Patient-related constraints on get- and be-passive uses in English: evidence from paraphrasing

Thompson, Dominic; Ling, Shuping; Myachykov, Andriy; Ferreira, Fernanda; Scheepers, Christoph

doi:10.3389/fpsyg.2013.00848

ORIGINAL RESEARCH article

Front. Psychol., 12 November 2013

Sec. Psychology of Language

Volume 4 - 2013 | https://doi.org/10.3389/fpsyg.2013.00848

Patient-related constraints on get- and be-passive uses in English: evidence from paraphrasing

$\r\nDominic Thompson*$ Dominic Thompson¹^*

S. P. Ling¹

Andriy Myachykov²

Fernanda Ferreira³

Christoph Scheepers¹

¹Department of Psychology, Institute of Neuroscience and Psychology, University of Glasgow, Glasgow, UK
²Department of Psychology, Northumbria University, Newcastle, UK
³Department of Psychology, Institute for Mind and Brain, University of South Carolina, Columbia, SC, USA

In English, transitive events can be described in various ways. The main possibilities are active-voice and passive-voice, which are assumed to have distinct semantic and pragmatic functions. Within the passive, there are two further options, namely be-passive or get-passive. While these two forms are generally understood to differ, there is little agreement on precisely how and why. The passive Patient is frequently cited as playing a role, though again agreement on the specifics is rare. Here we present three paraphrasing experiments investigating Patient-related constraints on the selection of active vs. passive voice, and be- vs. get-passive, respectively. Participants either had to re-tell short stories in their own words (Experiments 1 and 2) or had to answer specific questions about the Patient in those short stories (Experiment 3). We found that a given Agent in a story promotes the use of active-voice, while a given Patient promotes be-passives specifically. Meanwhile, get-passive use increases when the Patient is marked as important. We argue that the three forms of transitive description are functionally and semantically distinct, and can be arranged along two dimensions: Patient Prominence and Patient Importance. We claim that active-voice has a near-complementary relationship with the be-passive, driven by which protagonist is given. Since both get and be are passive, they share the features of a Patient-subject and an optional Agent by-phrase; however, get specifically responds to a Patient being marked as important. Each of these descriptions has its own set of features that differentiate it from the others.

Introduction

A transitive event is one involving two participants: an Agent, the “doer” of the action, and a Patient, the person or thing that “undergoes” the action¹. In English such events can be described in Active-voice or Passive-voice. While active is the canonical form, the passive serves distinct specific functions.

The function of the passive-voice in general is to allow focus to be directed toward a specific element, in a manner similar to other topicalizing constructions such as clefting (Keenan and Dryer, 2006). In the case of the passive, it is the Patient of a described event that is elevated above other elements; that is, the syntactic prominence of the Patient in the passive is utilized in communicating some form of significance. As Keenan and Dryer (2006) note, this has the complementary effect of allowing the backgrounding of another element, namely the Agent. While the Agent occupies subject position in active-voice descriptions, in the passive it is reduced to an agentive by-phrase, as in Mary was hired by the manager. The prominence of the Agent can be further reduced by removing the by-phrase entirely, as in Mary was hired.

It is notable that the passive is further divided, providing the options of be-passive and get-passive. Some have dismissed these two versions as equivalent, both syntactically (Chomsky, 1981) and semantically (Weiner and Labov, 1983). However, literature from both Linguistics and Psychology points to these two passive-types having their own distinct uses, structures, and connotations, as we will discuss in the subsequent sections.

Many distinguishing factors have been suggested, from the nature of the verb or the described event, to the language modality or variety of English. The passive Patient is frequently cited, though there are widely differing opinions as to the precise attribute of the Patient motivating the use of one passive-type over the other. Here, we aim to address this situation by experimentally investigating the role of the Patient, and how that role may differ between the two passive-types; get and be.

We will show that be-passives and get-passives have complementary responses to various Patient-related attributes: get displays a preference for important or focussed Patients, regardless of which protagonist is given, while be displays preference for a given Patient, regardless of any marked importance on Agent or Patient.

We will argue that active-voice, be-passive, and get-passive may be conceptualized as three distinct forms of transitive description, rather than primarily involving a voice distinction (active or passive), with passive-type (be or get) being a minor and largely syntactic matter.

