UNE ANALYSE DE CORPUS DE L’ALTERNANCE CAUSATIVE EN ANGLAIS
Cette thèse propose, par le biais d’une analyse d’un corpus comprenant plus de 10,000 occurrences de la construction intransitive non-causative et de la construction transitive causative, d’évaluer la quantité d’information partagée entre ces deux constructions et d’ajouter au débat entre alternance et généralisations de surface dans le cadre de la grammaire de constructions. Cette recherche peut être déroulée en trois étapes : (i) une description de l’alternance causative en anglais (The fabric stretched vs. Joan stretched the fabric) basée sur un corpus de 11,554 instances de 29 verbes dans ces deux constructions, (ii) une méthode d’analyse qui permet de mesurer la force d’alternance des verbes mais aussi la quantité et le type d’informations partagées par les deux constructions et (iii) une réflexion quant à l’importance des généralisations de bas niveau en complément des généralisations au niveau de la construction schématique. Pour ce faire, ce travail est divisé en deux parties. Tout d’abord, une partie “cadre théorique” qui propose un état de l’art sur la question des structures argumentales en linguistique cognitive mais aussi une comparaison des postulats de celle-ci avec d’autres approches. Dans cette première partie, nous abordons la question de la place de la construction et de l’organisation du “constructicon” chez les locuteurs.trices de l’anglais. Ensuite, une partie dédiée à l’exposition des données collectées et utilisées pour cette recherche, de la méthode et des techniques utilisées pour mesurer les similarités et différences entre les deux constructions ainsi que des résultats préliminaires et enfin un chapitre dédié à la présentation des résultats obtenus avec la méthode développée ainsi qu’une évaluation générale de cette méthode.
A description of the causative alternation
One of the reasons why we undertook this research is an effort to add to the discussion on alternating constructions in construction grammar, where it is often posited that constructions have meaning of their own and that the alternation itself does not play as important a role as may be believed by other approaches such as generative grammar (Chomsky 1965) or lexicalprojectionist approaches (Levin 1993). We believe that in order to assess the role played by argument structure alternations in the mind of speakers, it is important to look closely at the data and evaluate to what extent meaning is shared between two alternants, as was done by Cappelle (2006) and Perek (2015), for example. In order to reach this goal, we compared two argument structure constructions that share certain elements. That is, many causative verbs are known to alternate between an intransitive and a transitive construction. What is more, one of the participants of the event denoted by the verb is shared. This participant, which has been given several names in the literature but which we call “theme” is the participant that undergoes the event denoted by the verb. It is found in subject position in the intransitive non-causative construction and in object position in the transitive causative construction. Since one of our aims was to provide a description of the causative alternation we set out to analyse instances of both the intransitive non-causative construction, illustrated with the example in (1) and the transitive causative construction, as shown in example (2).1
(1) [… ] and return it to normal after mixture has frozen.
(2) Cover and freeze mixture 1 hour or until frozen around the edges. There are many verbs that can be found with these two constructions: Levin (1993) counts 355 verbs that enter the causative alternation. It is beyond the scope of this thesis to analyse 355 different verbs thoroughly. Therefore we decided to focus on 29 verbs out of the list of 355 verbs proposed by Levin. These verbs are grouped into five different sets, provided below. The first three were identified by Levin (1993) and the last two we created ourselves based on their supposed semantic similiarity. The verbs that make up these two sets were put together in the category “other verbs of change of state” by Levin.
