Agreement variation in French coordinate constructions: the case of ou


Greg Lessard, Michael Levison, Mark Olsen


Linguistic Symposium on the Romance Languages XXI

University of California, Santa Barbara

February 21-24, 1991


1.0 Agreement and Coordination

Agreement phenomena have received a fair amount of attention in the past few years (cf. Barlow and Ferguson 1988). Most grammatical frameworks (LFG, GB, GPSG) propose some mechanism or other to flow agreement information around a syntactic structure. A particularly intriguing aspect of agreement occurs in the case of coordinate subject NP's, which in French, English and many other languages, trigger number, gender or person agreement in the related VP, as in [1].


[1] Le garçon et la fille jouent/*joue dans la cour.

'The boy and the girl are playing/*is playing in the yard'


Note that most research on agreement under coordination deals primarily with NP's linked by and or its equivalent in another language. Relatively little attention has been paid to disjunctive coordination, based on or or its equivalents. One of the few exceptions is Peterson 1986, who examines the strategies which underly number agreement with NPs linked by or. This is not to say that research on conjunction by and is without application here: on the contrary, we will see that strategies for number and gender resolution proposed in the light of research on and or its equivalents also apply to examples of or or its equivalents.


2.0 Semantics of ou

It is sufficient here to allude to the well-known logical distinction between exclusive and inclusive disjunction, in which the first selects one of two options, but not both, while the second selects one of two options, or both. Peterson (1986) points out that, in English, the lexical item or is sometimes ambiguous between the two logical values. French is similar in this respect, as the following examples illustrate.


[2] La pièce est tombée pile ou face.

'The coin fell heads or tails'


[3] Sont admissibles au concours des hommes ou des femmes.

'Men or women are admitted to the competition'


In what follows, we shall attempt to specify where appropriate the intended value, but in general, since we are studying cases of number disagreement, among others, we will try to produce utterances which allow both readings.


3.0 Resolution Rules

Linguists have proposed a variety of 'resolution rules' to explain how number, gender, person or case agreement is determined where more than one controller is present. A survey of these rules, or strategies, or principles reveals a remarkable variety of levels. While many of these have been formulated to deal with and-conjunction, they are in principle applicable to or-conjunction as well.


3.1 Discourse Factors

Givon 1970 discusses the problem of Bantu gender resolution, and proposes a discourse strategy to alleviate conflicts. We quote his third rule:


[4] If at least one, but not all, of the nouns to be conjoined is/are [human], the normal schema of conjunction reduction may not apply; apply the comitative alternative instead (Givon 1970:254).


In other words, the speaker simply sidesteps the difficulty by using another sort of formulation. We will see that this sort of strategy is particularly important in French.


At another level, discourse factors may determine which element is considered primary, and which is secondary. The clearest case of this occurs in instances of parenthetical additions, which are subsumed under the first element. The is likely some interplay between such discourse factors and the various ordering hierarchies surveyed by Allan 1987.


3.2 Semantics

Hirtle 1982, among many others, points out that the semantics of conjoined elements may influence agreement resolution. He quotes the following English example, whose number may vary according to the way in which the two elements are perceived - either as a unit or as two distinct entities:


[5] Bread and butter is/are nourishing.


England 1976, in his work on a corpus of Old Spanish, illustrates that the distinction between animate, concrete and abstract nouns influences the proportion of singular verb concord, when NPs are linked by e. Animate nouns, which usually designate distinct individuals, tend to favour plural concord, while abstract nouns, which are easier to conjoin in order to designate two aspects of a distinct entity, favour singular agreement.


At another level, Peterson 1986 quotes from a paper by Bresnan which calls upon semantic principles to exclude inappropriate number specification. We will see below that normative French grammar also uses this device.


3.3 Syntax

By syntax is meant here not an abstract level of derivation. As Morgan 1972 points out, agreement occurs not with logical or underlying subjects, but rather with surface subjects. In fact, in many models, syntax is reducible to distance. Corbett's (1983b) Agreement Hierarchy aside, most accounts deal with ordering, either of conjuncts, or of controllers and targets. This is clearly a consequence of the linear nature of language, whose effects go far beyond agreement under coordination. For example, Frei 1929 quotes the following example, and draws the conclusion that 'français avancé' tends to obey a strictly linear agreement pattern:


[6] Je vous assure que la mort de mon pauvre père les ont bien frappé.


