically excluded: (i) … et vous dis bien, …, que se ce ne fut de peur de gaster mon lit, … 041066 … and I tell you, …, that if it were not for fear of spoiling my bed,

21

cases when a pronominal subject is present; in the case of a null subject (with or without an XP appearing between the conjunction and the verb), the conjunction appears 60 times as 'se' and only twice as 'si', one of those two cases being example (10). This example could therefore easily be reinterpreted as "s'i' l'eust", with a pronominal subject. (10) -et n'y avoit homme, si l'eust congneu et il l'eust veu en cest estat, qui … 014048 - and there wasn't [any] man, if [he] had seen him in this state, who … So, provisionally perhaps, I conclude that it is doubtful that conditional 'si' contributes in a productive way to the licensing of null subjects in the present text, and that the examples above are either fixed expressions, orthographic confusions, or, in the case where the null subject would correspond to 'ce', possibly the result of a "stylistic" deletion rule. The general conclusion is that there is no clear indication that the embedded clause introducers of category B are involved in the regular licensing of null subjects in the CNNV. We will see that things are different in the case of a number of WH clauses. 3.3. Category C. We will go into the details of the various clauses of part C of Tables 2a and 2b. But first, let us have a brief overview of the main facts regarding clauses of this category. These are summarized in table 4 (a subpart of Table 3).

Table 4: Spo vs Spr in subordinate clauses of type C

person 1 2 ce 3 31 4 5 6 total Sp0 0 0 0 20 31 1 16 5 73 Spr 122 13 19 403 68 19 109 79 832 % Spo 0 0 0 4.72 31.31 5 12.80 5.95 8.06 We may first note that there is not a single example of a null subject of person 1 or 2, while there are examples of null subjects for all the other persons. The facts are clear for person 1, as there are 122 examples with a pronominal subject of this person in clauses of type C; they are perhaps less clear for person 2, given that there are only 13 examples with a pronominal subject. The absence of null subjects of persons 1 and 2 in clauses of type C is what led us to suggest in section 1.2.2 that for persons 1 and 2, the capacity to identify the content of SpecIP via Spec-Head agreement is available only if INFL simultaneously formally

22

licenses the same position. As in subordinate clauses the power to formally license SpecIP is normally reserved to the C system, realization of persons 1 and 2 by a pro subject will be excluded in the environment discussed. The only examples which go against the general ban against null subject of person 1 or 2 in a V1 embedded clause are (5) and (6b), discussed in section 3.2.1.2. There are two possible analyses for these. Either the complementizer licenses SpecIP, in which case identification by person 1 or 2 is exceptionally not "parasitic", or INFL acts exceptionally both as a licenser and as an identifier for SpecIP in a V1 embedded context. In contrast to persons 1 and 2, the impersonal is frequently null, and person 5 also is, although clearly not as much. One unexpected fact is that person 5 is found more frequently in category A of Table 2a than in category C (40.81% against 12.80%). For the other persons, there are only between 4% and 6% of null subjects. One factor that we didn't discuss in section 1 is the opposition between impersonal vs non-impersonal verbs or uses of verbs in clauses of type C. In these clauses, the null subject is more frequent with impersonal uses than with person 3. The contrast may be attributable to discourse conditions or it may simply reflect a weaker requirement on local identification for the impersonal, i.e. person and number agreement could be a process independent from transmission of a referential index. We have no definite idea on this aspect of the data presently. These general considerations being made, let us take a closer look at the data of Category C. This will provide more justification of what was said about WH-'que' in section 1.2.4 as well as allow additional observations. In particular, we will see that there is no clear indication that null subjects are regularly allowed in indirect questions or in clauses introduced by 'comme'. This would leave only the introducer of relative clauses as a regular formal licenser of SpecIP. If other texts display the same characteristics, it would show either that the licenser of the empty category in subject position is not simply the WH word introducing the clause, or that other factors, still to be discovered, play a role. This detailed examination, some aspects of which may perhaps not be of immediate theoretical import for the analysis presented in this paper, will allow us to get a clearer picture of the grammar of the CNNV and may become important if new facts or new theoretical frameworks raise new questions. 3.3.1. Indirect questions

23

In the case of indirect questions, although null subjects are found in the CNNV, a close examination of the examples raises doubts as to whether they are productively licensed in this construction. Yes-no questions (introduced by 'si') and WH questions will be examined separately. The main facts are summarised in Table 5.

Table 5: Spo vs Spr in indirect questions (excluding 'que') Category C Sub Sp V 1 2 ce 3 31 4 5 6 Q.I.: si Spr 66 2 2 2 42 6 0 3 9 Q.I.: si Spo 3 0 0 0 2 0 0 0 1 % Spo 4.34 4.54 — 10 Q.I.: wh Spr 122 17 2 11 67 0 1 7 17 Q.I.: wh Spo 3 0 0 0 2 1 0 0 0 % Spo 2.4 2.89 3.3.1.1. Yes-no questions First, for the 'si' of indirect questions, the percentage of null subjects is low (4.34%) and is based on only three occurrences of a null subject. As these null subjects represent persons 3 and 6, we might want to give them a lot of weight, more than if they were of person 31 or 5 for instance. But on the opposite, we may also wonder if si here does not stand for s'i (sg. and pl.), in which case there would be no example of a null subject. Given previous work, we actually would not expect null subjects here, as the 'si' complementizer does not belong to the category of WH words. (11) a ... luy demanda s'il … et si ø3 seroit et demeureroit tousjours aussi asne et beste.