(3) Verbs and verb groups used for this research:
a. Break verbs: break, crack, crush, shatter, snap and tear
b. Bend verbs: bend, crease, crinkle, crumple, fold and wrinkle
c. Roll verbs: roll, drop, move, slide and turn
d. Grow verbs: grow, expand, increase, proliferate, stretch and thicken
e. Change of temperature verbs: burn, chill, cool, freeze, heat and warm
We decided to put grow verbs together because they all denote a change in the size of the undergoer. As to the last set, as its name suggests, it is made up of verbs that denote a change of temperature. While this obviously does not cover the entire scope of the causative alternation, the set of verbs analysed is quite varied: they describe fairly different types of events and are thus already a good indication of the mechanisms at play in the two constructions which compose the alternation. This set of 29 verbs served as the basis for our investigation of the causative alternation and of each construction individually. We extracted 32,355 instances of these verbs from the Corpus of Contemporary American English, which we then annotated manually for construction. In this sample, we identified 11,554 instances of the intransitive non-causative construction and the transitive causative construction combined. As we will argue throughout this thesis, a careful look at the data is essential when one wants to provide a description of a linguistic phenomenon. The constitution of a relevant and substantial corpus is one of the first steps in this direction. It allows a description of linguistic phenomena at different levels of generalisation : by studying a wide and varied array of instances of one or more argument structure constructions, one can identify item-specific constraints, verb-class constraints and even, abstracting away from individual instances, constraints at the level of the schematic argument structure construction itself. This research will present examples of constraints at these various levels. For example, in the instances presented in (1) and (2), one argument is shared: mixture and it undergoes the same change of state (denoted by the verb freeze) in each example; thus, it goes from “not frozen” to “frozen.” However, while going through the data we found that some themes are not shared, as exemplified in (4) and (5).
(4) I stretch the truth a smidge, or, you know, something like that.
(5) *The truth stretches (a smidge)
It quickly became clear that a vast majority of the verbs we chose were much more frequent in one construction or the other, thus already showing a substantial difference in their distribution between the two constructions. We also wanted to add one level of description in the behaviour of argument structure constructions. That is, in construction grammar, the analyses of argument structure constructions (and maybe more specifically of alternating argument structure constructions) usually revolve around one slot in the construction, that taken up by the verb (Croft 2003; Goldberg 1995; Gries and Stefanowitsch 2004; Perek 2014; Stefanowitsch and Gries 2003, among others). To counterbalance this, we decided to describe the kinds of themes found either in subject position in the intransitive non-causative construction and/or in object position in thetransitive causative construction. This adds considerably to the description of the mechanisms underlying these two constructions but is also the basis for the method we propose to measure the alternation strength of verbs that occur in these constructions, which we describe in the next section.
The theoretical framework
The first part of this thesis is divided into four chapters: Chapter 2 presents the cognitive background of construction grammar, Chapter 3 discusses the question or arguments and thematic roles, Chapter 4 introduces the notion of constructional meaning and offers a detailed description of the two constructions analysed for this research and finally, Chapter 5 explores the supposed dichotomy between alternations and constructions.Construction grammar is a theory that finds its roots in cognitive approaches to language. In Chapter 2 we will provide a definition of what counts as a construction and show that constructions are stored in a structured inventory sometimes called the “constructicon,” where there is a continuity between syntax and the lexicon. Among the key concepts that shape today’s construction grammar are prototype effects (Rosch 1973) and family resemblances (Wittgenstein 1953). We will see how these interact with argument structure constructions, and how it allows us to posit hypotheses with regard to the instantiation of the constructions of the causative alternation. For example, how certain semantically similar themes behave in the same way. As to the question of argument structure constructions, it is grounded in two different yet compatible theories: the concept of “frame semantics” developed by Fillmore (1977) and the notion that our organisation of event structure is grounded in our experience of the world. In Chapter 3 we will briefly present two prototypical thematic roles, namely agent and patient and show how traditional thematic roles may be problematic for the description of actual language data, notably with the alternative construction. Through this chapter we will discuss the role found in both the intransitive non-causative construction and the transitive causative construction: the theme. Our notion of theme is, among other things, based on Langacker’s concept of thematic relationship, which is centred around the theme participant (Langacker 1991). We will then move on to an overview of constructional approaches to the argument realisation and insist on the importance of local generalisations, following Lemmens’s argument that “a unified description of the grammar of processes and events requires the inclusion of a lexical perspective” (Lemmens 1998: 47) and the basic assumption in construction grammar that any participant role of a verb must be construable as an instance of the more general argument role of the construction (Goldberg 1995). This interaction between elements of the construction and the construction itself will be investigated further in Chapter 4, where we will explore the supposed dichotomy between lexical rules and constructions (Croft 2003). We will then offer a detailed description of the intransitive non-causative construction and the transitive causative construction, which we will allow us to show some of the differences and common properties of these two constructions. The differences between the two constructions will lead us to Chapter 5, where we will evaluate different approaches to the question of alternating argument structure constructions.We will first discuss approaches that are centred around the verb, namely lexical-projectionist approaches (Levin 1993) and valency approaches (Herbst and Schüller 2008). These approaches will be measured against the proposed status of surface generalisations defended by Goldberg (2002). Finally, we will compare two potential links between the two constructions: allostructions on the one hand (Cappelle 2006) and a subpart link on the other (Goldberg 1995).
