2. Describing and defining plagiarism

Plagiarism has widely been described (e.g., Eaton, Reference Eaton2021; Howard, Reference Howard2000) as difficult to define, perhaps almost to the point of impossibility. Elsewhere, I have taken issue with this contention (Pecorari, Reference Pecorari, Pecorari and Shaw2019), arguing that it is entirely feasible to define plagiarism in a robust and defensible way; the difficulty lies in defining it in terms that can readily be applied in practice, in order to determine which acts should be classified as plagiarism. The feasibility of a definition does not, however, imply a generally held consensus about what constitutes plagiarism; there is ample evidence of uncertainty and inconsistency in the academic community (Borg, Reference Borg2009; Pecorari & Shaw, Reference Pecorari and Shaw2012).

The lack of an unambiguous, commonly held understanding may well be related to a degree of vagueness or lack of specificity in the definitions many universities include in their policies on plagiarism (Eaton, Reference Eaton2017; Pecorari, Reference Pecorari, Belcher and Hirvela2001), but it is debatable which is the cause and which is the effect. That is to say, academics may have contradictory ideas of what plagiarism is because their universities fail to provide them with explicit and unambiguous definitions; or the definitions may be vague and unspecific because they are written by committees of academics who cannot agree on the details.

The inconsistency of academics’ views about plagiarism is all the more surprising because concise definitions of plagiarism tend to be very similar, suggesting broad agreement at a generic level. Two definitions taken essentially at random from university websites illustrate this:Footnote ¹

Both definitions are typical in describing plagiarism as involving the re-use of material originally produced by someone else, with the re-use happening in an inappropriate way. Those three criteria are widely accepted; much more contentious is a fourth criterion, the question of intention; that is, whether the misuse of the appropriated material must be an act of intentional wrongdoing (Pecorari, Reference Pecorari2013). Indeed, the University of Sydney definition continues with, ‘It doesn't matter whether you do this on purpose or accidentally’. Because of the significant disagreement in the academic community about whether intention to deceive is part of the definition, I have proposed (Pecorari, Reference Pecorari2003, Reference Pecorari2008) the umbrella term textual plagiarism, a term that embraces two subcategories. One subcategory is prototypical plagiarism, a form of cheating that involves deliberate deception. The acts that fall into the other subcategory are not motivated by deceptive or transgressive intent, acts that, in order to distinguish them from prototypical plagiarism have been given labels such as textual borrowing (Barks & Watts, Reference Barks, Watts, Belcher and Hirvela2001) and patchwriting (Howard, Reference Howard1995).

For the purposes of this article, unless otherwise indicated, plagiarism will be used to refer to this broader category, textual plagiarism. This is not because the question of intention is unimportant; it is of great importance both in understanding the causes of plagiarism and in meting out just and effective responses to it. It is also of great interest in setting out a research agenda for plagiarism from the perspective of the teaching and learning of language, because one of the subcategories – patchwriting – is especially closely intertwined with the development of language skills. However, intention is also very difficult to diagnose,Footnote ² and it must be considered that a text that is labelled as plagiarism within an instructional setting may have been misdiagnosed, or there may be insufficient information to make that diagnosis. In using plagiarism in that broader sense here, the intention is to exploit the neutrality ‘textual plagiarism’, which is agnostic on the question of intent.

It should also be noted that plagiarism can involve not only text in the narrow sense of ‘words’ but also many other kinds of material, including computer code and musical or artistic works (Eaton, Reference Eaton2021). However, in researching plagiarism from the perspective of language learning and language teaching, it is text in that narrower sense, texts made up of words, which is most relevant, and that will be used here.

Intentionality is problematic in large part because of an open disagreement amongst academics about whether it should be regarded as a criterion, but it is also problematic because of the uncertainty and subjectivity in deciding about it. How can we reliably distinguish between a student who made a genuine mistake and one who merely claims to have done so, to escape the consequences of an act of intentional wrongdoing once it is discovered? The same uncertainty and subjectivity apply to the other criteria as well. For example, how do we know whether a chunk of text has been copied or not? Two features are often referenced to help make that judgement: the extent of the (potentially) copied material, and the uniqueness of the phrasing. However, there are no accepted benchmarks for either characteristic. What volume of similarity should be regarded as suspicious and how unique is a unique phrase? These subjectivities can make it difficult to determine whether plagiarism has happened.