Hoybye provides a wealth of similar examples. With respect to ordering of coordinate elements, three factors are frequently mentioned.


3.3.1 Proximity:

First, it is noted that agreement often occurs with the closer of the two conjuncts, as in the following example from Francis 1986:


[7] The sheer bulk of our two databases warrant hard disks.


We might call this factor proximity.


3.3.2 Directionality

Secondly, England 1976, among others, has shown that verb anteposition is associated with a preference for the singular form. In other words, agreement appears to flow left to right, but not right to left. We might call this feature directionality.


3.3.3 Agreement Calculus

A third factor operating at the syntactic level is what we might call an agreement calculus, which evaluates the conjoined elements globally and takes decisions based on the combinations of elements found. Peterson 1986, dealing with cases of number disagreement, between items joined by or, postulates a plural wins strategy. We will see below that French uses a similar strategy.


3.4 Phonology:

Pullum and Zwicky (1986) propose a 'Resolution Principle' which argues that forms which are phonologically neutral with respect to two distinct lexical items may sometimes do double duty. For example:


[8] Jean l'a frappé et (l'a /0) mis à la porte. [9] Jean l'a frappé et (lui a /*0) donné des coups de pied. [10] Jean nous a frappés et (nous a/0) mis à la porte. [11] Jean nous a frappés et (nous a/0) donné des coups de pied.


The fact that nous (direct or indirect object pronoun) presents no difference in pronunciation allows it to be deleted in [10] and [11], while the phonetic difference between les and leur prevents deletion of leur in [9].


On another level, Hoybye (1944) quotes sources in French according to which, following determiners like combien, the pronunciation of past participle forms contributes to the judgement of their acceptability:


[12] Vous avez pêché des carpes, je sais combien vous en avez pris (*prises) [13] Vous avez poursuivi des grives, je sais combien vous en avez tué (tuées)


3.5 Idiolectal Variation

Several linguists have pointed out that resolution rules appear to apply not at the sentence level, but, to use Corbett's (1983b) term, at the 'corpus' level. In other words, they reflect tendencies not necessarily shared by all speakers. Peterson 1986 talks of strategies, rather than rules. Pullum and Zwicky 1988 provide examples such as:


[14] At present the project managers, but in the past the executive directors, set the research priorities.


for which different speakers will provide different responses. Morgan 1972 goes so far as to suggest that there will be cases where an individual's strategies will simply break down, so that neither of two alternatives is considered acceptable.


[15] ??Are/??Is either John or his parents here?


3.6 Diachronic Change

The agreement picture is further complicated by changes over time. For example, Antoine (1958) claims that proximity agreement has diminished in French over the past several centuries, while England's 1976 data tends to show an increase of plural agreement from Old Spanish to Modern Spanish.


3.7 Simultaneous Factors

Furthermore, the analysis is complicated by the interference of factors among themselves. A case in point is provided by Quirk et alii 1985 who point out that where or is used for coordinative apposition, "grammatical concord requires the number of the verb agree with the first appositive if the two appositives differ in number" as in:


[16] Gobbledygook, or the circumlocutions of bureaucratic language, is (*are)

intentionally difficult to understand.


Note, however, that example [16] surreptitiously introduces another factor: the NP immediately preceding the verb (beaureaucratic language) is singular, thus providing weight for a proximity analysis. If this NP is removed, the judgement is less clearcut, as in [17], unless a strong pause is introduced.


[17] Gobbledygook, or circumlocutions, (?is/?are) difficult to understand.


4.0 Data

In what follows, we will examine the situation in contemporary French with respect to number and gender agreement in VPs having conjoined subject NPs linked by ou. There are essentially three sources of data available in this area: normative grammar, corpus data, and metalinguistic judgements.


4.1 Normative grammar:

In the case of singular subject NPs linked by ou, Grevisse 1980:981 recognizes two options. Where it is the idea of conjunction which dominates, the VP is made plural:


[18] Le bonheur ou le conseil d'autrui peuvent préserver de certaines fautes. [19] La peur ou la misère ont fait commettre bien des crimes.


Where the idea of disjunction or opposition dominates, agreement occurs with the nearest element:


[20] L'affection ou la haine change la justice de face. [21] La douceur ou la violence en viendra à bout.


On the other hand, (1980:982) where one of the subject NPs is plural, the VP is made plural:


[22] Les menaces ou la douceur en viendront à bout.


