Navigation – Plan du site

Variations of chronometric measures of written production depending on clause packaging

Emilie Ailhaud et Florence Chenu

Résumé

A crucial way of revealing cognitive process at work during language use is by studying the temporal characteristics of language production. During written text production, the writer engages in many cognitive processes, such as planning, translating or reviewing (Hayes & Flower 1980). S/He has to write sentences and clauses which are linked, in order to constitute a coherent discourse. In our study, thirty undergraduate French students were asked to continue a fictive story. The text was written on a graphic tablet, connected to a computer. The Eye&Pen software (Chesnet & Alamargot 2005) was used to provide chronometric measures, such as the initial pause of each word. We hypothesized that the clause initial pause would depend on the degree of dependence between clauses: the more dependent the clause is, the shorter the initial pause would be. Results corroborate this hypothesis: the initial pause duration is affected by the clause type, suggesting that subordinated clauses start to be planned at the beginning of the matrix clause. A similar result was obtained when the clause required planning at the semantic level: pauses are shorter before restrictive relatives than before explicatives. We argue that the long pauses at the beginning of sentences can be explained by the planning of several dependent clauses.

Haut de page

Entrées d’index

Haut de page

Texte intégral

1. Introduction

  • 1 “subjects are asked to say aloud everything they think and everything that occurs to them while per (...)

1Much can be discovered about text production by examining texts as they are offered to the reader but it is also crucial to know about how texts are created. Text genetics examines the traces that writers leave on their texts (de Biasi 2002; Grésillon 2002; Lebrave 2006) in order to capture a better understanding of strategies used in literary production (e.g. Herschberg Pierrot 2012; Fenoglio 2009; Ferrand 2012), as well as, those used by professional and non-professional writers (Alamargot & Lebrave 2010) and those used by young writers during the development of text production (Fabre 1990; Plane 2003; Doquet 2003). To know more about the elaboration of a text, writing researchers have developed a variety of on-line research methods (Fayol 1997; Favart & Olive 2005). For example, Hayes & Flower (1980), who originated probably the most influential model of writing, adapted think aloud protocols1, primarily used to identify processes in problem solving tasks, to identify processes during writing. They identified three major writing processes: planning, translating and reviewing which are not to be considered as successive stages, and are refined in further versions of the model (e.g. Hayes & Nash 1996; Hayes 2012). These cognitive processes at work during written language production can be further explored by studying the temporal characteristics of language production. Since the availability of video recording (Foulin 1993) and subsequently digital recording (with, for example, softwares recording keyboard activity, Strömqvist et al. 2006; Leijten & van Waes 2013; van Waes et al. 2012; or software for recording handwriting activity on digitizing tablets, Chesnet & Alamargot 2005; Alamargot et al. 2006), researchers have access not only to the order in which writing activities occur but also to the precise timing of these activities (e.g. writing rates, pause durations) as well as eye movements occurring during writing (Alamargot et al. 2007; Torrance et al. 2015).

2Little is known about the temporal management of the conceptual and linguistic preparation of sentences: when ideas are selected, when lexical and syntactic choices are processed to insure coherence and cohesion. In the case of a multi-sentence text, it is first necessary to ascertain if this preparation occurs at the entire sentence level or more incrementally, and if it is only during pauses or also during motor execution. To this aim the present study analyses the link between certain linguistic features which characterize clause packaging and chronometric data (pause durations and time used to write the clause).

3We first introduce the issues involved in the study of chronometric measures for understanding writing processes, then we present the methodology used to collect the data and to analyse it. Finally, we discuss our results which reveal the link between chronometric measures and clause packaging.

2. The production of written texts

4Part of the originality of our work resides in the fact that we consider the temporal dynamics of handwriting in the framework of text production. Writing a text is a complex activity that involves setting objectives, imagining a recipient, forming ideas and translating them into linguistic forms, managing grapho-motor gestures, reading the text produced so far and comparing its computed effect on the reader to the intended one, etc. Writing a text requires writing words arranged in clauses that are organized in sentences which are linked and contribute to text parts, in order to produce a coherent discourse that will be understood by the reader and have the intended effect on him/her. Hence, during text production, the writer engages in many cognitive processes.

5Text quality and quantity depends on many choices made by the writer, concerning ideas and linguistic items (Beauvais et al. 2011; Beers & Nagy 2009). In this paper, we will use the term planning to refer to these processes, i.e. conceptualization (forming ideas, retrieving information from memory, structuration) and formulation (syntactic and lexical choices). Thus, a full understanding of writing requires observation of where and how planning occurs in real time.

2.1. Chronometric measures as manifestations of cognitive processes during written language production

6Analysis of pauses is one of the preferred methods for accessing the dynamics of writing and is based on the idea that pauses are behavioral correlates of cognitive processes (Schilperoord 1996). However, there are other behavioral correlates of cognitive processes. Behavioral manifestations of handwriting are scriptural activity (characterized by movements and pressure of the pen, writing rates), sometimes referred to as writing bursts (Alves et al. 2012; Alves et al. 2008) and scriptural inactivity (pauses). The variables that one can examine are diverse and include linguistic data (what was written, what is being or will be written) and their characteristics (type, frequency, etc.), and this in relation to chronometric measures (Plane et al. 2010) such as pauses, writing rates, pressure of the pen (e.g. location of these phenomena in relation to linguistic units).

7The view that pauses, as well as pen movements, are considered as behavioral correlates of cognitive processes, is to be taken carefully, as pointed out in Chenu et al. (2014), given that a particular cognitive process may be associated with different behavioral correlates, and a particular behavior might display different cognitive processes (Schilperoord 2002). This means that the relationship between cognitive processes and their behavioral correlates is complex and moreover, that those behaviors are not always correlated to the envisioned cognitive processes. We know, for instance, that a pause occurring during the writing of a word is not necessarily related to the processing of this word but could, for example, be associated to an anterior activity (e.g. fatigue, revision), or a posterior activity (anticipation). In addition, cognitive processes, whose behavioral correlates are observed, may well be different than what theories of writing have uncovered (e.g. Hayes & Flower 1980; Levelt 1989) and they may or may not be related to the task (e.g. related to the task, the writer may be rereading the text produced so far, see Alamargot 2005; Chesnet & Alamargot 2005; Alamargot et al. 2006; Alamargot et al. 2007, and not related to the task, the writer could be thinking of something else). Finally, elements in the data that are apparently pauses, accelerations or decelerations can not always be considered as correlates of cognitive processes involved in writing: for instance, when the writer moves the pen for a line return, or when s/he experiences muscle fatigue, cramps, etc. Thus, we should keep in mind that chronometric measures have to be taken with caution, as they can reflect cognitive processes related to the task or not, and they can reflect cognitive, physical or socio-psychological causes (Schilperoord 2002, following de Beaugrande 1984).