013036 … [he] asked him if he … and if [he] would be and remain always as stupid and

silly. b je ne sçay si ø3 se doubta de ce qu'estoit … 041056 I don't know if [he] suspected what was going on … c … Dieu sait si ø6 le raillerent … 052056 … God knows if [they] scoffed at him … 3.3.1.2. Wh questions

24

In the case of WH indirect questions, there are again only 3 cases of null subjects on a total of 125 examples (2.4%). These three examples are given in (12): (12) a … luy demanderent pour quoy ø3 s'en estoit allez. 031029 … [they] asked him why [he] had gone away b Ores escoutés de quoy ø3 se advisa. (I.Q rather than free rel.) 092023

Now listen to what he thought. c ... je leurs diroie si bien qu'ilz m'entenderoient, assavoir mont de quoy ilz se meslent ne … ne de quoy ø31 leur en appartient. 044052 … I would tell them so well that they would pay attention to me, that is what they

are meddling and … and what [it] matters to them. Although the percentage of Spo is small (but the percentage is close to 3% for person 3 and a little above 4% for 3 and 31 put together), these examples don't look problematic in any respect and we take them to be the first clear case where the choice of subordinate clause introducer participates in the licensing of a null subject. Indirect questions introduced by 'que' display very different properties, as they are followed 129 times by a pronominal subject (Table 6) and never by a null one.

Table 6: I.Q. introduced by 'que' with Spr preverbal subject

Category A Sub Sp V 1 2 ce 3 31 4 5 6 I.Q.: wh que 129 12 2 24 37 27 5 7 15 Indirect questions are the clearest case where 'que' introducing an embedded clause is a WH word; we consider it to occupy SpecCP, just as it would in main clause. The absence of null subject is not only an indication that WH 'que' does not formally license a null specIP, it also shows that the verb does not move to C in the context under consideration, as otherwise we would expect a number of postverbal null subjects. Particularly strong support for the view that indirect question 'que' is not a licenser comes from impersonal verbs, which represent the category most easily found with a null subject in relative clauses. Their total absence with a null expletive subject in 'que' indirect questions contrasts with the high number (i.e. 27) of impersonal verbs with a pronominal expletive subject.

25

(13) a Et luy, qui …, ne sentit rien et ne savoit qu'i luy fut advenu, 091288 And he, who …, didn't feel anything and didn't know what it happened to him b … je te dirés qu'il te fault faire. 094055 … I will tell you what it is necessary for you to do De ces parolles se print tres fort à rire le coustellier et veit bien quelle heure qu'il

estoit.9 091111 The variety of the verbs involved (advenir (15), falloir/failloir (7), sembler à (2), y avoir (2), être (1)) reduces the possibility of an accidental gap in the data. 3.3.2. Relatives The relative clauses have been divided into appositive relatives with a nominal antecedent, appositive relatives with a CP antecedent, and restrictive relatives. If the three types of relatives clauses are put together, it can be seen that the proportion of null subjects in this construction compared to that of Spr subjects is relatively important for some persons and that the number of examples may be high enough for the percentages to be reliable, in particular for persons 31 and 3, and perhaps 6.

Table 7: relatives not introduced by 'que' Type total 1 2 3 31 4 5 6 total (less p.31) Relatives Spo 50 0 0 16 28 1 1 4 22 Relatives Spr 363 65 2 216 23 17 9 31 341 % Spo 11.89 0 0 6.89 54.90 5.55 10 11.42 6.06% Table 7 clearly shows that relatives not introduced by 'que' allow null subjects, especially in the case of person 31, and to a lesser extend 3 and 6. Persons 1 and 2 seem to be categorically excluded. Some of the mechanisms involved have been discussed in section 1.2. The discussion there is however incomplete as we refrain from being concrete as to what the property of a subset of relatives and indirect questions (those not introduced by 'que') is that leads to formal licensing of SpecIP. But see Santorini's approach in footnote 5.

9 As the wh phrase "quelle heure" is followed by the complementizer 'que', the example has been classified with the I.Q. introduced by 'que'.

26

3.3.2.1. Appositive relatives Appositive relatives have been divided in two groups, those with NP antecedent and those with CP antecedent. They will be examined in turn. 3.3.2.1.1. NP appositive relatives NP appositive relatives have a large proportion of null subjects: 29.26% Spo against 70.73% of Spr. (To the examples of Spr, one may add 3 examples with 'on' which have not been included in the table, as 'on' as no null variant). It is important to note the weight of the impersonal: on 36 null subjects, there are 22 examples of expletive ones; this is to be compared to only 4 examples of impersonal verbs with an expressed pronominal subject. Null subjects appear to be the rule in the case of impersonal. The proportion of null subjects is high for person 3. It is higher for person 6, but here the total number of occurrences of Sp (7) is not high enough to make definitive claims regarding that person. Attention will therefore be focused on persons 3 and 31.