In the case of past participle agreement (1980:923), a similar choice exists between global agreement:


[23] Quel péril ou quelle menace n'a-t-il pas affrontés. [24] La peur ou la misère, que les moralistes ont considérées comme restreignant.


and agreement by proximity, in cases of clearcut disjunction:


[25] C'est sa perte ou son salut qu'il a risqué. [26] C'est son salut ou sa perte qu'il a risquée.


In the case of gender, agreement by proximity appears to favour equally the masculine and the feminine. However, in the case of past participle agreement, Grevisse (1980:924) claims that many examples are those of "grammairien vétilleux" and that usage often leaves the past participle invariable.


To conclude, with some simplifications (such as excluding apposition), the resolution rules inherent in Grevisse are the following:


[27] Normative Strategies for Number and Gender Resolution a) If both NPs are singular, use a plural VP to mark global meaning, and a

singular VP to mark disjunctive meaning.


b) If at least one NP is plural, make the VP plural.


c) If both NPs share the same gender, give the VP that gender.


d) If one NP is masculine and one is feminine, use a masculine VP to mark

global meaning, and a feminine VP to mark disjunctive meaning. Place the

NP which matches the VP in gender adjacent to the VP.


It is clear that the rules are asymmetrical, with gender marking offering greater freedom than number marking.


4.2 Corpus Data

Hoybye (1944) recognize the existence of significant variation in speaker preferences in the case of or-coordination. Little work has been done, however, to uncover more than anecdotal evidence for current usage.


In an attempt to expand the base of data, we decided to draw upon the ARTFL textbase in Chicago. A joint project of the CNRS and the University of Chicago, ARTFL is based on the machine readable corpus underlying the Trésor de la langue française, and contains 115 million word tokens primarily from the 18th, 19th and 20th centuries, with some texts from earlier periods. The approximately 2000 texts include a variety of genres, including novels, verse, journalism, essays, correspondence and treatises. Subjects range from literature to literary criticism, biology, history, economics and philosophy.


Since ou is part of the stopword list of ARTFL and thus inaccessible directly, the third author wrote a routine to print to a file all occurrences for the period 1950-1964, a total of 8627 tokens. The Oxford Concordance Program was used to produce a right-sorted KWIC concordance of all tokens. Since the task of searching all contexts would have taken many weeks, a subset of the corpus was defined. This subset includes all occurrences of ou in the 1950-64 subcorpus followed immediately by any of the following determiners:


[28] Definite article: le, la, les, l'

Indefinite article: un, une, des

Partitive article: de l', du, de la, des

Demonstrative adjective: ce, cet, cette, ces

Possessive adjective: mon, ma, mes, ton, ta, tes, son, sa, ses,

notre, nos, votre, vos, leur, leurs


The resulting occurrences were then searched by hand for cases of NP VP agreement in number or gender. It should be noted, though, that several constraints apply. First, the width of the concordance output (55 characters on either side of ou) makes inaccessible some cases of long-distance agreement. Second, nouns not preceded by any determiner which nevertheless agree, as in the following example:


[29] ...surfaces ou volumes conçus comme un ordre nouveau...


are excluded from the counts given. Note also that we have excluded from consideration cases where a single determiner precedes both members of an NP:


[30] les villes ou agglomérations qui touchent au Rhin


Curat 1986 has demonstrated that such examples represent, in fact, a single NP rather than two. Similarly, we have excluded cases of determiners linked by ou, as in:


[31] le ou les commandant(s)


The results of the count are as follows:


[32] Marked VP number agreement for ARTFL data 1950-64. Based on NPs with determiners.



83







***








***








***








***








***








***

53







***

***







***

***

46






***

***

***






***

***

***






***

***

***






***

***

***






***

***

***






***

***

***






***

***

***

10





***

***

***

***

1

0

0

0

SSS

SSP

PPP

SPP

PSP

SPS

PPS

PSS


Code: SSS=NP(sing)+NP(sing)+VP(sing), SSP=NP(sing)+NP(sing)+VP(plur), etc.


A random slice of the ARTFL textbase for 1900-1949 data was treated in a similar fashion, and gave comparable results.