It is well-established that the passive-voice is the sub-dominant form for descriptions of transitive events in English. It has also been found that passives are more likely in written as opposed to spoken language. Chafe (1982) notes that passives are as much as five times more common in the written modality. Likewise, in a study utilizing the BNC, Brown, Switchboard, and Wall Street Journal corpora, Roland et al. (2007) found that passives were less common in spoken data. Biber (1993) notes that passives are much more common in scientific writing than in spoken conversation or fictional writing.

Mair and Leech (2006) examined British and American English corpora: LOB, F-LOB, Brown, and Frown. They report a decline in the use of be-passives over time, as well as a rise in the use of get-passives. Interestingly, this pattern is observed in both British and American English. However, it does not seem to be the case that the be-passive is simply losing ground to the get-passive, since in raw numbers the fall of be far outweighs the rise of get. The authors relate this change to an apparent shift in written English toward the norms of spoken language.

While passives are less common than actives, within the passive it is get that is least frequent. Carter and McCarthy (1999) considered the CANCODE spoken English corpus and found just 139 instances of get-passive from a sample of 1.5 million words. Xiao et al. (2006) also note far fewer examples of get-passives than be-passives in both F-LOB and BNC.

Collins (1996) reports that get-passives are less common in formal language settings. This is compatible with Biber et al. (1999) who claim that get-passives are almost exclusively found in conversation. Likewise, Mindt (2000) reports that get-passives are mostly found in spoken language.

The most frequently reported feature of the get-passive in corpus literature is that it tends to appear more frequently without an agentive by-phrase. Collins (1996) as well as Carter and McCarthy (1999) found that more than 90% of get-passives did not include a by-phrase. Mindt (2000) reports 82%; while Rühlemann (2007) states that get-passives included a by-phrase with a frequency of just 0.079 per 1000 utterances. Xiao et al. (2006), who looked at F-LOB and BNC, state that agentless passives were the most common form for both get and be, with no strong difference between them in terms of by-inclusion. However, Guoliang and Lei (2010) report fewer by-phrases in get-passives than in be-passives for both British (BNC) and American (COCA) English.

With regard to the specific function and meaning of get-passives, there are numerous partially conflicting or even completely opposing claims. They range from get and be equivalence (e.g., Weiner and Labov, 1983) or non-equivalence (e.g., Lasnik and Fiengo, 1974; Chappell, 1980; etc.), to specifics such as get-passives communicating primarily negative outcomes (e.g., Sawasaki, 2000) or equally communicating both negative and positive outcomes (e.g., Sussex, 1982; Givón, 1993; Sasaki, 1999). Recent work (Thompson and Scheepers, 2013) has suggested a theoretical model to account for be-passive and get-passive syntax and semantics in a more parsimonious manner by way of a shared structural component.

Meanwhile, the majority of experimental data on get-passives is concerned with child language use, rather than adult use. As a result, the research questions addressed are primarily linked to development, such as the age at which children understand passives or can be primed to use passive-voice vs. active-voice (see Thompson, 2012 for a summary). With so much unresolved regarding how get-passive usage and semantics differ from the be-passive, there remains a great demand for further investigation.

Within the discordant arguments in the literature that try to separate the get-passive from the canonical be-passive, one aspect that is often discussed is the nature or role of the Patient. While there is not a great deal of agreement on the specifics, the Patient is frequently accorded some manner of “special” role or focus in the get-passive. It is the more prominent of these Patient attributes that we address in the present work.

Hatcher (1949) asserts that the Agent in the get-passive holds a subordinate role, which, as a result, gives the primary role to the Patient. For Palmer (1974), get communicates an action as well as the state that results from it, implying a strong transitivity or change of state for the Patient.

Several authors use the term “affected,” though not consistently. In general, if an entity is affected, it is understood as undergoing some experience resulting from the action described by the main verb. Authors including Sasaki (1999) believe that the get-passive suggests greater subject (i.e., Patient) affectedness. In a similar vein, Carter and McCarthy (1999) suggest that the get-passive has a tendency to focus on the event itself, along with the way in which the event impacts the Patient.