8Two approaches can be considered for the study of chronometric measures in written text production. A first approach, driven by temporality, defines thresholds and observes where temporal phenomena (pauses, accelerations, etc.) occur. A second approach, linguistically driven, defines structural units and examines the chronometric measures at the boundaries of these units or within these units. In the temporally driven approach the units are not defined a priori but are expected to emerge, while the linguistically driven approach, which avoids the thorny definition of a threshold, is confronted with the problem of the definition of a unit and cannot investigate whether or not a processing unit emerges from the behavioral traces of writing processes. In this study, we take the second approach: avoiding the definition of a threshold, we gather chronometric measures (pause durations at the boundaries of selected linguistic units, i.e. clauses, clause packages, sentences and the time spent to write these units).

9Chronometric measures here, as behavioral correlates of cognitive processes, are taken into account in the framework of the limited capacity theory (Just & Carpenter 1992) adapted to written language production by McCutchen (1996). In this view, cognitive resources are limited and hence are shared between processes. The cost of a process (the amount of resources it takes) depends on its degree of automaticity. For instance, in the case of writing, if low level processes involved in grapho-motor execution, for example, are not strongly automatized, they weigh on the system and prevent adequate operation at higher levels. Hence, our hypotheses will assume that the higher the cost of a cognitive process (e.g. the more the writer has to plan), the longer the pause and the time used to write; we will try to explain the variation of chronometric measures by a variation in cognitive cost.

2.2. The study of planning scope in writing

10Some studies used chronometric measures with a linguistically driven approach to explore cognitive processes involved in the written production of a text, either at a local level (a word and its context), either at a more global level (structure of the text). At a local level, the study of Maggio et al. (2012) examined chronometric data (pauses before and inside words) in relation to lexical and infra-lexical characteristics of words in texts and helped to uncover immediate, delayed and anticipated influences (i.e. the chronometric measures of a word have been shown to be dependent on the lexical and infralexical characteristics of the preceding and the following words). In this study, words preceded by a punctuation mark were put aside, since the authors needed to use lexical and infra-lexical characteristics of preceding word, which means that words at the beginnings of sentences were ignored. Thus, this raises the question of the management of higher level writing processes which may occur at sentence boundaries.

11At a more global level, previous studies have found a correlation between the pause length before the unit and the size of the unit: the larger the unit, the longer the initial pause (Chanquoy et al. 1996; Foulin 1998; Schilperoord & Sanders 1999). Indeed pause durations followed this hierarchy: paragraph > sentence > clause > phrase > word. Thus, a hierarchical model has been substituted to the first word-by-word production models, as pointed out by Matsuhashi (1982), when she reviewed previous studies on speech production. Subsequently, it has been proposed that the more dependent a clause, the shorter the initial pause, since juxtaposed clauses are preceded by longer pauses than coordinate clauses and subordinate clauses (Immonen & Mäkisalo 2010; van Hell et al. 2008). Ailhaud et al. (2016) examined chronometric data as a function of clause packaging for identifying units of production in a developmental perspective. Analysis of chronometric data (pauses and writing rates) revealed that younger writers (10-11-year-olds) seem to plan their texts clause by clause, but that older writers (15-16-year-olds) plan more at a multi-clausal level.

12In the present study, our objective is to identify how and where planning occurs in multi-sentence narrative texts produced by French speaking young adults.

3. Methods

3.1. Data collection

13Thirty undergraduate French students were asked to complete a story; the title and the first sentences of the story were given, as illustrated in Figure 1 (with its translation). The story deals with a prisoner who attempts an escape.

Figure 1. Beginning of the story which had to be continued by the participants (and its translation in English)

L’évasion

Dimitri regardait le ciel étoilé à travers les barreaux. Il devinait les ombres des gardiens dans le mirador. Il entendit la porte de sécurité s'ouvrir dans le couloir. Le gardien ouvrit la porte de sa cellule. Il passa alors à l'action.

‘The escape

Dimitri was looking at the starry sky through the window bars. He could see the shadows of the guards in the watchtower. He heard the security door open in the corridor. The guard opened the door of his cell. He then took action.’

14The participants were asked to write their text on a graphic tablet, connected to a computer. The Eye&Pen software (Chesnet & Alamargot 2005) was used to obtain chronometric measures. We then elaborated a data processing script to obtain, for each word, the initial pause and the time spent to write the word.

3.2. Coding

15We first segmented the texts using three units: sentences, A-units, clauses. “Sentences” were graphically defined as a unit which begins with a capital letter and is finished by a full stop. Autonomous units, that will be called “A-units”, were syntactically defined as a main clause plus any clause subordinated to this clause. “Clause” are defined as units that contain a unified predicate (Berman & Slobin 1994). In the example (1) there is one sentence composed of two A-units delimited by “//”; the first A-unit is composed of two clauses delimited by “/”.

  • 2 The code at the end of each example identifies the participant. 

(1)

Dimitri s’empara alors du bandeau de tissu / qu’il avait déchiré sur le bas de son drap la nuit précédente, // il en entoura la tête du gardien par deux fois au niveau de la bouche. [S15]2

‘Dimitri then took the strip of cloth / that he had torn off the bottom of his sheet the night before, // he wrapped the head of the guard twice at the level of the mouth.’

16This segmentation was performed by one main coder and 10% of the corpus was coded by a secondary coder to assess coding liability. Inter-coder agreement for this 10% reaches 92% agreement. An example of a full text, segmented for sentences, A-Units and clauses is provided in Annex A.

17Our coding provides a clause packaging coding for each clause. First, at the syntactic level, we took into account 3 features which are each illustrated below by an example.