The first observation is that given singular NPs linked by ou, the preferred choice of number for the VP is singular. According to Grevisse, this should represent agreement by proximity, and presupposes opposition between the two NPs. And yet, this idea of opposition is seriously weakened by examples like the following:


[33] le physicien ou l'analyste qui prétend à la renommée

[34] un récit de chasse ou un roman policier conviendrait mieux à votre humeur


The second observation is that there is much less tendency to link unlike number NPs (11 cases in all). Where this does occur, plural resolution occurs, and in 10 of the 11 cases, the plural element immediately precedes the VP.


With respect to gender agreement, the 1950-1964 corpus provided the following results:


[35] Marked VP gender agreement for ARTFL data 1950-64. Based on NPs with determiners.


19







***







***

17







***

***







***

***







***

***







***

***







***

***

12






***

***

***






***

***

***






***

***

***






***

***

***






***

***

***






***

***

***

6





***

***

***

***





***

***

***

***





***

***

***

***





***

***

***

***





***

***

***

***

0

0

0

0

MMM

FFF

FMM

MFM

FMF

MFF

FFM

MMF


Code: MMM=NP(masc)+NP(masc)+VP(masc), MMF=NP(masc)+NP(masc)+VP(fem), etc.


Note first that the number of occurrences is much lower than was the case in number agreement. Second, there is a strong tendency to same-gender linking, which obviously avoids any problem of gender conflict. Third, where gender conflict does occur, resolution is to the masculine VP, and of the 18 examples found, 12 use the order which puts the NP sharing the VP gender next to the VP. We found no examples of feminine VP resolution, such as Grevisse had mentioned, with the second NP next to the VP.


It is important to note that we have counted cases of explicit gender marking. In fact, a common strategy in the corpus consists in using ambiguous forms in the VP, which make no difference between masculine and feminine agreement:


[36] l'entrepreneur ou la firme dynamiques


At a more general level, the distribution of the various possibilities shows clearly that writers in the corpus appear to adopt a more general strategy for number and gender resolution, which may be formulated as follows:


[37] Corpus Strategies for Number and Gender Resolution

a) Avoid number or gender conflict.

b) Hide conflict by use of ambiguous lexical items.

c) If number conflict occurs, resolve to the plural; if gender conflict

occurs, resolve to the masculine.

d) Where possible, place unlike number or gender NPs such that the NP

which shares features with the VP is closest to the VP.


4.3 Metalinguistic Judgements

4.3.1 Acceptability Judgement Task

Based on Russian data, Corbett (1983b:34-35) has demonstrated that normativity (preference for syntactic as opposed to semantic agreement) increases with education, and that those who use language professionally (such as writers) tend to be the most conservative of all. Given the primarily literary nature of the ARTFL corpus, it is therefore not surprising to find little tendency to use other than plural VPs following mixed-number subject NPs linked by ou.


In an attempt to evaluate this conservative tendency more closely, we presented a short questionnaire on number and gender conflict between linked subject NPs to a group of francophone university professors and graduate students in literary studies. Several sample questions follow:


[38] L'avocat ou la coiffeuse sera choisi.

L'avocat ou les chimistes seront choisis.


The format adopted was an acceptability judgement task, with subjects declaring sentences to be acceptable (bonne), unacceptable (mauvaise) or unevaluable (Ne sais pas). Although the number of subjects was small, two tendencies stood out. First, the normative rule was almost universally respected, according to which in cases of singular and plural NPs linked by ou, the VP is plural. Similar results applied when masculine and feminine NPs were linked by ou, in that preference was given to masculine VP agreement. It is clear that the subjects tested were applying consciously a normative strategy: when asked to comment on the questionnaire, several alluded to the explicit rule.


Second, and more importantly, many subjects commented that the format of the test sentences, which explicitly raised the agreement problem, was itself unacceptable. Several typical reactions are included below:


[39] La plupart de ces énoncés seraient soigneusement évités (évitée) par un

francophone.


[40] ...la plupart de ces phrases sont pour moi maladroits, à éviter, puisqu'un

accord par le sens est impossible.


Some subjects simply declared that all test sentences were bad.


Taken together with the ARTFL data, these tests suggest that sophisticated users of French simply do not face the problems we are discussing, to the extent that they are capable of avoiding structures and forms which explicitly recognize and resolve number or gender conflict. This is similar to the rules proposed by Givon (1970) to deal with problems of Bantu gender conflict.


4.3.2 Forced Choice Task

Given the tendency of (at least) more highly educated subjects to simply avoid the problem of number or gender conflict, we adopted an alternative strategy. Following the model of Peterson's 1986 work on or in English, we produced a series of three questionnaires on number and gender resolution. The model used was a forced choice task, with both options being presented in each case:


[41] Selon sa biographie, les croyances religieuses ou la spiritualité a/ont très peu marqué ses options politiques.