Table 8: Spo vs Spr in appositive relatives. Category C Sub (Sp) V 1 2 ce 3 31 4 5 6 R.A. Spo 36 0 0 0 11 22 1 0 2 R.A. Spr 87 15 1 0 55 4 3 4 5 %Spo 29.26 0 0 0 16.66 84.61 25 0 28.57 First of all, a closer look at the examples seems to justify a tentative division of NP appositive relatives in two classes. The first one, with 13 Spo examples, is characterized by the fact that the relative is introduced by a WH word acting as a determiner of a noun or by a WH word starting a new sentence after a full stop (or both).

Table 9: Appositive relatives of type 1 Catégorie C Sub (Sp) V 1 2 ce 3 31 4 5 6 R.A.1 Spo 13 0 0 0 3 10 0 0 0 R.A.1 Spr 6 0 0 0 4 1 0 0 1 %Spo 68.42 — — — 42.85 90.90 — — 0 In all the examples of this type, the relative is preceded by a preposition, as in (14): (14) a … mais qu'elle volcist … obeir à sa requeste. À laquelle requeste ø3 ne volut

oncque obeir … 039019

27

… as long as she would … obey his request. To which request [he] would never obey …

b Et premier arriva en Ytalie, auquel pays ø3 trouva pluseurs choses estranges, 100055

And first [he] arrived in Italy, in which country [he] found several strange things …

c et en icelluy prey y avoit pluseurs belles fleurs,…, entre lesquelles fleurs ø31 y avoit de ces belles grandes marguerites. 043038

… and in this meadow [there] were several pretty flowers, …, among which flowers [there] were [a number] of these nice and tall daisies

The ten examples with an impersonal all involve '(y) avoir', which is never found with a pronominal subject in this group. These ten examples should perhaps therefore have no more weight than the unique occurrence of an impersonal verb with a lexical expletive subject. (15) Et advint ung jour et assés tost aprés que les Rogacions … vindrent, auquel jour il

fault aller en pourcession trois jours durant … 023008 And [it] happened one day and soon after that the 'Rogacions' … came, in which

day it is necessary to walk in procession for three days Although there is not enough data here to come to a conclusion, one might have the impression that null subjects here are restrcted to cases where the initial WH is a PP, as in the four cases where it is an NP, the subject is Spr. However, as we will see, data from the second class of NP appositive relatives militates against that impression. Let us turn to the second class of NP appositive relatives (Table 10), which is simply made of all of those which don't fit the description of the first class.

Table 10: Appositive relatives of type 2 Catégorie C Sub (Sp) V 1 2 ce 3 31 4 5 6 R.A.2 Spo 23 0 0 0 8 12 1 0 2 R.A.2 Spr 81 15 1 0 51 3 3 4 4 %Spo 22.11 0 0 0 13.55 80 25 0 33.33

28

In this category, null subjects are not limited to person 3 and 31. There also is also a large variety of verbs: 'se pouvoir émerveiller', 'souvenir' (2), 'y avoir' (6), 'être compté', 'prendre dévocion', 'se doubter', 'penser', 'tarder', 'être congneu', 'prendre', 'venir', 'donner', 'pouvoir aller', 's'accompagner', 'parler', 'avoir', 'se trouver' (once for each verb). The fact that null subjects are well distributed among persons and that it is not restricted to a few particular verbs indicates that the V1 construction with Spo reflects a productive grammatical phenomenon in appositive relatives. With 80% of Spo and the bulk of the cases, the impersonal is responsible for the high proportion of null subjects in the appositive relatives of type 2. These 80% have to be qualified by the fact that half of the cases of Spo involve 'y avoir'. Even if we take the examples involving this expression out, we still have 66.66% of expletive Spo. The examples involve 'souvenir à quelqu'un' twice, 'tarder à quelqu'un', 'prendre devotion à quelqu'un', 'venir quelqu'un' and 'être conté de quelqu'un par quelqu'un', i.e. it involves obligatory impersonal verbs, unaccusative ones, and one impersonal passive. With 17.18%, on a total of 64 Sp, person 3 confirms that null subjects are not an accident in NP appositive relatives. A few examples are given in (16). (16) a … une grainge joindant de leur maison, en laquelle ø3 pourroit aller facillement et

secrettement de nuyct... 047012 … a barn joining their house, in which [she] could easily and secretly go at night

… b … il demanderoit l'ausmone, laquelle ø3 ne pensoit jamais qu'elle luy deust

refuser ... 018030 … he would ask for charity, which [he] never though that she would refuse him