  • The clause position within the graphical sentence: we distinguished between the first clause (in bold in example (2)) and the following clauses.

(2)

Dimitri s’empara alors du bandeau de tissu / qu’il avait déchiré sur le bas de son drap la nuit précédente, // il en entoura la tête du gardien par deux fois au niveau de la bouche. [S15]

Dimitri then took the strip of cloth / that he had torn off the bottom of his sheet the night before, // he wrapped the head of the guard twice at the level of the mouth.

  • The clause type: we distinguished between juxtaposed (3), coordinated (4) and subordinate clauses; the subordinate clauses could occur before (5) or after (6) the matrix clause.

(3)

Mais Dimitri était très intelligent, / il avait pensé à tout. [S04]

‘But Dimitri was very smart, / he had thought of everything’

(4)

Alors Dimitri poussa de toutes ses forces / et la grille s’arracha. [S24]

Then Dimitri pushed with all his strength / and tore the grid off

(5)

Afin de ne pas être capturé une nouvelle fois /, il s’enfuit à travers la ville. [S24]

So as not to be caught again / he fled through the city’

(6)

Il descendit discrètement du camion / lorsque celui-ci s’arrêta. [S19]

‘He discreetly got off the truck / just when it stopped’

  • Matrix or non matrix clauses: we identified matrix clauses, which are clauses followed by a subordinate clause (7). The matrix clause can be subordinated, as illustrated by the second bold clause of the example (8).

(7)

Et Dimitri sortit / pour aller diner. [S10]

And Dimitri went out / in order to eat’

(8)

Il rit de ce même rire / qu’entendait Dimitri durant les séances / où on le torturait. [S30]

‘He laughed that same laugh / that Dimitri had heard during the sessions / in which he was tortured’

18Finally, the coding distinguished between 3 types of subordinate clauses relative to the necessity of their mention from a syntactic, semantic or discursive perspective. We made one category with clauses which are obligatory to complete the valence of the predicate (OV) (9) or to delimitate the referent (OR) (10), a second category for the clauses which are obligatory from a communicational perspective (OC) (11) and a third category for the clauses which are optional (FC) (12). This distinction is particularly useful to identify defining relative clauses (10), narrative relative clauses (11) and explicative relative clauses (12).

(9)

Dimitri vit / que les étoiles avaient disparu [S03]

‘Dimitri saw / that the stars had disappeared.’

(10)

Sa main gauche était toujours bandée depuis la dernière altercation / qu’il avait eue avec Dimitri [S03]

‘His left hand was still bandaged since the last altercation / that he had had with Dimitri’

(11)

Dimitri assomma le gardien / qui tomba évanoui [S15]

‘Dimitri knocked the guard / who fainted.’

(12)

Et ils partirent tous les deux en bateau pour la Crête / où Dimitri possédait une ferme léguée par son parrain [S19]

‘And they went both by boat to Crete / where Dimitri had a farm bequeathed by his godfather.’

19These linguistic features will constitute our independent variables while dependent variables will be the chronometric measures: the initial pauses before clauses and the total duration of the writing of clauses (adding time used for motor execution and pauses between words). The initial pause will be expressed in ms, and the mean duration of clause writing will be expressed in ms/word (the total duration is divided by the number of words in the clause).

3.3. Analyses used

  • 3 « S » means an inter-sentential pause, « A » a pause which is at the beginning of a new A-unit but (...)

20The aim of our analyses is to explore how chronometric measures vary depending on syntactic location or linguistic features of clauses. The texts examined here were produced by 30 participants. We could use either ANOVA or multiple regression analysis. However, these methods fail to take into account inter-individual variability while preserving the precision of the data. To this end we selected a linear mixed model, which fits this kind of data and analysis (Baayen et al. 2008). With this approach, fixed and random effects have to be indicated. Fixed effects are factors with several modalities chosen by the experimenter: for instance, “S”, “A” and “C”3 are three modalities of the fixed effect “syntactic location”. Random effects are variables which are a population sample, without being controlled by the experimenter: in our study, the variable “participant” is a random effect. Linear mixed-models can be used to explain dependant variables (i.e. initial pause length or writing duration) by fixed (i.e. linguistic features) and random (i.e. participants) effects.

21Several steps are needed to establish the most relevant results. First, an empty model (“model 0”) is established with only three variables: one fixed effect (the intercept), a random effect for subjects and the variance of the residual error. With this model we can calculate the within-subject correlation coefficient, i.e. the between-subjects variance. Then we include variables which can explain the variance of the dependent variable: these variables are included as fixed-effects; they will be described for each analyse. The final step is to establish the most parsimonious model - the model which is the most adjusted (using the Maximum Likelihood deviance) with the minimum number of parameters. These analyses were conducted by using the SPSS Statistics 20 software.

22Since our dependant variables are chronometric measures, we used a log-transformation for normalization. However, results will be given in milliseconds to give a better idea of the durations.

3.4. Hypotheses

23The main question of this study is: what is the scope of planning? Is language planned over several clauses during text production?

24Our hypotheses were:

  • H1: The syntactic location effect is present in our corpus: pauses before sentences are longer than those before A-Units, which are longer than before clauses.

  • H2: The more the writer has to plan, the longer the initial pause and/or the longer the writing time.

- H2a: The pause before the 1st clause of sentences is longer than the pause before the following clauses in the same sentence.

- H2b: The pause before matrix clauses is longer than the pause before non-matrix clauses.

- H2b’: Writing time for matrix clauses is longer than that of non-matrix clauses.

- H2c: The pause before subordinate clauses which precede the matrix clause is longer than the pause before other subordinate clauses.

- H2c’: Writing time of subordinate clauses which precede the matrix clause is longer than that of other subordinate.

  • H3: The more the planning of the clause is anticipated, the shorter the pauses and/or the shorter the writing time.

- H3a: The pause before obligatory clauses (OV and OR) is shorter than the pause before optional clauses.

- H3a’: Writing time of obligatory clauses (OV and OR) is shorter than in optional clauses.

- H3b: The pause before subordinate clauses is shorter than the pause before coordinate or juxtaposed clauses.

- H3b’: Writing time of subordinate clauses is shorter than in coordinate or juxtaposed clauses.