Plagiarism also comes in many forms. The following are examples (not a comprehensive list) of acts that have been described as plagiarism:

• • handing in an assignment identical to one that another student submitted for the same class in a previous semester;

• • handing in an assignment identical to one that the same student submitted for a different class in a previous semester;

• • including several copied paragraphs in a text that otherwise appears to have been written by the person whose name is on the cover page;

• • including one or more paragraphs from a source with changes – more or less liberal – to aspects of the wording; for example, replacing words with synonyms, changing the order of items in a list, rewriting to make a sentence more concise or more elaborated.

The variety of different textual acts that come under the heading plagiarism can make it a slippery construct to discuss; the word can evoke very different understandings in different people.

For the reasons described above – because the textual indicators of plagiarism are numerous and varied – studies that have aimed to document the relationship between new texts and their sources have been challenged to describe their findings. Different approaches have been taken, which presents an obstacle to comparisons amongst texts in the same study and across studies. For example, Gilmore et al. (Reference Gilmore, Strickland, Timmerman, Maher and Feldon2010) used text-matching software to identify plagiarism, and described the proportion of texts in their research corpus in which it was found. They did not, however, describe the threshold at which they considered that matches constituted plagiarism, so a researcher who wished to replicate the study, or repeat it in a different context, would be hard put to compare their results. (Of course, there is an inherent difficulty in setting such a threshold, because the proportion of overlapping text is not the sole consideration in determining whether a text is plagiarised.)

By contrast, both Howard et al. (Reference Howard, Serviss and Rodrigue2010) and Pecorari (Reference Pecorari2003) inspected texts manually. The latter looked for five categories of source use, two of which – patchwriting and copying not signalled as quotation – could be understood as plagiarism (although it should be noted that Howard (Reference Howard1999) rejects the idea that patchwriting should be considered a form of plagiarism). They also considered whether citations were present for propositions that needed them, and whether the content of the source was reported correctly. They reported the proportion of texts having or lacking one of these features.

Pecorari (Reference Pecorari2003) also considered the accuracy of reported content, along with two other features: source identification (were propositions from a source attributed to the correct source in the target text?); and the origins of the wording (i.e., if the phrasing used was taken from the source, was that signalled?) The extent of source dependence was quantified with three metrics: the overall proportion of words in the target text that originated from the source; the proportion of words that were copied from a source but not signalled as quotation; and the length of copied strings.

Keck (Reference Keck2006) investigated paraphrasing strategies and took as her basic unit of analysis the ‘unique link’, a lexical word or phrase that could be traced to one (only) part of a source text. She grouped paraphrases into four categories based on the proportion of words in the paraphrase that consisted of unique links, ranging from ‘substantial revision’ (no unique links) to ‘near copy’ (50% or more of the words in the paraphrase were unique links). She also documented ‘exact copies’; that is, passages in which paraphrasing was not attempted. Wiemeyer (Reference Wiemeyer2020) built on Keck (Reference Keck2006) by examining paraphrase and exact copy, but also looked at source use strategies such as quotation and summary.

In short, different approaches to measuring and describing potential plagiarism have been taken, and this makes it difficult to compare findings across studies. The first task, therefore, is to create and validate a descriptive rubric, the terms of which can be used to describe the relationships between new texts and their sources. In terms of development, carrying out this task could involve looking at the studies that have attempted to describe intertextual relationships to produce a list of potential descriptors (e.g., proportion of copying, extent of changes to the copied chunks, acknowledgement or not of the source relationships). Testing or validating the functionality of the rubric could take different forms, but one productive approach would be to apply the rubric to several texts and see if practitioners find that the descriptions thus produced facilitate their determinations about whether the texts use sources appropriately and, if not, whether the problematic usage constitutes plagiarism. This approach to validation illustrates the potential pedagogical applications such a rubric would provide, alongside research applications. The tendency on the part of busy practitioners to place overreliance on the metrics produced by text matching software has frequently been observed, and significant problems with that overreliance have been documented (e.g. Canzonetta, Reference Canzonetta, Pecorari and Shaw2019; Weber-Wulff, Reference Weber-Wulff, Pecorari and Shaw2019a, Reference Weber-Wulff2019b). A rubric that facilitates a fuller understanding of intertextual relationships could be a useful aid.