… c Symonnat, à qui ø31 ne souvenoit plus de l'an passés, vint à confesse. 002036 Symonnat, to whom [there] was no recollection of the previous year, went to

confession … d … il arivait en l'abayee aucun gentil homme de Lorraine, auquel ø31 fut comptez

des fais de messire Jehan Pare ... 005067 … there came to the abbey a gentleman from Lorraine, to whom [it] was told of

the deeds of master Jehan Pare … e … je vueil icy compter ceste presente adventure laquelle ø31 n'a guiere qu'elle est

advenue à Mets. 041002

29

… I want to tell here this present event which [it] is not long ago that it happened in Mets. [with he resumptive pronoun]

f … nous avons trouvez ung homme mort et murtry … lequel ø4 avons prins ... 020189

… we have found a man dead and killed …, which [man] [we] took … g … là où on ce devisoit de pluseurs besongnes, entre lesquelles ø6 parloient des

divers marchez qui se tenoient en divers lieux par les citez ... 020007 … where people were talking about different things, among which [they] talked

about the different markets that were held in different places in the towns … Regarding the hypothesis (sugested by the first class of appositive relatives) that Spo might be excluded if the fronted WH phrase is an NP rather than a PP, the three examples with an NP that are found in the second class of NP appositive relative go against it. In section 1.2.1, we adopted the often taken position that the WH-phrase itself played a role in formally licensing SpecIP. There is however another very natural hypothesis which deserves a few words. One might be tempted to associate the examples of NP appositive relatives of type 2 with Spo subjects to V2 constructions in main clauses: the relative here can often be paraphrased by 'et' ('and') followed by a non-WH fronted phrase, which would normally induce V-to-Comp. At the present time, we reject that approach to the data, given the absence of Spo of person 1, as these are well attested in a postverbal position in the text. 3.3.2.1.2. CP appositive relatives In CP appositive relatives, Spo makes up 9.48% of the total, which is just under half of what we saw for the second type of NP appositive relatives. There are enough cases to be sure that Spo is a regular phenomenon here. Null subjects are found for person 3, 31 and 6. There is no example of either Spo or Spr subject for person 2, 4 and 5. In the context of the absence of null subjects of person 1 in category C, we take the absence of Spo of person 1 here against 12 Spr as significant.

Table 11: CP relatives

Catégorie C total 1 2 ce 3 31 4 5 6 R. Ph.Spo 11 0 0 5 4 2 R. Ph. Spr 99 12 4 63 8 12

30

% Spo 9.48 0 0 7.35 33.33 14.28 The examples are characterized by the variety of verbs used with Spo ('respondre', 'ne pouvoir faire', 'se prendre à', 'ne savoir', 'estre mandé', 'estre commandé', 'avoir été possible', 'estre ri' (2), 'ne vouloir souffrir', 'dire'), which provides independent confirmation for the regularity of the construction. Some examples are given in (17). (17) a Si luy demande le seigneur devant-dit qu'il voulloit faire de ses escailles d'oefz. A

quoy ø3 respont qu'i les voulloit vendre … 020032 The Lord asked him what he wanted to do with the shells of the eggs. To which

[he] answers that he wanted to sell them … b Ores advint une nuyt qu'il avoit permis d'aller veoir et visiter icelle saincte dame,

laquelle chose ø3 ne pouvoit bonnement faire … 048006 There came a night when he got permission to see and visit this saint woman,

which thing [he] could not really do … c … il luy souvint du verlet au barbier et de son beau membre, par quoy ø3 se print

si très fort à rire que … 093041 … he remembered the barber's servant and his nice organ, by which [he] started to

laugh so much that … d …il y eust ung Cordellier preschant la Resurrection à qui on avoit comptez

l'hystoire nouvellement advenue, qui recita ce compte audit sermon et en plaine predicacion, de quoy ø31 fuit assés ris. 072083

… there was a Cordelier … , who recited this story in the sermon and in the midst of the predication, as a result of which [there] was a lot of laughter.

3.3.2.2. Restrictive relatives There are only 3 cases of Spo against 183 cases with Spr, i.e. 1,61%. The percentage goes up to 2.04% if person 1 and 2 are excluded, on the basis of the fact that they are never realized as Spo in Category C. In the case of person 3 there are 98 cases of Spr against none of Spo. Impersonal verbs fare better with 2 Spo against 11 Spr, i.e. 15.38%. The one case of Spo for person 5 against 5 with Spr tells us nothing, given that person 5 in principle allows Spo in any sort of subordinate clause. The 3 examples are given in (18):

31

(18) a … elle pensa à demander le nom du recepveur à qui ø31 luy failloit parler,… 082023

… she though of asking the name of the tax collector to whom [it] was necessary that she talked, …

b … car ledit prebtre fist faire ung gros pastez … auquel ø31 y avoit dedans des bons oysellet … 048031