Cameron (1990) suggests that, to warrant the use of get, it is not sufficient for the Patient to simply be affected; rather the Patient must be “materially affected,” that is, the event must involve a material action, as opposed to an emotion or thought. Orfitelli (2011) even claims that “the ‘affectedness’ requirement is so strong that predicates that do not affect their internal argument are typically illicit with the get-passive, although they are allowed in the be-passive.” The latter two claims are actually not well-supported by our own corpus searches. A brief interrogation of COCA (COCA, 2012) or BNC (BNC, 2012) indicates that get-passives are commonly used with “non-affecting” verbs, including Orfitelli's specific example. Conducting a search in COCA for “[get] seen” (i.e., “lemma GET immediately preceding the exact form seen”), returns results such as “14 billion videos get seen on YouTube every month,” “it might make it slightly easier for me to get seen,” “it would not get seen by anyone,” etc. This precise and constrained search returns 34 tokens, with comparable results for numerous other “non-affecting” verbs, such as watched or spotted. Clearly, get-passives are not impossible with such verbs.

Get-passives are also widely conceptualized as communicating a sense of adversity for the Patient of the action. For some, adversity can only refer to events with negative outcomes, for example Sawasaki (2000), who suggests this tendency is most apparent with human protagonists. Carter and McCarthy (1999) also suggest that get-passives communicate adversity; however, they specify that the adversity is interpreted by the speaker, rather than by the actual Patient of the action.

Other authors hold a broader view of adversity, taking it to refer equally to both positive and negative outcomes (Chappell, 1980; Sussex, 1982; Siewierska, 1984; Givón, 1993; Gronemeyer, 1999; Sasaki, 1999). The adversity may be due to a sense of struggle; McIntyre (2005) describes the get-passive as suggesting “the result is hard to attain.” Sussex (1982) suggests that, while get-passives can communicate both positive and negative events, they have more semantic flexibility when they are negative.

Many authors note some implication of initiative, control, or responsibility in get-passive descriptions (among many others, Hatcher, 1949; Lakoff, 1971; Barber, 1975; Givón and Yang, 1994; Downing, 1996; Sasaki, 1999). A similar feature is noted by Vanrespaille (1991), who suggests that “resultativeness” (an action that leads to a result that cannot be undone) is a major feature of the get-passive, and also that the Patient is at least partly responsible for the occurrence of the action. Arrese (1999) notes this “partial responsibility” of the Patient. Sussex (1982) claims that get-passives can imply varying degrees of purposefulness, blame, and responsibility, as well as other meanings.

Lasnik and Fiengo (1974) conceptualize the semantics of control in a rather more complementary manner between the two passive-types. The authors suggest that a passive sentence formed with get implies Patient control, while the same sentence formed with be implies Agent control. For example, they state that (1) implies that the Patient, John, intended to cause the event; while in (2), it is the Agent, Mary, who intended to cause it.

(1) John got fouled by Mary on purpose

(2) John was fouled by Mary on purpose

Givón (1993) provides comparable examples.

According to Cameron (1990), the Patient's responsibility in the get-passive comes from association with causative get (as in He got her fired), not from any inherent feature of get-passive semantics. A similar claim is made by Hatcher (1949), suggesting that the responsibility is extended from the overt responsibility of reflexive get (as in He got himself fired).

Another potentially highly relevant constraint on passivation as a whole, is information structure. In the linguistic literature, information structure is a term that is used to cover numerous related concepts such as topicality, givenness, and focus within a wider discourse. These concepts are also used with varying definitions. Birner and Ward (1998) generalize givenness as the level of availability of information. For our purposes we consider givenness to represent what is old or new to a hearer: information that is old is given; it is the entity that is the topic of a sentence or discourse, and has been available for longer. Once an entity has been established as given (for example, A new delivery boy arrived at our house.), any time that it re-occurs in the same discourse, it is likely to be referred to with a pronoun (e.g., He …).

This distinction of given vs. new is important here because given referents in a discourse tend to assume a prominent syntactic role, typically the subject position in English. Therefore, the givenness or newness of the Agent or Patient of an action should influence the selection of active-voice vs. passive-voice, with a given Patient increasing the use of passives. Relatedly, in a study by Meints (2003), scene descriptions were prompted via Patient-directed questions. This established the Patient as referentially given (Gundel, 1988, 1999), and resulted in passive responses being the most frequent (60.8% overall).