3. The prevalence of plagiarism

It is not uncommon to encounter claims in the literature that the incidence of plagiarism has increased dramatically. Sometimes these claims are accompanied by a reference to the internet as an explanation, apparently on the basis that the internet makes it easier for students to find sources from which to copy. These claims are widely disseminated, despite the fact that relatively little research has attempted to address the question, and that which has (e.g., Curtis & Tremayne, Reference Curtis and Tremayne2019; Curtis & Vardanega, Reference Curtis and Vardanega2016) does not provide strong evidence for such a contention.

As Kaposi and Dell (Reference Kaposi and Dell2012) note, the discourse of these claims merits interrogation. The notion of the supposed dramatic increase in plagiarism is used to suggest an urgent need to address the problem. The same is true for another discourse about plagiarism: the public scandal. When a politician, author, or other celebrity is involved in allegations of plagiarism, the ensuing public discussion of it similarly tends to invoke the episode as an illustration of a serious social malaise.

If all textual plagiarism were prototypical plagiarism – that is, if every unattributed quotation or patchwritten paragraph were a sign of dishonesty and a deliberate attempt to gain by cheating – then it would be reasonable to interpret documented instances of it as a worrying sign, and to take an increase in textual plagiarism (if true) as evidence of a growing problem. However, as noted above, it is widely accepted that some textual plagiarism is a developmental issue, so it might be expected to be as common as the pool of novice academic writers is large. Understanding more about the real incidence of textual plagiarism – how much of it there is, and who does it – can potentially shed light on the developmental/patchwriting explanation, and that in turn can lead to a better understanding of how to address it. The more widespread it is, the more evidence there is that the current prevailing strategy – ‘the triad of prevention, detection and punishment’ (Kaposi & Dell, Reference Kaposi and Dell2012, p. 813) – is of limited efficacy. The next two research tasks take two different approaches to investigating the prevalence or incidence of textual plagiarism.

Many studies have described intertextual relationships in student writing that could be identified as plagiarism (e.g., Abasi et al., Reference Abasi, Akbari and Graves2006; Angélil-Carter, Reference Angélil-Carter2000; Ouellette, Reference Ouellette2008), producing rich understandings of intertextual relationships in the context of the writers’ perceptions, experiences and identities. Fewer studies have attempted to quantify these intertextual relationships, either in terms of how much dependence on sources can be found in individual texts, or in terms of how many texts in a corpus contain textual plagiarism. The Citation Project (e.g., Howard et al., Reference Howard, Serviss and Rodrigue2010; Jamieson & Howard, Reference Jamieson, Howard, McClure, Purdy and McClure2013) used a corpus of writing produced by undergraduates at a US university, the majority of which were believed to have L1 English. Pecorari (e.g., Reference Pecorari2003, Reference Pecorari2006) examined texts written by postgraduates at a UK university, all of whom had another L1 than English. Both studies identified a widespread propensity toward intertextual relationships that could be diagnosed as plagiarism. In addition, they found other features, such as distortion of the meaning of the source material, which would not ordinarily be regarded as plagiarism, but that do suggest that the disciplinary literacy skills of the writers were still under development.

The more widespread textual plagiarism is, the greater the need to approach it proactively rather than reacting to it when it is discovered. The greater the evidence that it is associated with the development of academic literacy, the clearer it is that supporting academic writers to become more skilled and confident will be effective in preventing textual plagiarism. More evidence is needed to help understand how pedagogical resources can best be focused.