… because the said priest had someone made a large pie … in which [there] was in-it good birds

c …, car avec ceste hystoire de Braie laquelle ø5 avés icy devant ouy … 083064 … because, with this story abiut Braie that [you] just heard … The general conclusion is that if it weren't for the two cases of impersonal Spo, we would say that characteristics of the CP system don't provide for formal licensing of SpecIP in restrictive relatives in the CNNV. 3.3.2.3. Conclusion on relative clauses As we saw in Table 7 at the beginning of section 3.3.2, when the three types of relatives clauses are put together, the proportion of null subjects in this construction compared to that of Spr subject is relatively important for some persons and the number of examples may perhaps be high enough for the percentages to be significant, in particular for person 31 (54.90% on 51 examples), 3 (6.89% on 232 examples), and perhaps 6 (11.42% on 35 examples). However, looking more closely, it is clear that the vast majority of examples with Spo come from NP and CP appositive relatives, and that restrictive relatives not introduced by 'que' contribute nearly no Spo example. It is therefore clear that some factor should allow us to distinguish between the two types of relatives, either a grammatical one, a discourse one, or some other. One hypothesis that we briefly considered was that in NP appositive relatives, and the same hypothesis could have been made for CP appositive relatives, the fronted WH phrase was taken as a topicalized constituent triggering V-to-C, exactly in the way that any non-WH phrase fronted to SpecCP results in V-to-C. We don't adopt that approach, given the 27 cases of Spr of person 1, against none with Spo, in NP and CP appositive relatives. As made clear earlier, we have not identified the particular factor responsible for formal licensing of a null subject in SpecIP in some WH clauses. As a result it is useless to speculate at this point on what the relevant

32

difference between NP and CP appositive relatives vs restrictive relatives, and for that matter, WH indirect questions perhaps, is. Clearly, this area is open for further research. 3.3.3 Comme In order to be complete, one last type of clause must be considered, i.e. clauses introduced by 'comme' ('as'). In this case, apart from person 5, where null subject is licit independently of the complementizer, there are only two examples of a null subject, both impersonal.

Table 12: comme Catégorie C Sub Sp V 1 2 ce 3 31 4 5 6 comme Spr 275 38 7 0 78 39 1 90 22 comme Spo 17 0 0 0 0 2 0 15 0 % of Spo 5.82 0 0 — 0 4.87 0 14.28 0 On the 15 cases of person 5, we find 14 times the same verb in the same formula ("comme avés ouy" 13 times and "comme avés cy devant ouy" once), and once in a variant of it ("comme avés tousjours dit"), which results in the percentage for Spo being very low if the one expression commonly used with Spo is discarded. If, as should be the case, person 5 is not taken into account, facts are very similar to what was seen in restrictive relatives not introduced by 'que', i.e. the remaining null subjects are limited to the impersonal. (19) a … tout fut perdus comme ø31 en est de nous, … 078164 … everything was lost, as far as we are concerned b … je volroie des poullaine à grant soulez comme ø31 se pourtoit au vy temps.

091099 … I would like to have pointed shoes as [it] was the fashion in the old days. It is not inconceivable that the rarity of the V1 construction here is simply the result of a stylistic preference resulting in impersonal verbs being preceded by a complement in initial position of IP if there is no formal subject expressed. This would be supported by the large number of V2 constructions with a non subject in initial position embedded under 'comme', i.e. there are 54 occurrences of 'comme dit est' and 37 other cases, among which 22 are with an impersonal expression and 15 with a personal use of verb. Examination of the 22 impersonal verbs involved in the V2 construction, the two examples in (19), and the

33

39 examples with the expletive 'il' appearing before the verb did no lead us to see anything distinguishing the cases with a fronted complement from those with the expletive 'il'. 3.4. Summary on Spo in embedded V1 We have looked in minute detail at the different types and subtypes of constructions, in order to ascertain whether there were finely grained differences between them. Excluding the examples with person 5, it appeared that null subjects in a V1 construction existed to some degree in WH constructions, but with unexpected differences. First of all, it seems that in the text considered, WH clauses introduced by 'que', whatever their type, don't allow the construction. The construction is also extremely rare in WH indirect questions, in the remaining restrictive relatives, as well as in the examples introduced by 'comme'. The construction seems to be relatively common in appositive relative clauses not introduced by 'que'. Under the approach described in section 1, where it is assumed that some property of C or SpecCP (whose particular nature I have not committed myself to) is associated to formal licensing of a null SpecIP, we don't expect those differences. As far as I can tell, although it may very well be the correct explanation for the data for which it was originally put forward, the account in terms of a feature [+finite] located in C does not seem to provide more light on the data considered here than saying that a feature associated with WH phrases is transferred from SpecCP to C, which feature is responsible for formal licensing of an empty SpecIP. There are also important differences according to the person considered in WH constructions. There is no example of a null subject for persons 1 and 2, and there seem to be enough examples of a pronominal subject for person 1 (122 with Spr in Category C, and 27 in NP and CP appositive relatives) for it not to be an accidental gap. There are examples for all the other persons, and the numbers are quite high in the case of the impersonal one. Although the percentages are not very high in the case of person 3, it reaches 6.89% in relative clauses not introduced by 'que', and goes up to 11.94% (i.e. 16 Spo against 118 Spr) in the case of appositive relatives (NP+CP). It is high enough, and it involves enough examples, to be confident that it reflects a real grammatical feature of the language. The same conclusion is probably correct for person 6, but the total number of relevant examples (5.95% Spo on a total of 84 examples for all WH clauses not introduced by 'que', and 11.42% on a total of 35 examples for relative clauses not introduced by 'que') may make the conclusion weaker. Person 5 is well represented in Category C (12.80%), but not as much as in 'que' clauses, something for which I have no explanation. And despite its