4. Analysis of results

4.1. General description: pause length at syntactic boundaries

4.1.1. Repartition in quartiles

25To begin our study, we need to verify that the « syntactic location effect » is present in our data. This would provide the first cue of planning over several clauses.

26Pauses have been classified depending on their length relative to the other pauses of the text: they have been divided in quartiles. Thus, pauses in “Q1” are the 25% shortest pauses, whereas pauses in “Q4” are the 25% longest pauses. Table 1 presents the pause distribution depending on their syntactic location and their length expressed in quartile.

  • 4 « S » means an inter-sentential pause, « A » a pause which is at the beginning of a new A-unit but (...)

Table 1. Pause distribution depending on their syntactic location and their length in quartile4

Q1

Q2

Q3

Q4

Total

« S »

5

1

23

530

559

« A »

5

13

53

170

241

« C »

31

55

145

251

482

« I »

2181

2046

1945

1223

7395

27A chi² test has been used to check for the dependence of the variables “syntactic location” and “pause length in quartile”. The result is significant, (χ(9)=2282, p<0.001), meaning that these variables are dependent. An equal repartition in each column would have been observed if the pause length was not linked to the syntactic location, and this is not the case. Bold values in Table 1 indicate observed values which are superior to theoretical values: more “S” and “A” pauses are very long (“Q4”), “C” pauses are more present in “Q3” and “Q4”, whereas there are less “I” pauses than expected in “Q4”. Syntactic boundaries (“C”, “A”, “S”) can therefore be considered as most common locations for long pauses.

Figure 2. Proportion pauses of each syntactic location per quartile

Figure 2. Proportion pauses of each syntactic location per quartile

Figure 3. Proportion of pauses (in %) of each quartile per syntactic location

Figure 3. Proportion of pauses (in %) of each quartile per syntactic location

28Figure 2 and Figure 3 show that almost all pauses of Q1 and Q2 are “I” pauses, whereas almost 45% of Q4 pauses are “C”, “A” or “S” pauses. The 55% of “I” pauses in Q4 represent only 18% of all “I pauses”. Among the 1223 “I” of Q4, 64 pauses are before a revision, 57 pauses are just after a revision, and 13 pauses are the last sequence of the text, which can explain why some “I” pauses are very long despite their syntactic location. At the opposite, inter-sentential pauses are very long, since the 25% of “S” pauses in Q4 represent 95% of all “S” pauses.

4.1.2. Mixed-model: length of initial pauses depending on syntactic locations (“C”, “A”, “S”)

29We used a linear mixed-model to explain the clause initial pause length depending on its syntactic location: we indicate if the clause was at the beginning of a sentence (“S”), of an A-unit (“A”) or within an A-unit (“C”). The random effect was the participant.

  • Model 0:

30With the empty model, the within-subject correlation coefficient (i.e. the between-subjects variance) is 9.54%. Thus, it is important to use a mixed-model to take into account variability due to participants.

  • Model 1:

31We added the fixed effect with 3 modalities: “S”, “A”, “C”. Model 1 is significantly better than the model 0 (χ(2)=380.485, p<0.000). Post-hoc analyses (Bonferroni) confirm that each syntactic location is significantly different from the others. In this model, the pseudo R², which describes the amount of variance explained by fixed effect (here the syntactic location), is 26.58%.

  • Model 2:

32We wanted to verify that it was useful to distinguish between “A” and “C” pauses. Thus, we made a new model with only two modalities for the syntactic location: at the beginning of a sentence (“S”) or within a sentence (“A” and “C”). This model is significantly worse than model 1 (χ(1)=26.588, p<0.000). Thus, the best model is the model introducing the three syntactic locations “”S”, “A”, “C”.

33The analyses undertaken in this first part confirm that the pause length at the syntactic boundaries is not randomly distributed. First, “C”, “A” and “S” pauses are mainly long (in the Q4 quartile), while “I” pauses are more frequent in the Q1, Q2 and Q3 quartiles. Second, the analyses with linear mixed models show that the pause length of initial clauses is best predicted if we distinguish between pauses before sentences (“S”), A-unit (“A”) and clauses (“C”). Thus, hypothesis 1 is confirmed.

4.2. Chronometric measures of clauses explained by linguistic features

  • 5 OV stances for clauses which are obligatory to complete the valence of the predicate; OR for those (...)

34In this part we again used a linear mixed-model. The model 0 indicates the part of variance explained by between-subject variation. In model 1 four fixed-effects were added: the clause position within the sentence (2 modalities: 1st clause / following clauses), the position of the A-unit within the sentence (4 modalities), if the clause is matrix or non-matrix (2 modalities: matrix / non-matrix), if it is obligatory or facultative (5 modalities: OV, OR, OC, FC5, non-relevant), the clause type (6 modalities: coordinate, juxtaposed, juxtaposed after subordinate, correlative, subordinate before matrix, subordinate after matrix) and if it is a subordinate or an independent clause (2 modalities: subordinate / independent). After indicating the within-subject correlation coefficient issued from the Model 0, we present only the results of the final model, which is the most parsimonious model.

4.2.1. Initial pause before clauses

  • Model 0:

35In the empty model, the within-subject correlation coefficient was 0.1026, i.e., between-subjects variance represented 10.26% of the total variance.

  • Final model:

36The significant fixed-effects introduced in the final model to explain clause initial pause length are: “clause position within the sentence” (2 modalities) ; “matrix / non-matrix” (2 modalities) ; “obligatory / facultative” (4 modalities: “OV” and “OR” are combined) ; clause type (6 modalities). The improvement of the maximum likelihood deviance for model 2 over the model 0 was significant (χ(10)=424.796, p<0.000).

37The significant effects are summarized in Table 2 and they are illustrated by the following graphics. The results indicate significant differences between modalities, obtained with a Bonferroni test. As was predicted by the hypothesis 2a, the pauses at the beginning of a sentence are longer than pauses before clauses within the sentence (Figure 4). Matrix clauses are initiated by a longer pause than non-matrix clauses (Figure 5), in accordance with the hypothesis 2b, and pauses before obligatory clauses are shorter than facultative clauses (hypothesis 3a, Figure 6). Finally, pauses before subordinate clauses are longer if the subordinate clause is before the matrix clause, as was suggested by the hypothesis 2c (Figure 7).