The second research task, therefore, is to extend the work of Howard et al. and of Pecorari to other contexts. Of particular interest is the rapidly growing area of English as a medium of instruction (EMI). In EMI settings, most teachers as well as learners find themselves in the position of engaging in the complexities of academic discourse in a language that is not their first. The ramifications of this for textual plagiarism/patchwriting bear investigation. More generally, additional evidence about how students use sources from other educational levels, geographical settings, academic disciplines, and other languages than English would be of great value in shaping a broader understanding of the phenomenon.

Such an investigation could involve an open approach, in which authentic assessment texts (e.g., thesis chapters or essays written for a particular class) could be used, or a closed approach, in which research participants would all be given the same topic to write on, and potentially a small number of texts to use as sources. The choice of approach would involve weighing up the relative merits of a more naturalistic investigation versus the benefits of having a small pool of potential sources and greater comparability across the texts.

Note: For tasks such as this, which involve a comparison of texts with their possible sources, the comparison can either be manual or can be performed with text-matching software (i.e., products such as Turnitin). An automated approach has the advantage that a larger sample can be analysed against a larger volume of possible sources. However, such tools can give unhelpful results, including false positives and false negatives (see Weber-Wulff, Reference Weber-Wulff, Pecorari and Shaw2019a, Reference Weber-Wulff2019b). A manual analysis can produce a more nuanced understanding of the relationship between texts, particularly if the writer has made extensive changes to the source language, but the time-consuming nature of a manual comparison limits the amount of text that can be analysed.

When allegations of plagiarism are made against high-profile individuals, the effect is often to discredit the individual. Joseph Biden's first presidential campaign was scuppered on accusations of plagiarism (Dionne, Reference Dionne1987). A biographer of Martin Luther King Jr. speculated that had the plagiarism scandal surrounding his university work become public sooner, it could have irreparably damaged the US civil rights movement (McKnight, Reference McKnight1998). If plagiarism is understood as an act of deliberate wrongdoing, then reputational damage is a reasonable outcome for what one would hope are a small group of individuals who choose to perpetrate it. However, the idea of patchwriting as a developmental stage that novice academic writers tend to pass through suggests that it is likely to be widespread. If that is the case, then singling out prominent individuals for criticism is not only unfair, but it also deflects the criticism away from the educational setting that failed both to identify students using sources inappropriately and to support them in learning to create intertextual ties appropriately.

This task starts with a high-profile case of (alleged) plagiarism involving student writing. For the sake of feasibility, it should be a genre that is archived by the university, such as the doctoral thesis, and the task will be greatly facilitated if, as part of the public attention the case has received, the alleged plagiarism has been analysed and described. For example, The Martin Luther King Papers Project released a statement analysing Dr King's source use (Reference King1991), and a body of research and commentary has sprung up around it (e.g., Miller, Reference Miller1991; Reagon, Reference Reagon1991). Germany has famously had many cases involving prominent politicians, and many of these are documented in the VroniPlag wiki (https://vroniplag.wikia.org/de/wiki/Home). Other cases can readily be found in news archives.

The task begins with collecting a selection of (sections of) doctoral theses from the same university or one that can be regarded as similar, and from the same time period and academic discipline, and then analysing their source use. The intention is to see whether the high-profile case is an isolated exception in producing the intertextual relationships that have been labelled plagiarism, or whether those source-use strategies were common among similar students. If it is the latter, that suggests that more attention should be given to prevention (including teaching the skills that allow writers to use sources in less controversial ways) and/or to gatekeeping than to retrospective criticism.

4. Causes of plagiarism

When plagiarism is understood as a deceptive act of deliberate cheating – that is, prototypical plagiarism – then causes can be sought amongst the factors known to promote academic dishonesty (see Eaton, Reference Eaton2021, pp. 49–51 for a review). This section is specifically concerned with the other sort: the intertextual relationships that may be labelled plagiarism but are not the product of an intention to cheat or deceive.