34

morphological characteristics, person 4 is represented less than what might have been expected, with 5%, but only 1 example of Spo. As for the impersonal, the contrast that exists between Category C (and in particular appositive relative clauses not introduced by 'que') and Category A, as well as the fact that, as we will see, Stylistic Inversion, resulting in a V1 construction, is possible in WH clauses introduced by 'que' while the impersonal V1 construction is excluded from this context, indicates that there is a null subject (pro) that needs content identification in standard impersonal uses, while there is no such null subject in SpecIP in the Stylistic Inversion construction (or that if there is, it does not need identification). INFL not having an active relationship with SpecIP in this case (as it relates to the expressed postverbal subject), the +WH complementizer (which is not phonetically realized in the case of indirect questions, 'que' being a WH-phrase in SpecCP, and which is realized as 'que' in C0 in the other constructions) is able to formally license the SpecIP position. This concludes our discussion of Spo in the embedded V1 construction. In the next section, we briefly examine cases where the embedded clause introducer is not immediately followed by the verb. 4. Sub XP (Subject) V One position adopted at the beginning of this paper, following insights from Roberts, is that, in embedded clauses, whether SpecIP is formally licensed depends on properties of C, and not on INFL. This was discussed in relation to the embedded V1 construction. In the analysis defended here, the V1 construction is an [IP proVX] construction. The SVX construction may be part of a larger construction where a first constituent (or a series of such constituents) is adjoined to IP, as in the following examples: (20) a … mais d'une chose je vous prie … 002045 … but one thing I ask you … b Et ce fait, il bouta ce maistre tuppin dedans le mur… 011025 And this being done, he put this large pot in the wall We saw that this construction allows null subjects, i.e. INFL or AGR licences and identifies a pro subject on its left in SpecIP, as in (21), which is similar to (1d) above. As the participial phrase is never found with a postverbal pronominal subject, it is clear that the null subject is to the left of the tensed verb (21) Et cela faict, marchanda audit poinctre de luy faire ung sepulchre 011032

35

And this done, [he] made a deal with the painter that he would paint a sepulchre The same type of constructions are found in embedded clauses, i.e. with Spr or Spo, as clearly shown in (1c) above where Spr and Spo alternate. The distribution of each construction is given in Table 13 and Table 14. with examples following each table:

36

Table 13: Sub XP Spr V (72+9=81)

A: Que Sub XP Sp V 1 2 3 31 ce 4 5 6 on purpose 1 1 cause 5 1 2 1 1 Complement 27 2 1 12 3 2 3 4 misc 1 1 concessive result 19 13 1 1 1 1 2 optative comp.-superl. 1 1 Rest.Rel.: que 2 2 total 56 6 2 28 4 1 4 5 6 B: Spec. cases Sub XP Sp V 1 2 3 31 ce 4 5 6 'quant' ('when') 3 3 Temporal phrase conditional 'si' 2 1 1 total 5 3 1 1 C: Wh ≠ 'que' Sub XP Sp V 1 2 3 31 ce 4 5 6 Q.I.: si 1 1 Q.I.: wh≠que 0 NP Appos. Rel. 6 1 4 1 CP Appos. Rel. 5 4 1 R.Rel. ≠ 'que' 6 2 3 1 'comme' ('as') 2 1 1 total 20 4 11 1 1 3 Total A, B, C 81 10 2 42 4 1 4 2 7 9 (22) a … vecy toutes les chandeilles que oncques en ma vie je acheta 091125 … here are all the candles that ever in my life I bought b Je ne sçay si depuis ilz en joyrent. 047061 I don't know if since then they enjoyed them. c … comme cy aprés vous oyrés. 020000 … as thereafter you will hear

37

Table14: Sub XP V X (with null subject)10 (335/514 = 65.17 % embedded Spo )

A: Que Total Sub X V ø 1 2 3 31 4 5 6 purpose 7 2 1 1 cause 16 10 1 5 4 Complement 114 92 1 36 43 1 4 7 misc. 13 8 1 4 2 1 concessive 7 4 2 1 1 result 47 45 1 28 8 1 7 optative 2 2 1 1 comp.-superl 6 4 2 2 Rest.Rel.: que 28 9 1 1 6 1 total 240 176 4 4 83 61 1 5 18 B: Spec. cases Total Sub X V ø 1 2 3 31 4 5 6 'quant' ('when') 13 8 1 7 Temporal phrase 4 1 1 conditional 'si' 63 29 4 15 8 2 total 80 38 0 1 4 23 0 8 2 C: Wh excl. que Total Sub X V ø 1 2 3 31 4 5 6 Q.I.: si 6 3 1 1 1 Q.I.: wh≠que 5 2 1 1 NP Appos. Rel. 42 6 4 2 CP Appos. Rel. 26 15 2 4 7 2 R.Rel. ≠ 'que' 7 4 2 1 1 'comme' ('as') 108 comme dit est 54 54 comme XP V 37 2 23 1 9 2 total 194 121 2 0 14 88 2 9 6 Total A, B, C 514 335 6 5 101 172 3 22 26 (23) a … puis que aultrement ø3 n'en pouvoit faire… 016042 … as differently [he] could not do … b … demanderent au painctre qu'i luy failloit que tant de fois ø3 les rappelloit …