Table 2. Summary of significant fixed effects for initial pause of clauses

Fixed effects 

ddl

F

Sig.

Results (“>” means “longer pause”)

Intercept

1

1115.43

0.000

Clause position within the sentence

1

67.54

0.000

1st clause > following clauses

Matrix / non-matrix

1

4.48

0.035

matrix > non matrix

Obligatory / facultative

3

5.18

0.001

obligatory “OV” or “OR” > facultative

Clause type

5

4.40

0.001

Subordinate before matrix clause > after matrix

Figure 4. Pause length before clauses (in ms) depending on the position in the sentence

Figure 4. Pause length before clauses (in ms) depending on the position in the sentence

Figure 5. Pause length before clauses (in ms) depending on the feature “matrix / non-matrix”

Figure 5. Pause length before clauses (in ms) depending on the feature “matrix / non-matrix”

Figure 6. Pause length before clauses (in ms) depending on the feature “obligatory (“OR” or “OV”) / facultative” (“FC”)

Figure 6. Pause length before clauses (in ms) depending on the feature “obligatory (“OR” or “OV”) / facultative” (“FC”)

Figure 7. Pause length before clauses (in ms) depending on the position of the subordinate clause in relation to the matrix clause

Figure 7. Pause length before clauses (in ms) depending on the position of the subordinate clause in relation to the matrix clause

4.2.2. Mean duration of clause writing (in milliseconds per word)

  • Model 0:

38The within-subject correlation coefficient was 0.1635, i.e., between-subjects variance represented 16.65% of the total variance.

  • Model 2:

39The improvement of the maximum likelihood deviance for model 2 over model 0 was significant (χ(7)= 39.13, p<0.000). In this model, only two linguistic variables were included as fixed-effects: the position of the A-unit within the sentence (4 modalities) and if it is subordinate or independent clause (2 modalities: subordinate / independent). Subordinate clauses are written slower than coordinated and juxtaposed clauses (p<0.001), which confirms hypothesis 3b’ (Figure 8). Writers spend less time writing the clauses in the second A-unit of the sentence than in the first (p<0.05) or fourth (p=0.05) A-unit (Figure 9).

Figure 8. Mean duration of clause writing (in ms/word) depending on the position of the clause within the A-unit

Figure 8. Mean duration of clause writing (in ms/word) depending on the position of the clause within the A-unit

Figure 9. Mean duration of clause writing (in ms/word) depending on the position of the A-unit within the sentence

Figure 9. Mean duration of clause writing (in ms/word) depending on the position of the A-unit within the sentence

5. Discussion

40First, the “syntactic location effect “ is strongly observed in our corpus, as was reported in many previous studies (e.g. Foulin 1998; Schilperoord 1996). Results of mixed-models, which take into account “participants” as a random effect supported that pause length at clause boundaries was best predicted when the three levels were distinguished: between sentences, between A-units and between clauses. This effect is also apparent when we observe the distribution of pauses for each text: almost all sentence boundaries are included in the 25% of longest pauses, whereas pauses before words within clauses (coded “I” in this study) are shorter. The study of quartiles allows us to capture the distribution of pauses of individual writers, since the cognitive value of durations may vary across writers (Chenu et al. 2014; Fortier & Préfontaine 1994). Furthermore the presence of this syntactic location effect supports our methodological choice of a linguistically driven approach. Some pauses before clause-internal words (“I pauses”) are indeed very long, but the strong dependence between syntactic location and pause length suggests that particular cognitive processes are involved at these boundaries, and are not the same at the different levels. In the view of a cascading model (Olive 2014; van Galen 1991), which suggests that higher-level processes process larger units of language than lower level ones, we could hypothesize that conceptualization could occur at the sentence level, while formulation could occur first at an A-unit level (e.g. syntactic structure) and then be specified (e.g. exact words) at the clause level.

41Second, the study of linguistic features of clauses in relation to chronometric measures corroborate the hypothesis of planning over several clauses. This hypothesis can be defended by observing two types of chronometric measure variations: a longer pause or writing duration when much information has to be planned, and a shorter duration when information has already been planned. Matrix clauses (i.e. clauses followed by a subordinate clause) are initiated by a longer pause than non-matrix clause. This difference could be due to the anticipation of the subordinate clause. We could suppose that pauses preceding subordinate clauses would be shorter than before non-subordinate clauses, as was found in previous studies (Immonen & Mäkisalo 2010; van Hell et al. 2008). It was not exactly the case in our data, but we did observe that this type of clause was written faster, which could suggest that writing duration benefits from the anticipatory planning. Thus, it is necessary to consider both pauses and writing time in order to better understand the temporal management of conceptualization and formulation, since planning does not occur only during scriptural inactivity. Furthermore, our study revealed that the anticipated preparation of subordinate clauses, which was found in previous studies, should be nuanced depending on its linguistic status. Clauses which are obligatory to complete the valence of the verb or to delimitate a referent are preceded by a shorter pause than clauses which are facultative. It can be argued that facultative clauses are more incrementally planned than obligatory clauses, because writers primarily plan elements which are obligatory to obtain a syntactically complete and fully communicative structure. A last case of clear over-clausal planning concerns A-units with subordinate clauses preceding matrix clauses. Subordinate clauses written before matrix clauses have the longest initial pauses, probably because it is necessary to plan both clauses simultaneously; this pattern was already observed in 9th grade students in the developmental study of Ailhaud et al. (2016).

42The results concerning planning at the sentence level are less clear. Inter-sentence pauses are longer than pauses at other locations, lasting on average almost five seconds, which could be interpreted as the time required to plan the whole sentence. However, we observed that the fourth A-unit of a sentence is written more slowly. This deceleration could have several explanations. First, the planning scope could be limited, because of limited cognitive capacities (McCutchen 1996): thus, the writer would have to plan the end of his long sentence when he is writing. Second, he could begin to read his sentence during the writing of its end in order to check the adequacy with the intended goal, and thus the long pause at sentence boundaries could also be devoted to reading the text produced so far (Alamargot 2005). Another explanation of this deceleration could be simply physiological: the writer experiences fatigue when he is writing a long sentence, and decreases his writing speed (de Beaugrande 1984; van der Linden et al. 2008). These explanations are not mutually exclusive and illustrate the difficulty of associating chronometric measures with specific cognitive processes.