Several factors have been suggested as causes for this non-deceptive act, and they can be grouped into three broad categories. The first is an imperfect understanding of the expectations of the disciplinary discourse community. Gilmore et al. (Reference Gilmore, Strickland, Timmerman, Maher and Feldon2010) found an inverse relationship between the presence of plagiarism in postgraduate student writing and disciplinary expertise as indexed by two factors: the amount of time students had spent in postgraduate study; and the presence of references to primary literature, which they argued was a disciplinary expectation, and Keck (Reference Keck2014) had a similar finding with respect to undergraduates. Pecorari (Reference Pecorari2006) found that for aspects of source use that are visible to a reader (e.g., the preference in some disciplines for research articles over books), postgraduate writers approximated the disciplinary norms. For other, less visible features (such as whether text not in quotation marks was actually copied from a source), they were more likely to deviate from the disciplinary norms, because the lack of visibility meant that the students were not alerted to those norms. Davis (Reference Davis2012) found that international students at a UK university had gaps in their understandings about plagiarism and reported that their training in using sources had been minimal. In sum, the evidence supports the idea that students who are aware of what their academic disciplines expect, either because they have been instructed in it, or because they have learnt it through exposure, are less likely to plagiarise.

A second explanation places the responsibility for plagiarism with cultural differences. More specifically, the basic notion is that cultures vary in terms of which acts are considered transgressive. The cultural explanation exists in many variations; one commonly expressed version is the idea that students from certain cultures (Confucian-heritage cultures are frequently named) are trained in rote learning and repetition, but not that originality is required in their work. Another version is that students from cultures with a high power distance are taught to have great respect for teachers. From this viewpoint, the practitioner, who is wiser and more widely read than the student, will certainly recognise material repeated from a source. Explicitly marking it, for example with a citation or quotation marks, is not only unnecessary, it is potentially offensive.

The cultural explanation for plagiarism has generated a significant debate (reviewed in Pecorari & Petrić, Reference Pecorari and Petrić2014) and has been subjected to an extended analysis (Li & Flowerdew, Reference Li, Flowerdew, Pecorari and Shaw2019) in the Chinese context. However, little empirical research has addressed it directly. One exception is an experimental study (Bikowski & Gui, Reference Bikowski and Gui2018) in which Chinese university students (some of whom were in China, and some in the US) were shown videos of a writer working from sources. The videos illustrated inappropriate, questionable or appropriate use of the sources, and after viewing each one, the students were asked to describe what had happened. The analysis focused on the connotations of the words and phrases used in the description. They found that the US-based students were more likely to use Chinese words that translate as plagiarism or otherwise have a negative connotation, while the Chinese-based students were more likely to use a neutral term. The authors interpret this as evidence of a shift in understandings of plagiarism associated with exposure to the US educational system; this in turn suggests that the two cultures may have different understandings.

In the Japanese context, Wheeler (Reference Wheeler2009) asked students to assign scores to texts created for the purpose of the study. They assigned a high score to the first. They were then shown the ‘source’, and this allowed them to see that the first text had copied a great deal from it. Asked to assess it again, they gave it a lower score. Next, they were shown a second text that used the same source, but with non-identical wording; that is, it used the source in a way that was arguably less problematic than the first. They rated it more highly that the first. As Wheeler (Reference Wheeler2009, p. 26) notes, the findings ‘suggest that one should be careful about concluding that plagiarism is inherent in Japanese culture’.

Keck (Reference Keck2014) re-analysed the data used in her 2006 study of paraphrasing to look at differences between L1 and L2 users of English, and between first-year students and more advanced students. She found the latter factor – experience of university study – to be associated with the tendency to rely on the words of source authors and found little evidence to support the idea that copying is more common amongst L2 users.