087055

10 Some of the subjects are preverbal, others postverbal. See Vance (1988) for a discussion on how preverbal null subjects may have arisen in Middle French as a result of the coexistence of 'XP Spr V', 'XP V Spr' and 'XP V' with null subject.

38

… [they] asked the painter what he was missing so that so often [he] would call them back …

c … comme cy aprés ø5 orrés … 091008 /… comme cy aprés ø31 s'ensuit. 092005 … as hereafter [you] will hear /… as hereafter [it] follows d Je ne sçay se depuis ø6 en eurent quelques parolle 065111 I don't know if since [they] talked about it. e la journée vint que une fois ø3 se trouva leans aveques ses desirées amours …

041023 the day came when once [he] found himself there with his beloved f Laquelle … luy fist meilleur semblant que les aultres fois ø3 n'avoit faict.… 068012 Which (one) … was nicer to him than the other times [she] had been … pensant que bien facilement le pourroit faire … 016043 g Et fut long temps en ceste peine sans ce que jamais ø3 osait dire son piteux cas à

son mary, jayçoit que pluseurs fois s'en mist en voie 045101

And [she] was for a long time in this pain without that ever [she] dared tel her pitiful

A look at these examples and at the additional ones in the CNNV shows that the construction is found in all types of embedded clauses, i.e. complement clauses, relatives, comparatives, introduced by 'comme', etc. This is important because, in contrast to this situation, all the 25 examples of postverbal pronominal subjects in embedded clauses are found only in complement clauses, as in (24):11 (24) a … il affermoit … que se n'estoit il point. 005025 … he was swearing … that that this was he not b Les autres moinnes ont dit que voirement avoient il une geline … 005126 The other monks said that for sure had they a chicken … c Je te promés… que tout ainsi luy coupperez je la teste comme … 043052

11 These examples are also distinct from those with a null subjects in that the types of initial XPs are much more restricted. The XP preceding the verb may be an adverb or an adverbial phrase: 'si'(twice), 'ainsi' (twice), 'aussi' (twice), 'à peine' (5 times), 'voirement' (3 times), 'tout ainsi', 'en mal an', 'au moins', 'encore au debout de quinze' jours'; an object as 'se' (=ce') (2 fois) and 'moult de telles femmes'; or a combination of several phrases: "au fait de bien mentir, à cela", "par Dieu, voirement" and "s'il se courrouçait, encor".

39

I promise you … that exactly in this way will I cut his head as… d Luy … voyant … que, s'il se courrouçoit, encor pourroit il estre batus … 053088 He … seeing … that, if he became angry, in addition could he got beaten up … Along the lines of Adams (1988), examples of that type can be analyzed as embedded CPs, with the first phrase after the subordinator in SpecCP and the verb in C. These examples, with the verb governing SpecIP, allow for a null subject, but as I indicated just above, examples like those in (23) are not of the V2 type, since they are found in all types of embedded clauses, in particular clauses in which a postverbal Spr is excluded. In addition, they are often introduced by phrases that are not found with a postverbal pronominal subject, i.e. that don’t trigger V-to-C and are attested in conjunction with a preverbal pronominal subject, as can be seen by comparing some of the examples in (22) and (23). The array of facts discussed suggests that null subjects in examples of the type of (23) must be accounted for exactly like the parallel cases in main clauses, i.e. INFL is responsible for both formal licensing and identification, despite the presence of a subordinate clause marker. I assume that here INFL is able to act as the formal licenser of the embedded SpecIP because it is the only potential formal licenser for SpecIP. The first C up is not a potential formal licenser of SpecIP because it does not govern it, as the result of the intervention of the XP adjoined to IP. Given that INFL is the formal licenser of SpecIP in the construction discussed, we expect null subjects of person 1 and 2 to be found. There are 6 examples of person 1 and 5 of person 2 in the text (the fifth examples is in (1c)): (25) a je me repens fort que jamais ø1 vous feis venir ceans. 018137 I regret very much that ever [I] made you come here b nonobstant qu'en ma vie plus chandeilles ø1 n'acheta 019128 despite [the fact] that in whole my life I had not bought more candles c Et bon gré en ait Dieu … de la mesure que ainsi ø1 as oubliés 029023 … of the measure that so [you] forgot d va à tous les deable, que jamais plus ø1 ne te voie. 066046 go to the devils, that never more [I] see you e par quoy à ceste heure ø1 vous en vueil dire aucune chose 080001 as a result of which at this hour [I] I want to tell you something about it