In this paper, we aim to contribute experimental data to these discussions. We use a paraphrasing task, since givenness and focus manipulations can be achieved through basic linguistic attributes, such as clefting as discussed below. Using linguistic rather than visual stimuli (e.g., picture description) avoids any influence of visual prominence, and allows a natural way to introduce one protagonist before the other. We also utilize Patient-directed questions, but with additional implications regarding the Patient's role in the event.

First, we test the impact of Givenness on the description of transitive events. Experiment 1 considers the effects of given information, vs. new information, on the rate of passive production, establishing baseline probability levels for active- vs. passive-voice (and the various passive-types) as a function of information structure.

Following this, we combine Givenness with additional manipulations in order to investigate specific aspects of the Patient in be-passives and get-passives. Experiment 2 considers a general sense of importance or focus applied to the Patient established via clefting (which is another type of topicalizing construction; see above), while Experiment 3 investigates potential effects of implied blame, control, or responsibility on the part of the Patient, which we take together as forms of “Patient agency.”

Experiment One

Introduction

Experiment 1 manipulated only givenness of the Agent or Patient. This allowed us to establish baseline levels of active-voice vs. passive-voice, and be-passive vs. get-passive, as a function of information structure. This serves as a comparison for the findings of Experiments 2 and 3, in which additional experimental factors are manipulated.

Participants

Twenty-four native English speakers (age 18–56, mean age 24; 29% males) were tested in individual sessions, with each lasting approximately 30 min. They all gave informed consent and received subject payment or course credits for their participation. All participants were recruited through the University of Glasgow's subject database².

Stimuli

Twenty-four short stories were created (as in 3), each consisting of two sentences. The second sentence described a transitive event in active-voice, while the first provided a preamble to the event. One of the two protagonists of the event, namely either the Patient (3a) or the Agent (3b) of the event in the second sentence, was introduced in the first sentence and was therefore given; i.e., when participants encountered the sentence describing the transitive event, one of the two protagonists was already known (given), while the other was not previously known (new). This constituted the conditional manipulation information status.

(3a) The cowboy rode across the desert into the small dusty town.

When he arrived a thief attacked him.

(3b) The thief made her way through the sandstorm to a small town.

When she arrived she attacked a cowboy.

The transitive description was always in active-voice, meaning that any increase in passive production would be motivated by the experimental factor. This produced a two-level design with two Information Status levels. A full list of materials is given in the appendix (Table A1).

Procedure

All experiments within this paper employed a paraphrasing paradigm. In this instance, the 24 (items) × 2 (conditions) were assigned to 2 separate lists such that each item appeared precisely once per file, and in a different condition in each of the two files, using a Latin square. This resulted in twelve items per condition per file, ensuring an equal frequency of each condition in each file. Further to the 24 critical items per file, 50 filler items were also included. These fillers varied structurally with various features to distract participants from the intentions of the study. This gave a total of 74 trials per file.

The task was presented on a 12· LCD monitor running at 60 frames per second and was run using SR Research Experiment Builder. Participants interfaced with the task using a keyboard. The spacebar was used to advance through trials and to advance from one screen to the next within trials. A short practice session preceded the main experiment to familiarize participants with the procedure and type of sentences they would be encountering in the experimental trials and fillers.

Trials proceeded as shown in Figure 1. They always began with a central fixation cross. The next screen displayed the preamble sentence in the center of the screen. After reading this aloud, participants pressed the spacebar to advance. The next screen displayed the transitive event sentence in the center of the screen. Again, this sentence was read aloud. After a brief (500 ms) pause, the next screen displayed a prompt for participants to retell the transitive event described in the second sentence in their own words. They responded to this out loud, and then pressed the spacebar to end the trial and begin the next.

FIGURE 1

Figure 1. Example of paraphrasing trial procedure.

Participants' spoken responses were audio recorded on the experimenter computer and coded for “voice” at three levels: active (transitive sentence in active-voice, with the subject referring to the Agent and the direct object to the Patient of the critical event), be-passive (sentence in passive-voice using a form of be as auxiliary verb and with the subject referring to the Patient of the critical event), or get-passive (sentence in passive-voice using a form of get and with the subject referring to the Patient of the critical event). Passive responses were also coded for agentive by-phrase inclusion (i.e., whether the sentence contained a prepositional phrase headed with by that referred to the Agent of the critical event). Each participant was presented with one of the two files (74 trials), which were split into 3 blocks, allowing for breaks to maintain attentiveness.