Rather than expectations – either disciplinary or cultural – language skills (and whether they are adequate to meet expectations) are the focus of a third cluster of explanations for non-prototypical plagiarism. Flowerdew and Li (Reference Flowerdew and Li2007) studied Chinese postgraduates in the sciences as they wrote articles intended for publication. The students perceived a need to improve their English, and adopted strategies toward that end:

Notably, when the first strategy was used, the ‘useful expressions’ were rather short and generic chunks; for example, ‘In this article’ or ‘Here we report’ (p. 449), chunks that would in all likelihood be considered appropriate to reuse by most academics (cf. Davis & Morley, Reference Davis and Morley2015). The second strategy, however, resulted in long and quite specific strings being reproduced from articles into the study participants’ texts. The authors do not conclude from this that the science students committed plagiarism; indeed, they argue that the standards of the humanities regarding plagiarism cannot and should not be applied to the sciences. However, in universities with one-size-fits-all plagiarism policies and in which academic conduct committees are drawn from across the university, the risk is significant that this copying behavior could be adjudged plagiarism. The students’ choice to adopt these strategies suggests a perceived need to look for a supply of academic words and formulae outside of their own mental lexicons.

For Flowerdew and Li's (Reference Flowerdew and Li2007) students, productive language skills were at issue, but receptive skills have also been implicated in textual plagiarism. For example, Jamieson and Howard (Reference Jamieson, Howard, McClure, Purdy and McClure2013) found that students misreported the ideas they took from sources; that they reported ideas at sentence level, rather than the larger point of the source as a whole; and that the sentences they worked with tended to come from the first page or two of the source. They interpreted these findings as casting doubt on the students’ reading skills.

These three possible sets of explanatory factors for patchwriting – that students need to learn about disciplinary expectations for academic writing; to learn the expectations of a new culture; and to learn to understand and produce academic discourse in an L2 – are naturally not mutually exclusive. Nor are they unrelated; indeed, they might be expected to develop together, as incidental outcomes of being in a given academic setting. However, as indicated above, very little empirical evidence exists to show what their respective and relative contributions to patchwriting might be. The next tasks are designed to produce such evidence.

As noted above, relatively little research has attempted to investigate empirically the possible role of cultural differences in explaining student plagiarism. One of the few exceptions is Wheeler (Reference Wheeler2009). He asked a group of Japanese students to perform judgement-elicitation tasks based on three short texts on the same subject, two of which were purported to have been written by fictional students and one of which was said to have been published in an academic journal. Participants were first presented with the student text and asked to evaluate it, which they did generally positively. They were then shown the authoritative, published text, and at this point it became clear that the first ‘student’ had copied from it. When asked to re-evaluate the first student text, they were much more negative, and specifically criticised the copying, characterising it as plagiarism. Participants were then given and asked to evaluate the second ‘student’ text, which put forward the content of the ‘published’ text but the wording was to some extent changed. They were negative about the clear dependence on the source, but less negative than toward the first, more source-dependent text.

The procedure of having participants evaluate the first text twice, once in ignorance of its source and then again in awareness of it, isolates the plagiarism as an element in their judgement. Their criticism in the second round of evaluation indicates that they were not inclined to condone plagiarism. Their more restrained criticism of the second ‘student’ writer, who varied the wording of the source, can be interpreted as evidence that they valued the effort to write more autonomously. All of this makes it difficult to argue that Japanese culture does not regard plagiarism with disfavour, leading Wheeler to conclude that ‘it is a lack of understanding of the act, rather than cultural values, that is the root cause of plagiarism committed by students’ (Reference Wheeler2009, p. 17).

Research task 4 is to reproduce this study in a different geographical or cultural setting, in order to provide more direct evidence speaking to the cultural explanation. Although analysis and commentary from both cultural insiders and external observers (e.g., Scollon, Reference Scollon1995; Ha, Reference Ha2006) is of value, it currently exists in much greater volume than empirical data. Reproductions of Wheeler's study in other contexts can redress this imbalance.

Wheeler (Reference Wheeler2009) gives the short texts in an Appendix and provides a clear and detailed description of his methods, making it feasible to follow them closely. The participants in his study were L1 speakers of Japanese. If a multicultural group of participants is available, the judgement-elicitation procedure could be combined with a short questionnaire asking about participants’ backgrounds, thus permitting possible patterns to emerge in the analysis.