40

f Par quoy, en acquitant ma promesse, ø1 vous dirés de celuy chareton, 082069 By which, by accomplishing my promise, [I] will tell you of this carter, (26) a Ne quel affaire a tu eu que au moins ø2 ne t'es venu confesser 013019 And what happened to you that at least [you] didn't come to confess yourselves b … quant à moy ø2 ne peust touchier. 039083 … when me [you] can not touch c et me dis … comment celle faulte que envers moy ø2 as commis t'est advenue

094044 and tell me … how this misdeed that against me [you] committed happened to you d … je te le pardonne, mais que plus ø2 n'y renchiesse 094046 … I forgive you …, as long as [you] don't fall again in it 5. Conclusion Most of the main points having been stressed several times in the text, this conclusion will be brief. – First of all, we saw that it is necessary to establish a tripartition between the various types of subordinate clause introducers according to whether they contributed to the formal licensing of SpecIP, or not. The division is not simply between elements that do and elements that do not formally license SpecIP, since we have seen that WH-'que' did not license SpecIP when it was connected to INFL (in the sense that this category was the source of the potential identification of pro in SpecIP), while it did in the Stylistic Inversion construction (where we took the position that SpecIP is not associated to INFL). – Second, we saw that it is necessary to establish a tripartition between the various persons. Person 5 is always allowed, persons 1 and 2 never (when adjacent to a complementizer), and persons 3, 31 and 6 are allowed in WH clauses. – Moreover, a detailed examination of the various types of WH clauses has shown that a null subject in a V1 construction is not equally well represented in all of them. At this point, we have no clear idea as to why this is so, in particular as to whether this is to be attributed to the way formal licensing works, or whether it follows from different considerations, having to do with discourse for example or with the author’s stylistic preferences. In conclusion, I hope that the observations that have been made have shown that in order to uncover regularities – and also to discover that some facts which appear at first

41

sight to go in one direction are, upon close examination, cases of unproductive formulas – it is important to make much more detailed descriptions than those which are usually offered in the literature, however tedious they may be at times. * This research has been made possible thanks to grants 410-87-0332 and 410-89-1131 from the Conseil de Recherches en Sciences Humaines du Canada. BIBLIOGRAPHIE Adams, M. (1987a) “From Old French to the theory of Prodrop,” Natural Language and

Linguistic Theory 5, 1-32. Adams, M.(1987b) Old French, Null Subjects, and Verb Second Phenomena. Ph.D.

dissertation, UCLA. Adams, M. (1988) “Embedded PRO,” NELS 18, GLSA, University of Massachusetts,

Amherst, 1-21. Deprez, V. (1988) “Stylistic Inversion and the Structure of COMP,” Escol Proceedings. Dupuis, F. (1988) “Pro-drop dans les subordonnées en ancien français,” Revue québécoise

de linguistique théorique et appliquée 7.3, 41-62. Dupuis, F. (1989) L'expression du sujet dans les propositions subordonnées en ancien

français, thèse doctorale, Université de Montréal . Hirschbühler, P. et M.-O. Junker (1988) “Remarques sur les sujets nuls en subordonnée en

ancien et en moyen français,” in Aspects de la syntaxe historique du français, Revue québécoise de linguistique théorique et appliquée 7.3, 63-84.

Hirschbühler, P. (1990) “La légitimation de la construction V1 à sujet nul en subordonnée dans la prose et le vers en ancien français” Revue québécoise de linguistique 19.1, 33-55

Martin, R. et M. Wilmet (1980) Syntaxe du moyen français, Bordeaux, Sobodi. Platzack, Ch. and A. Holmberg (1989) “The Role of AGR and Finiteness in Germanic VO

Languages”, Working Papers in Scandinavian Syntax, vol. 43, 51-76. Rizzi, L. (1986) “Null Objects in Italian and the Theory of pro ,” Linguistic Inquiry 17,

501-597. Roberts, I. (to appear) Verbs and Diachronic Syntax. Kluwer Santorini, B. (1990) “Variation and Change in Yiddish Subordinate clause word order,”

Ms.

42

Vance, B. (1988) Null subjects and Syntactic Change in Medieval French, Ph.D. thesis, Cornell University.

Vance, B. (1989) “The evolution of Prodop in Medieval French,” in J. de Cesaris and C. Kirschner, Studies in Romance Linguistics, 413-441, John Benjamins, Amsterdam. (also appeared as «L'évolution de Pro-drop en français médiéval», Aspects de la syntaxe historique du français, Revue québécoise de linguistique théorique et appliquée 7.3, 85-109.

Zink, G. (1987). “«Quant ce vint au congiet prendre ». De ce anaphorique à ce auto-référentiel en ancien français,” in Etudes de linguistique générale et de linguistique latine offertes en hommage à Guy Serbat, Bibliothèque de l'Information Grammaticale, Paris, 1987: 417-426.