[42] De toute évidence, les réactions ou la réponse publique de l'équipe nationale pourra/pourront déterminer la direction de cette dernière dans les années à venir.


and subjects being asked to accept one or the other, accept both, or refuse both. Subjects were requested to use a quick intuitive judgement and to refrain from consulting previous responses.


We made three principal changes to the Peterson model. First, whereas Peterson attempted to measure number resolution in very simple sentences:


[43] A rabbit or a goat has/have eaten all my lettuces.


we used more complex sentences, typically containing an initial PP, whose lexicon and syntax resembled journalistic prose.


Second, whereas Peterson used occurrences of or in which the disjunctive meaning was clearly dominant, we chose to produce utterances ambiguous between the two meanings. (See above examples). Finally, we used exclusively declarative sentences, as opposed to a mix of declarative and interrogative sentences.


The revised version of the questionnaire was submitted to undergraduate francophone university students at the Université de Sherbrooke, in Quebec. The first questionnaire, dealing with number, was completed by 39 subjects, while the second, dealing with gender, was completed by 49 subjects.


4.4.2.1 First Questionnaire: Conjoined Singular NPs In the first questionnaire, the object of study is NPs which share the same number (the singular). We observed in the corpus data that in this case, there is a global preference for VP number resolution in favour of the singular. In the 8 questions with linked singular NPs, results were as follows, for 39 subjects:


[44] Global Responses: Conjoined Singular NPs


VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

219 (70)

70 (22)

22 (8)

1 (0)

312 (100)


Code: s=singular, p=plural, sp=singular or plural, 0=neither singular nor plural, N/R=number of responses, %=percentage of total responses.


These results confirm the corpus data, that in the case of conjoined singular NPs linked by ou, resolution strongly favours the singular VP. This disproportion obviously merits detailed analysis. For the moment, let us examine two possible factors which might influence the strength of this tendency, beginning with the animate/abstract distinction on subject NPs.


[45] Responses (animate/abstract NPs)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP anim

120 (77)

25 (16)

10 (7)

1 (0)

156 (100)

NP abstr

99 (63)

45 (29)

12 (8)

0 (0)

156 (100)


Surprisingly, animate subject NPs favour the singular more than do abstract subject NPs. This runs counter to the results obtained by England (1976) for NPs linked by e. Although the difference is small, it merits further study.


The second factor examined was the position of the VP, either anteposed or postposed to the NPs.


[46] Responses (Postposed/Anteposed VP)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

VP postposed

95 (61)

50 (32)

11 (7)

0 (0)

156 (100)

VP anteposed

124 (80)

20 (13)

11 (7)

1 (0)

156 (100)


In this case, the anteposed VP favours singular resolution more than does a postposed VP, in accordance with predictions by England 1976 and others.


4.4.2.2 Second Questionnaire: Conjoined Singular/Plural NPs The second questionnaire maintains the animate/abstract and the verb initial or verb final distinctions, while adding an additional factor: in the examples given, the linked subject NPs differ in number, with the singular NP preceding in half the cases, and the plural NP preceding in the other half. The result is a 16 item test. The first and second questionnaires are presented on the same 24 item form, and the order of questions randomized.


In this case, given that the default rule is use of the plural on the VP, we will be primarily interested in cases where the singular appears. Of course the first question is simply whether in fact the singular does appear, contrary to grammatical norms. In the 16 questions characterized by conjoined singular and plural NPs, results were as follows, for 39 subjects:


[47] Global Responses: Conjoined Singular/Plural NPs


VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

123 (20)

458 (73)

39 (6)

4 (1)

624 (100)


Clearly, the normative grammar rule which predicts that all 624 VPs should be plural is not applying across the board in the case of these subjects. The question is then which of the factors analyzed support the tendency to use the singular VP. We begin with the animate factor.