This task involves examining the extent of patchwriting in a writing sample in the light of the receptive and productive language skills of the writer who produced it. Many variations on this basic theme are possible, some of which are described below.

A writing prompt is first constructed. Participants should be asked to write on a topic that could be addressed by consulting sources, but for which they are not strictly necessary (i.e., the topic does not require significant technical or background information). An opinion text, such as ‘Does the world need more vegetarians?’ would be suitable. A small number of sources – possibly just one – relevant to the topic are supplied to participants. The number, length and readability of the sources can be determined with reference to participants’ language level. Participants are first given time to read the source(s) and are then given the prompt and asked to write on it. Their texts are then analysed to identify whether and how they have used the source(s): is language from the sources re-used? If so, how much, and to what extent is it adapted (e.g., by substituting synonyms) as opposed to reused verbatim? Are the intertextual relationships acknowledged appropriately or not? Depending on the length of the source text(s), the first part of the analysis, the comparison, may be done manually or using text-matching software.

To measure productive language skills, a very simple approach would be to analyse the same text for any of a range of features that are thought to correlate with the development of writing skills. These could include length; sentence complexity; lexical richness; the presence of academic vocabulary; error analysis; and so forth. A possible objection to this procedure could be that some of the language in the participants’ texts would likely come from the sources, so it would be the ability of the authors, rather than the participants, which was measured to some extent. Against this is the fact that integrating copied chunks into a text is time-consuming and can be difficult, so writers who depend more on their sources may also introduce errors, write shorter texts, and so forth. However, if it is possible to secure participants who are willing to invest additional time, a good solution would be first to ask them to write one text without recourse to sources, and subsequently to perform the source-based writing task. The first, non-source-based text would provide the material to assess their productive skills.

It would also be possible to test productive knowledge of grammar and vocabulary, both highly relevant to writing skills, by a number of means, including a Cloze test and, for English, the Productive Vocabulary Levels Test (Laufer & Nation, Reference Laufer and Nation1999) or the Productive Academic Vocabulary Test (Pecorari & Malmström, in preparation). Formulaic language is also highly relevant (cf. Simpson-Vlach & Ellis, Reference Simpson-Vlach and Ellis2010) and various approaches to testing it – both productive and receptive – exist (Gyllstad & Schmitt, Reference Gyllstad and Schmitt2018).

Receptive language skills can be measured in similar ways. For English, possible measures include a test of reading comprehension and reading speed, such as the Nelson-Denny; a test of receptive general vocabulary, such as the Vocabulary Levels Test (e.g. the versions produced by Schmitt et al., Reference Schmitt, Schmitt and Clapham2001); and/or the Academic Vocabulary Test (Pecorari et al., Reference Pecorari, Shaw and Malmström2019).

All other things being equal, more varied measures of receptive and productive language skills will give richer and more reliable results, while also being more time-consuming both for the researcher and participants, thus possibly limiting the number of results. Through a combination of one or more of these measures, it will be possible to understand the relationship between source-dependent writing and receptive and productive skills in the target language.

This task also begins with the analysis of source-based writing, and follows the first steps in Task Five, creating a writing prompt, asking participants to read one or more sources on the same topic before producing a writing sample, and analysing the writing samples to see how they have used sources.

The next step is to conduct text-based interviews with the writers. The attention of each writer is directed in the interview toward specific places where language from the source(s) has been repeated, and the writer is asked to explain the choice to use that chunk of language in that way. The answers – including the answers to follow-up questions, which are likely to be needed – can include two possible explanations. One is that writers may have felt unsure about their understanding of the source, and as a result chose not to paraphrase more extensively, in order to avoid the risk of distorting the meaning. This would suggest that the decision to use the source's language lay in the participant's receptive skills. Alternatively, they may recount that they felt the idea was expressed very nicely and doubted their own ability to express it as well, indicating that productive skills were central to the decision to adopt a copying strategy.

Of course, some responses might provide a more nuanced blending of both these factors, and others could arise, and this is why a qualitative, interview-based approach is suggested for this task. The opportunity to hear unexpected explanations, and probe, follow up and negotiate understanding with participants has the potential to produce rich data speaking to this complex topic.