A brief overview of construction grammars
Construction grammar emerged in the late 1980’s with major works such as Fillmore (1988), Fillmore, Kay, and O’Connor (1988), Kay (1984, 1988), and Lakoff (1987). Two branches of construction grammar developed in parallel: Berkeley Construction Grammar (taking its name from the corresponding university) on the one hand and Langacker’s own version of cognitive grammar (Langacker 1987, 1991). Construction grammarians stood against so-called Chomskyan approaches to language and notably against the notion of a clear division of labour between the syntax on the one hand (syntactic rules) and the lexicon on the other. They instead argued that grammatical constructions, then defined as “any syntactic pattern which is assigned one or more conventional functions in a language, together with whatever is conventionalized about its contribution to the meaning or the use of structures containing it” (Fillmore 1988: 36) should be given a central place in the study of language. This approach blurred the border between syntax and the lexicon, notably by focusing on partially lexically filled idioms and syntactic structures. The definition of what counts as a construction has since then evolved and become broader.1 One of the definitions that stood out particularly is that of Goldberg (1995): “C is a construction iffdef C is a formmeaning pair <Fi , Si> such that some aspect of Fi or some aspect of Si is not strictly predictable from C’s component parts or from other previously established constructions” (Goldberg 1995:4). This particular definition excludes expressions whose meaning can be seen as componential; however, Goldberg later changed her definition. In Goldberg (2006), the definition is extended to include patterns that are fully predictable “as long as they occur with sufficient frequency” (Goldberg 2006: 5) and thus Goldberg concluded: “it’s constructions all the way down” (Goldberg 2006: 18). Croft (2001), while agreeing with the main principles of Berkeley and “Goldbergian” construction grammars, goes even further and argues that syntactic categories only exist in relation to constructions, and not the other way around. Overall, constructionist linguists agree that a construction is a pairing of form and meaning,similar to Saussure’s linguistic sign (Saussure 1916/1995). Therefore, a construction is a symbolic unit, as represented in Figure 2.1, borrowed from Croft and Cruse (2004: 258). Figure 2.1: The symbolic structure of a construction (Croft and Cruse 2004: 258) As is visible from Figure 2.1, the form (the syntactic, morphological and phonological properties) of a construction is associated to its (conventional) meaning (semantic, pragmatic and discourse-functional properties) via a symbolic correspondence link in the same way that the signifiant is associated with the signifié in Saussure (1916/1995). This definition applies to a wide variety of elements of language, from the word, e.g. wombat, to more schematic constructions such as [the X-er, the Y-er], e.g. the older, the wiser. All these constructions constitute what is often referred to as the “constructicon” (Jurafsky 1992), which is a structured inventory of linguistic knowledge (Langacker 1987: 73). Following the main assumptions of cognitive linguistics, construction grammarians posit that “knowledge of language is knowledge” (Goldberg 1995: 5). Therefore, language is part of a wider array of cognitive abilities. The next two sections will focus on the cognitive mechanisms associated with language and explore the cognitive background of construction grammar.