[48] Responses (Animate/Abstract NPs)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP anim

67 (22)

226 (72)

16 (5)

3 (1)

312 (100)

NP abstr

56 (18)

232 (75)

23 (7)

1 (0)

312 (100)


Once again, animate subject NPs favour VP singular resolution slightly more than do abstract subject NPs, although the difference is minuscule. If we turn to the order of NPs and VPs, the following results are found:


[49] Responses (Postposed/Anteposed VP)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

VP postposed

41 (13)

247 (79)

22 (7)

2 (1)

312 (100)

VP anteposed

82 (26)

211 (68)

17 (5)

2 (1)

312 (100)


In was the case earlier, the anteposed VP favours singular resolution more than does a postposed VP. The final factor is the order of NPs, whether singular-plural or plural-singular. The overall results are as follows:


[50] Responses (Order of NPs)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP(s)-NP(p)

57 (18)

234 (75)

17 (5)

4 (2)

312 (100)

NP(p)-NP(s)

66 (21)

224 (72)

22 (7)

0 (0)

312 (100)


We see that the plural-singular order favours singular agreement slightly more than does the singular-plural order. In fact, though, this way of expressing things misses a significant factor. Linguistically, the order of NPs is important only to the extent that it places a particular NP closer to the agreeing VP. So in fact, it is important to separate out two classes: that where the NP(p)-NP(s) order occurs with a postposed VP and that where it occurs with an anteposed VP. Let us begin with the first:


[51] Responses (Order of NPs, with Postposed VP)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP(s)-NP(p)

10 (7)

137 (88)

7 (4)

2 (1)

156 (100)

NP(p)-NP(s)

31 (20)

110 (70)

15 (10)

0 (0)

156 (100)


Where the singular NP and the VP are in close proximity, the number of cases of singular VPs is three times higher then when this proximity is not present. Let us turn now to cases of anteposed VPs:


[52] Responses (Order of NPs, with Anteposed VP)


Factor

VP(s)

VP(p)

VP(sp)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP(s)-NP(p)

47 (30)

97 (62)

10 (7)

2 (1)

156 (100)

NP(p)-NP(s)

35 (23)

114 (73)

7 (4)

0 (0)

156 (100)


We can see, then, that if we measure proximity to the VP, as opposed to simple order, the number of cases of singular agreement rises to 78, versus 45 cases of plural agreement. This confirms the tendencies discussed earlier.


4.4.2.3 Third Questionnaire: Conjoined Masculine/Feminine NPs The third questionnaire deals with gender resolution in the VP, given two NPs linked by ou. Again, the factor of animate/abstract NPs is examined. However, since only the masculine/feminine distinction is perceptible orally:


[53] La petite fille était surprise. [syrpriz]

Le petit garçon était surpris. [syrpri]

Les petites filles étaient surprises. [syrpriz]

Les petits garçons étaient surpris. [syrpri]


(although only in a few lexical items) we chose to vary the gender of the NPs rather than the number. Similarly, since Hoybye had suggested that the oral perceptibility of gender marking influences agreement, we chose to vary the VPs according to the oral marking of the gender distinction. The fact that number and gender agreement on the VP is marked only in the case of anteposed NPs is another reason for dealing with a different variable. The third questionnaire appears on a separate form, again with randomized ordering of questions.


Note to begin with that we will have nothing to say about the syntactic analysis of past participle agreement, as discussed by Kayne 1985, 1989, Bouchard 1987, etc. The fact of agreement itself is sufficient for our present purposes.


Given that the default global agreement in gender would produce a masculine VP, we will be primarily interested in cases of feminine VP. For the 16 questions characterized by conjoined masculine and feminine NPs, results were as follows, for 49 subjects:


[54] Global Responses: Conjoined Masculine/Feminine NPs


VP(f)

VP(m)

VP(mf)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

212 (27)

502 (64)

26 (3)

44 (6)

784 (100)


Code: f=feminine, m=masculine, mf=masculine or feminine, 0=neither masculine nor feminine, N/R=number of responses, %=percentage of all responses


We noted above that according to Grevisse, past participle gender agreement may occur with both conjuncts (which favours the masculine if one is masculine) or with the last conjunct. Let us examine the factors in play, beginning with proximity:


[55] Responses (Order of NPs)


Factor

VP(f)

VP(m)

VP(mf)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP(m)-NP(f)

194 (49)

152 (39)

18 (5)

28 (7)

392 (100)

NP(f)-NP(m)

18 (5)

350 (89)

8 (2)

16 (4)

392 (100)


Clearly, proximity to the VP plays an important role. There is little support for a feminine VP where the preceding NP is masculine. Let us examine now the pronounced/unpronounced factor.