Results

In this experiment we were concerned with the production of syntactic alternatives, thus it was deemed necessary for participants to demonstrate the availability of at least two forms. Those who did not produce any passive responses were discarded and further participants were tested to replace them; in this instance, only one participant was replaced. Following this, the data were filtered before analysis. Responses that were not transitive were coded as errors and were discarded. Less than 2% of the data were excluded.

Table 1 provides the raw descriptive counts for five response types (active-voice, be-passive with or without an agentive by-phrase, get-passive with or without a by-phrase) in each condition (given Agent/given Patient).

TABLE 1

Table 1. Cross-tabulation results giving raw counts of responses (AV, active-voice; B0, be-passive without by-phrase; BB, be-passive with by-phrase; G0, get-passive without by-phrase; GB, get-passive with by-phrase) in each of the two conditions (given Agent, given Patient).

Table 1 indicates a strong overall preference for active-voice responses, especially in the presence of a given Agent. Get-passive responses were extremely rare, always including an agentive by-phrase when they occurred. While be-passive responses were less frequent than active-voice responses overall, in the presence of a given Patient, be-passives became slightly more frequent than active-voice.

To statistically corroborate these observations, we employed Generalized Estimating Equations (GEE; e.g., Hanley et al., 2003; Hardin and Hilbe, 2003). GEE is an extension of Generalized Linear Models (Nelder and Wedderburn, 1972) that is capable of handling repeated-measures designs as well as mixed designs. Unlike ANOVA—but very much like, e.g., Generalized Linear Mixed Models (GLMMs)—GEE can statistically accommodate a wide range of different data types (categorical frequencies, ordinal data, continuous data, etc.) by providing a variety of appropriate distribution and link functions. However, contrasting with GLMMs, separate by-participant and by-item analyses need to be carried out in GEE, analogous to F1 and F2 in ANOVA. One advantage of GEE over GLMMs (especially when the latter are used in their recommended “maximal” form, see Barr et al., 2013) is that GEE is computationally less complex and more likely to converge on the type of data we consider here (an issue that becomes more severe with the more complex experimental designs that we introduce later).

Here, the GEE model was used to predict the likelihood of producing a passive-voice (vs. active-voice) paraphrase as a function of Information Status. (Due to rare occurrences of B0, G0, and GB responses, all passive responses were combined into a single category.) Since the dependent variable was binary, a binomial distribution and logit link function were used (thus implementing a binary logistic regression model). To investigate whether effects generalize across participants and items, two types of binary logistic GEE analyses were carried out: the first took Information Status as a within-participant factor and the second as a within-item factor, each time assuming an exchangeable covariance structure for repeated measurements.

Generalized Score Chi-Squares derived from this analysis showed a clear effect of Information Status both by participants [χ²₍₁₎ = 15.440, p < 0.001] and by items [χ²₍₁₎ = 13.879, p < 0.001], whereby passive-voice responses were reliably more likely when the Patient was given (0.54 ± 0.10) (mean ± SE by participants) than when the Agent was given (0.22 ± 0.07).

Discussion

In Experiment 1, passive-voice responses were overall less frequent than active-voice responses. This is not surprising, since active is the canonical form of transitive description and there was no explicit instruction or other encouragement to utilize the passive-voice.

The givenness manipulation produced a very clear effect, with passive use increasing when the Patient of the event was given (i.e., the Patient was introduced earlier in the discourse). This is in agreement with existing subject-assignment literature (see Myachykov et al., 2011): that is, the assignment of a protagonist to subject position drives subsequent structural selection. In this instance, whichever protagonist was given in the stimulus, participants were likely to maintain as the subject in their paraphrase. The most frequent way of topicalizing a protagonist (in English) is via assignment to sentential subject position (others, such as clefting, are much less common; see Keenan and Dryer, 2006). When the subject is the Agent of an action, the canonical active-voice is selected; when the subject is the Patient, passive-voice is selected.