6. Conclusion

43In order to understand when and how conceptualization and formulation occur in a multiple-sentence text, we attempted to identify the scope of planning. Our results corroborate the hypothesis of planning over several clauses. The fact that initial pauses are longer before matrix clauses than before non-matrix clauses and that subordinate clauses display shorter writing durations suggests that subordinate clauses are anticipated. This is supported in addition by the observation that subordinate clauses written before matrix clauses show the longest initial pauses. The results relative to necessity of clauses suggest a difference in the anticipation of subordinate clauses: more obligatory subordinate clauses (such as defining relatives, for example) display shorter initial pauses than facultative subordinate clauses (such as explicative relatives, for example). A further step for our analyses would be to uncover what exactly is being planned over several clauses by linking more precise semantic and lexical features of clauses with chronometric data. Our results would suggest studying separately matrix clauses, subordinate clauses and autonomous clauses. A fine-grained description of semantic (e.g. accessibility of referents, temporal and spatial details) and lexical aspects (e.g. number of words, frequency) could be linked with chronometric measures to better understand temporal management of conceptualization and formulation. The planning function of the intersentence pauses – which is very long - is less clear. An additional study is needed to untangle the multitude of other processes that may be at play. For example, an analysis of eye-movements at these locations would help to sort out rereading strategies.

Haut de page

Bibliographie

Ailhaud, Emilie, Florence Chenu, & Harriet Jisa. 2016. A developmental perspective on the units of written French. In Joan Perera, Melina Aparici, Elisa Rosado, & Naymé Salas (eds.), Written and spoken language development across the lifespan: Essays in honour of Liliana Tolchinsky, 287–305. Springer.

Alamargot, Denis. 2005. Le rôle de la lecture au cours de l’écriture : ce que nous indiquent les mouvements oculaires du rédacteur. Rééducation Orthophonique, 223, 189–201.

Alamargot, Denis, David Chesnet, Christophe Dansac & Christine Ros 2006. Eye and Pen: a new device to study reading during writing. Behaviour Research Methods, Instruments and Computers, 38(2), 287-299.

Alamargot, Denis, Christophe Dansac, David Chesnet & Michel Fayol. 2007. Parallel processing before and after pauses: A combined analysis of graphomotor and eye movements during procedural text production. In Mark Torrance, David Galbraith & Luuk van Waes (eds.), Cognitive Factors in Writing. Dordrecht-Boston-London: Elsevier Sciences Publishers

Alamargot, Denis & Jean-Louis Lebrave. 2010. The study of professional writing: A joint contribution from cognitive psychology and genetic criticism. European Psychologist, 15(1).

Alves, Rui A., Marta Branco, São Luis Castro & Thierry Olive. 2012. Effects of handwriting skill, output modes and gender of fourth graders on pauses, written language bursts, fluency and quality. In Virginia W. Berninger (ed.), Past, present, and future contributions of cognitive writing research to Cognitive Psychology, 389-402. New York: Psychology Press.

Alves, Rui A., São Luis Castro & Thierry Olive. 2008. Execution and pauses in writing narratives: Processing time, cognitive effort and typing skill. International Journal of Psychology, 43, 969-979.

Baayen, R. Harald, Douglas J. Davidson & Douglas N. Bates. 2008. Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59, 390–412.

Beaugrande, Robert de. 1984. Text production. Toward a science of composition Norwood, N.J: Ablex.

Beauvais, Caroline, Thierry Olive, Jean-Michel Passerault. 2011. Why Are Some Texts Good and Others Not? Relationship Between Text Quality and Management of the Writing Processes. Journal of Educational Psychology, 103(2), 415–428.

Beers, Scott F. & William Nagy. 2009. Syntactic Complexity as a Predictor of Adolescent Writing Quality: Which Measures? Which Genre? Reading and Writing: An Interdisciplinary Journal, 22, 185-200.

Berman, Ruth A. & Dan I. Slobin. 1994. Relating Events in Narrative: a Crosslinguistic Developmental Study. Hillsdale, NJ: Lawrence Erlbaum Associates.

Biasi, Pierre-Marc de. 2002, 2nd edn. La critique génétique. In Daniel Bergez (ed.), Introduction aux méthodes critiques pour l'analyse littéraire, 37-73. Paris : Nathan.

Chanquoy, Lucile, Jean-Noël Foulin & Michel Fayol. 1996. Writing in adults: A real-time approach. In Gert Rijlaarsdam, Huub van den Bergh & Michel Couzijn (eds.), Theories, models and methodology in writing research, 36–43. Amsterdam.

Chenu, Florence, François Pellegrino, Harriet Jisa & Michel Fayol. 2014. Interword and intraword pause threshold in writing. Frontiers in Psychology, 5, 1–7.

Chesnet, David & Denis Alamargot. 2005. Analyse En Temps Réel Des Activités Oculaires et Grapho-Motrices Du Scripteur : Intérêt Du Dispositif « Eye and Pen ». L’année psychologique, 105(3), 477–520.

Doquet-Lacoste, Claire. 2003. Écriture et traitement de texte à l’école élémentaire : modes d'analyse et pistes de travail. Langage et Société, 103, 11–29.

Fabre, Claudine. 1990. Les brouillons d’écoliers ou l’entrée dans l’écriture, Grenoble : Ceditel / L'atelier du texte.

Favart, Monik & Thierry Olive. 2005. Modèles et méthodes d’étude de la production écrite Models and methods for the study of writing. Psychologie Française, 50(3), 273–285.

Fayol, Michel. 1997. Des idées au texte : psychologie cognitive de la production verbale, orale et écrite. Paris : PUF.

Fenoglio, Irène. 2009. La fin de Tre cavalli d’Erri de Luca : le livre à la place de l’arme, Genesis, 25, 151-156.

Ferrand, Nathalie. 2012. Brouillons des Lumières, Presses de l’Université Paris-Sorbonne.

Fortier, Gilles & Clémence Préfontaine. 1994. Pauses, relecture et processus d’écriture. Revue des Sciences de l’Éducation, 20(2), 203–220.