[56] Responses (Phonetically Marked/Unmarked VP Agreement)


Factor

VP(f)

VP(m)

VP(mf)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

Marked

98 (25)

265 (68)

8 (2)

21 (5)

392 (100)

Unmarked

114 (29)

237 (60)

18 (5)

23 (6)

392 (100)


We note a small difference in favour of the masculine. Finally, let us examine the animate/abstract distinction.


[57] Responses (Animate/Abstract NPs)


Factor

VP(f)

VP(m)

VP(mf)

VP(0)

Total

N/R (%)

N/R (%)

N/R (%)

N/R (%)

N/R (%)

NP anim

84 (21)

271 (69)

19 (5)

18 (5)

392 (100)

NP abstr

128 (33)

231 (59)

7 (2)

26 (6)

392 (100)


Here, the abstract form favours VP feminine agreement.


5.0 Conclusions, Modelling and Future Problems


The data examined seems to show that number and gender resolution occurs at a variety of levels, from discourse strategies (avoid conflicts), to semantics (abstract/animate distinction) to linear phenomena (proximity and directionality), to phonetic factors.


Second, despite claims made by Antoine, and the rules of normative grammar, the linearity/proximity principle principle appears to be alive and well in modern French.


Third, the semantics of conjoined singular NPs remains to be worked out. Quite clearly, the simple model which associates singular VP agreement with disjunctive meaning and plural agreement with global meaning is not adequate. It remains to be shown why the singular predominates as strongly as it does.


Fourth, there are several aspects of the study which require further development. The descriptive statistics provided here must be analyzed more rigourously for significance (in progress), and in particular, the questionnaire responses must be analyzed to determine whether there exist separate populations with respect to number and gender resolution strategies.


6.0 References

Allan, K. 1987. Hierarchies and the choice of left conjuncts (with particular attention to English). Journal of Linguistics 23:51-77.


Antoine, G. 1958. La coordination en français. Paris: Editions d'Artrey.


Barlow, M., Ferguson, C.A. (eds.) 1988. Agreement in Natural Languages. Stanford: Center for the Study of Language and Information.


Bouchard, D. 1987. A Few Remarks on Past Participle Agreement. Linguistics and Philosophy 10/4:449-474.


Corbett, G. 1979. The agreement hierarchy. Journal of Linguistics 15:203-224.


Corbett, G. 1983a. Resolution rules: agreement in person, number and gender. In Gazdar, G., Klein, E., Pullum, G. (eds.), Order, Concord and Constituency. Dordrecht: Foris, pp. 175-206.


Corbett, G. 1983b. Hierarchies, Targets and Controllers: Agreement Patterns in Slavic. London: Croom Helm.


Corbett, G. 1986. Agreement: a partial specification, based on Slavic data. Linguistics 24:995-1023.


Curat, H. 1986. Référence, accord et formation de syntagme. Langues et linguistique 12:1-26.


England, J. 1976. 'Dixo Rachel e Vidas': Subject-Verb Agreement in Old Spanish. Modern Language Review 71:812-826.


Farkas, D., Ojeda, A. 1983. Agreement and coordinate NPs. Linguistics 21:659-673.


Francis, W.N. 1986. Proximity Concord in English. Journal of the English Language 19/2:309-317.


Frei, H. 1929. La grammaire des fautes. Paris.


Grevisse, M. 1980. Le bon usage. Paris: Duculot. (11e édition)


Hirtle, W. 1982. The Singular Plurality of Verb Discord in English. Canadian Journal of Linguistics 27/1:47-54.


Hoybye, P. 1944. L'accord en français contemporain. Copenhague: Andr. Fred. Host & Sons Forlag.


Kayne, R. 1985. L'accord du participe passé en français et en italien. Modèles linguistiques 7/1:73-89.


Kayne, R. 1989. Facets of Romance Past Participle Agreement. In Benincà, P. (ed.) Dialect Variation and the Theory of Grammar. Dordrecht: Foris. Pp. 87-103.


van Oirsouw, R. 1987. The Syntax of Coordination. London: Croom Helm.


Peterson, P. 1986. Establishing verb agreement with disjunctively conjoined subjects: strategies vs principles. Australian Journal of Linguistics 6:231-249.


Pullum, G., Zwicky, A. 1986. Phonological Resolution of Syntactic Feature Conflict. Language 62:751-773.


Quirk, R., Greenbaum, S., Leech, G., Svartvik, J. 1985. A Comprehensive Grammar of the English Language. London: Longman.