Prototypes, frames and ICMs
While exploring constructional meaning it is essential to pay attention to the mechanisms that underlie the organisation of linguistic knowledge. Lakoff claims that “we organize our knowledge by means of structures called idealized cognitive models, or ICMs, and that category structures and prototype effects are by-products of that organization.” (Lakoff 1987: 68). To better understand what Lakoff means, we will start by presenting the theory known as “prototype theory” and its main claims (Rosch 1973).It is widely agreed that categorisation is one of our most basic cognitive activities as humans: we organise our knowledge of the world into categories, for example, we know that a cat is a member of the category ANIMAL. Our ability to construe a cat as an instantiation of the category ANIMAL comes from our ability to categorise and to organise our experience of the world in kinds of things that can be grouped into a larger and more abstract category, such as ANIMAL. This type of mental construct is called a “conceptual category”—these categories are cognitive tools which Croft and Cruse claim have at least four different functions: learning, planning, communication and economy (Croft and Cruse 2004: 74). In what is known as the classical model of category structure, categories are based on sets of necessary and sufficient features. Members of a category need to exhibit a certain number of necessary features to be part of a category and an entity that exhibits a certain number of sufficient features will automatically be a member of a corresponding conceptual category. Although formerly widely accepted, this theory has now been discarded by many researchers, following work by Wittgenstein on “family resemblances” (Wittgenstein 1953), which then influenced the work of Rosch (1973, 1978/2004), whose prototype theory will be discussed in more detail in this section. One of the first scholars to point out the inadequacies of the classical theory of categorisation (sets of necessary and sufficient features, clear-cut boundaries) was Wittgenstein (1953) who showed, via a description of the category GAME, that there actually were major flaws in the classical theory. He explains that while there are many things that we call game, itis extremely difficult, even impossible, to find features that are common to all of them. Rather, he argues, these various items such as chess, poker and ring-a-ring-of-roses are related by what he calls “family resemblances” in much the same way that members of a family resemble one another. That is to say, members of a family do not usually share a fixed set of common features: for example, some will have dark hair and blue eyes, while others will have blue eyes but blond hair, some will be tall and sturdy, while others will be short and sturdy. In this sense, family members share many or few of these features, but rarely all of them. Members of the category GAME exhibit the same kind of family resemblances. Some games are fun and not competitive, some are competitive but a large part is left to chance (such as board games with a roll of the dice), some involve physical activity such as tennis, while others mostly rely on wit, such as chess. As Lakoff puts it: “games, like family members, are similar to one another in a variety of ways. That, and not a single, well-defined collection of common properties, is what makes game a category” (Lakoff 1987: 16).
Table des matières
List of Figures
List of Tables
1.1 Research questions
1.1.1 A description of the causative alternation
1.1.2 A methodology to measure alternation strength and shared meaning
1.1.3 Theoretical implications
1.2 Structure of the thesis
1.2.1 Part 1: The theoretical framework
1.2.2 Part 2: Method, data and results
I Theoretical framework
2 The cognitive background of construction grammar
2.1 A brief overview of construction grammars
2.2 Prototypes, frames and ICMs
2.3 The experiential grounding of event structure
3 Arguments, participants and roles
3.1 The semantics of roles
3.1.1 Agent: a prototype-based category
3.1.2 Experiencer, Patient, Theme: semantic and syntactic constraints
3.2 The interaction between constructions and roles
3.2.1 Mapping from lexical semantics to syntax
3.2.2 Constructionist approaches to argument realisation
4 Constructional meaning
4.1 Lexical rules vs. constructions
4.2 A description of the non-causative and the causative constructions
5 Alternation or construction?
5.1 Alternation vs. surface generalizations
5.1.1 Verb-centric approaches
5.1.2 Surface generalizations
5.2 Inheritance links and allostructions
5.2.2 Inheritance links: the subpart link
II Method, data and results
6 Data and method
6.1 Collection and annotation
6.1.1 The corpus
6.1.2 Defining the scope of each construction
6.2 Measuring the alternation strength of causative verbs
6.2.1 Distinctive collostructional analysis
6.2.2 Theme overlap
6.2.3 Semantic grouping of themes
7 Evaluation and results
7.1 BREAK verbs
7.2 BEND verbs
7.3 ROLL verbs
7.4 GROW verbs
7.5 CHANGE OF TEMPERATURE verbs
7.6 General evaluation
8.2 A corpus-based description of the causative alternation
8.3 A method for the analysis of alternations
8.4 Theoretical implications
8.5 Research prospects