Foulin, Jean-Noël. 1993. Pauses et débits : les indicateurs temporels de la production écrite, Thèse de Doctorat, Université de Bourgogne.

Foulin, Jean-Noël. 1998. To what extent does pause location predict pause duration in adults’ and children's writing? Cahiers de Psychologie Cognitive, 17(3), 601–620.

Galen, Gerard P. van.1991. Handwriting: Issues for a psychomotor theory. Human Movement Science, 10, 165–191.

Grésillon, Almuth. 2002. Langage de l’ébauche : parole intérieure extériorisée. Langages, 147, 19–38.

Hayes, John R. 2012. My Past and Present as Writing Researcher and Thoughts about the Future of Writing. Past, Present, and Future Contributions of Cognitive Writing Research to Cognitive Psychology, 3–26.

Hayes, John R. & Linda Flower. 1980. Identifying the Organization of Writing Processes. In Lee W. Gregg & Erwin R. Steinberg (eds.), Cognitive processes in writing, 3–30. Hillsdale, NJ: Erlbaum.

Hayes, John R., & Nash, Jane G. 1996. On the nature of planning in writing. In C. Michael Levy & Sarah Ransdell (eds.), The Science of Writing: Theories, Methods, Individual Differences and Applications (pp. 29–55).

Hell, Janet G. van, Ludo Verhoeven & Liesbeth M. van Bijsterverldt. 2008. Pause time patterns in writing narrative and expository texts by children and adults. Discourse Processes, 45, 406–427.

Herschberg Pierrot, Anne. 2012. Flaubert. L'Empire de la bêtise, Nantes, Éditions Cécile Defaut.

Immonen, Sini & Jukka Mäkisalo. 2010. Pauses reflecting the processing of syntactic units in monolingual text production and translation. Hermes–Journal of Language and Communication Studies, 44, 45–61.

Just, Marcel A. & Patricia A. Carpenter. 1992. A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122-149.

Lebrave, Jean-Louis. 2006. La critique génétique : une discipline nouvelle ou un avatar moderne de la philologie?, Item, Disponible sur: http://www.item.ens.fr/index.php?id=14048

Leijten, Mariëlle & Luuk van Waes. 2013. Keystroke Logging in Writing Research: Using Inputlog to Analyze and Visualize Writing Processes. Written Communication, 30(3), 358–392.

Levelt, Willem J. M. 1989. Speaking: From intention to articulation. Cambridge & London: MIT Press.

Linden, Dimitri van der, Michael Frese, Theo F. Meijman (2008). Mental fatigue and the control of cognitive processes: effects on perseveration and planning, Brain Research Reviews Brain. 59(1), 125-139

Maggio, Séverine, Bernard Lété, Florence Chenu, Harriet Jisa & Michel Fayol. 2012. Tracking the mind during writing: immediacy, delayed, and anticipatory effects on pauses and writing rate. Reading and Writing, 25(9), 2131–2151.

Matsuhashi, Ann. 1982. Explorations in the real-time production of written discourse. In Martin Nystrand (ed.), What writers know: The language, process, and structure of written discourse, 269–290. New York: Academic Press.

McCutchen, Deborah. 1996. A capacity theory of writing: Working memory in composition. Educational Psychology Review, 8(3), 299–325.

Olive, Thierry. 2014. Toward a Parallel and Cascading Model of the Writing System: A Review of Research on Writing Processes Coordination. Journal of Writing Research, 6(2), 173–194.

Plane, Sylvie. 2003. Stratégies de réécriture et gestion des contraintes d’écriture par des élèves de l’école élémentaire : ce que nous apprennent des écrits d’enfants sur l’écriture. Rivisita Italiana de Psicolinguistica Applicatta Anno, 3(1), 57–77.

Plane, Sylvie, Denis Alamargot & Jean-Louis Lebrave. 2010. Temporalité de L’écriture et Rôle Du Texte Produit Dans L’activité Rédactionnelle. Langages, 177(1), 7–28.

Schilperoord, Joost. 1996. The distribution of pause time in written text production. In Gert Rijlaarsdam, Huub van den Bergh & Michel Couzijn (eds.), Theories, models and methodology in writing research, 21–35. Amsterdam: Amsterdam University Press.

Schilperoord, Joost & Sanders, T. 1999. How hierarchical text structure affects retrieval processes: Implications of pause and text analysis. In Mak Torrance & David Galbraith (eds), Knowing What to Write. Conceptual Processes in Text Production, 13–33. Amsterdam: AUP.

Schilperoord, Joost. 2002. On the cognitive status of pauses in discourse production. In T. Olive & M. Levy (eds.), Contemporary tools and techniques for studying writing, 61–87. Dordrecht, The Netherlands: Academic Publishers.

Strömqvist, Sven, Kenneth Holmqvist, Victoria Johansson, Henrik Karlsson & Åsa Wengelin. 2006. What keystroke logging can reveal about writing. In K. P. H. Sullivan & E. Lindgren (eds.), Computer Keystroke Logging: Methods and Applications, vol. 18, 45–71. Amsterdam: Elsevier Ldt.

Torrance, Mark, Roger Johansson, Victoria Johansson & Åsa Wengelin. 2015. Reading during the composition of multi-sentence texts: an eye-movement study. Psychological Research, 1-15.

Waes, Luuk van, Mariëlle Leijten, Åsa Wengelin & Eva Lindgren. 2012. Logging tools to study digital writing processes. In Viginia W. Berninger (ed.), Past, present, and future contributions of cognitive writing research to cognitive psychology, 507-533. New York/Sussex: Taylor & Francis.

Haut de page

Annexe

Legend: “~” indicates a boundary between sentences (“S”); “//” indicates a boundary between A-units (“A”); “/” indicates a boundary between clauses (“C”).

(1)    ~ Le couteau [dérobé le midi même à la cantine] était tenu fermement par sa main droite ;

The knife [stolen at noon in the canteen] was firmly held in his right hand;

// d’un pas rapide, il se plaça devant le gardien, // lui mit la lame sous la gorge // et lui intima

// with a quick step he placed himself in front of the guard, // put the blade under his throat, // and intimated him

/ d’être silencieux. ~ Il ne s’agissait pas / d’ameuter le service de sécurité au complet

/ to be silent. ~ It was not a question / of rousing the entire security service

dès le commencement de l’action. ~ Il demanda au gardien / d’avancer plus en avant

from the beginning of the action. ~ He asked the guard / to move forward

dans la cellule / et de s’assoir sur le lit. ~ Il murmurait tous ces mots / qu’il s’était déjà

in the cell / and sit on the bed. ~ He murmured all those words / that he had already

répétés cent fois. ~ Oui, il avait planifié jusqu’aux mots / qu’il dirait au gardien,

repeated a hundred times. ~ Yes, he had planned right up to the words / he would say to the guard,

// il ne fallait / laisser aucune place à l’imprévu. ~ Une fois sur le lit, assis à la droite de celui

// there was no room for the unexpected. ~ Once on the bed, seated to the right of him

/ qui était désormais son prisonnier, / Dimitri lui ordonna / d’ôter son uniforme.

/ who was now his prisoner, / Dimitri ordered him / to remove his uniform.

~ Le gardien était resté étrangement silencieux depuis le début de son " enlèvement", 

~ The guard had remained strangely silent since the beginning of his "abduction",

// il se déshabilla / toujours sans ouvrir les lèvres.

// he undressed / still without opening his lips.

~ Dimitri s’empara alors du bandeau de tissu / qu’il avait déchiré sur le bas de son drap

~ Dimitri then grabbed the strip of cloth / which he had torn off the bottom of his sheet

la nuit précédente, // il en entoura la tête du gardien par deux fois au niveau de la bouche.

the night before, // he surrounded the guardian's head twice at the level of the mouth.

~  Une fois le bâillon en place, / il prit les menottes / qui pendaient à la ceinture de l’uniforme

~ Once the gag was in place, / he took the handcuffs / hanging from the belt of the uniform

// et attacha le gardien au lit. ~ Il était désormais temps / qu’il quitte à son tour ses vêtements.

// and tied the guard to the bed. ~ It was now time / he took off his own clothes.

~ Il enfila ensuite ceux / qu’il avait prélevés au gardien. ~ Avant de passer la porte de ce / qui

~ He then put on the ones / he had taken from the guard. ~ Before passing the door of / what

avait été sa cellule, / il veilla / à ce que la casquette soit suffisamment enfoncée sur sa tête

had been his cell, / he ensured / that the cap was sufficiently pressed on his head

/ pour que son visage soit en partie recouvert. ~ Il sortit // et regarda sa montre.

/ so that his face would be partly covered. ~ He went out // and looked at his watch.

~ Dix-huit heures trente, il était parfaitement à l’heure. ~ C’était l’heure de la relève

~ Eighteen thirty, he was perfectly on time. ~ It was the time of the changing

entre gardiens, // ceux du service de nuit étaient arrivés,

between guards, // those of the night service had arrived,

// il pouvait donc s’en aller. ~ En saluant vaguement deux ou trois " collègues ",

// so he could leave. ~ Vaguely saluting two or three "colleagues",

/ il traversa la cour / qui menait à la porte d’entrée.

/ he crossed the courtyard / which led to the front door.

~ À dix-huit heures trente-cinq, il était libre.  [S11]

~ At eighteen-thirty-five he was free. [S11]

Haut de page

Notes

1 “subjects are asked to say aloud everything they think and everything that occurs to them while performing a task, no matter how trivial it may seem.” Hayes & Flower (1980:4)

2 The code at the end of each example identifies the participant. 

3 « S » means an inter-sentential pause, « A » a pause which is at the beginning of a new A-unit but not a new sentence, « C » a pause which is at the beginning of a new clause but not a new sentence. Pauses before words within clauses will be referred to as “I”.

4 « S » means an inter-sentential pause, « A » a pause which is at the beginning of a new A-unit but not a new sentence, « C » a pause which is at the beginning of a new clause but not a new sentence. Pauses before words within clauses are referred to as “I”.

5 OV stances for clauses which are obligatory to complete the valence of the predicate; OR for those which are obligatory to delimitate the referent (OR); OC for clauses which are obligatory from a communicational perspective and FC for clauses which are optional.

Haut de page

Table des illustrations

Titre Figure 2. Proportion pauses of each syntactic location per quartile
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-1.png
Fichier image/png, 12k
Titre Figure 3. Proportion of pauses (in %) of each quartile per syntactic location
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-2.png
Fichier image/png, 15k
Titre Figure 4. Pause length before clauses (in ms) depending on the position in the sentence
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-3.png
Fichier image/png, 12k
Titre Figure 5. Pause length before clauses (in ms) depending on the feature “matrix / non-matrix”
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-4.png
Fichier image/png, 11k
Titre Figure 6. Pause length before clauses (in ms) depending on the feature “obligatory (“OR” or “OV”) / facultative” (“FC”)
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-5.png
Fichier image/png, 13k
Titre Figure 7. Pause length before clauses (in ms) depending on the position of the subordinate clause in relation to the matrix clause
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-6.png
Fichier image/png, 11k
Titre Figure 8. Mean duration of clause writing (in ms/word) depending on the position of the clause within the A-unit
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-7.png
Fichier image/png, 12k
Titre Figure 9. Mean duration of clause writing (in ms/word) depending on the position of the A-unit within the sentence
URL http://journals.openedition.org/cognitextes/docannexe/image/992/img-8.png
Fichier image/png, 18k
Haut de page

Pour citer cet article

Référence électronique

Emilie Ailhaud et Florence Chenu, « Variations of chronometric measures of written production depending on clause packaging », CogniTextes [En ligne], Volume 17 | 2018, mis en ligne le 13 juillet 2018, consulté le 20 septembre 2018. URL : http://journals.openedition.org/cognitextes/992 ; DOI : 10.4000/cognitextes.992

Haut de page

Auteur

Emilie Ailhaud et Florence Chenu

Laboratoire Dynamique Du Langage, UMR 5596 (Université Lyon2 & CNRS), France

Haut de page
  • Logo DOAJ – Directory of Open Access Journals
  • Logo ERIH PLUS | NSD
  • Logo AFLiCo – Association française de linguistique cognitive
  • OpenEdition Journals