5. Teaching away from plagiarism

Over 30 years ago, Hull and Rose (Reference Hull and Rose1989) described a writing strategy that is very reminiscent of that which Howard came to call ‘patchwriting’. Their informant was a student in need of extra academic support as she learned to write her way into the discourse community of her chosen profession: nursing. One of her strategies was relying on the language of her sources but putting her imprimatur on it to the extent she could, by varying some wordings. The researchers observed that

Some years later, Howard commented that Hull and Rose

The idea that a ‘free-wheeling pedagogy of imitation’ can be beneficial is supported by the account of Li & Flowerdew's (Reference Li and Flowerdew2007) students, who mined the articles they read for the phrases they hoped to be able to use effectively and, perhaps, master. A similar possibility emerged from a case study of a bilingual secondary school pupil (Villalva, Reference Villalva2006). And yet more than 30 years on, we not only have not explored the potential benefits of encouraging a source-repetitive strategy for learning to write, but we have also arguably taken a step backward. Increased institutional attention to plagiarism, increased use of text-matching software, the tendency toward penalty tariffs for plagiarism and the like make it more difficult than ever to avoid treating patchwriting ‘negatively, as a “problem” to be “cured” or punished’. The final research task seeks evidence to guide pedagogy in this area.

This task is a longitudinal investigation of the writing development of one or more participants, in the light of their patchwriting behavior. Before describing the task, a note on its potential sensitivity is needed. The task will explore the effects of encouraging writers to adopt Hull and Rose's ‘freewheeling pedagogy of imitation’, and it is important that participants understand that the rules that apply in the study are different from the ones that apply in the classroom.

The project begins by recruiting participants prepared to spend approximately two hours on each of several occasions across the period of the study. As with any longitudinal study, establishing the precise parameters is an act of balancing the ideal with the realistic. A duration of an academic year, with four data collection points, would be sensible. If cooperation from participants can be gained, a longer period and/or more data collection points would be better still. A large group of participants is not necessary (a single case study could produce findings of value) but because of the risk of drop-out, erring on the side of caution and recruiting above target is advisable.

At each data collection point, participants will first read one or more sources and then write a text in response to a prompt on a related source. Each prompt should involve different topics but be similar in character (e.g., ask participants to write the same genre, a text of similar length, etc.).

After producing each sample, participants are asked to take part in a text-based interview. The first asks participants how they used the sources and, where relevant, why. Subsequent interviews will also ask about strategies named in earlier interviews. For example, if a student reports that ‘this sentence is a paraphrase of this part of the source’, the next interview might ask, ‘in the previous task you paraphrased from the source. Have you done that this time?’ If a student says, ‘On reflection, I think this paraphrase may be too close to the original’, a follow-up question could be ‘Last time you mentioned that you thought one of your paraphrases could be too close to the original. Do you have that concern this time?’

The relationship between each writing sample and its sources is analysed, and this will indicate whether the writers grow less dependent on their sources over time. The writers’ own accounts in the interviews will add an important additional set of information and perspectives. This will indicate whether giving novice academic writers permission to imitate is an effective learning strategy.

Two optional steps would make this study somewhat more complex procedurally, but also provide richer outcomes. The first would be to introduce a measure of the quality of the texts, ideally using at least two independent raters who are experienced in assessing academic writing. If their holistic rating of each writer's texts is higher over time, that is an additional indication that the license to engage in textual mimicry is beneficial.

A second step would be to ask participants to provide copies of the assessment writing they do during the duration of the study. As part of the analysis of each writing sample, the words and phrases which are copied from the source(s) are recorded. The texts subsequently produced by the participants, both for coursework and the later writing samples, can be profiled to see if the copied words and phrases become part of that writer's lexicon.

The findings would need careful interpretation, because some writing development would be expected over an academic year in any event. Equally, the absence of visible development would not necessarily mean that the license to copy produces no benefits; it could simply mean that benefits were too small to be measured by this method. However, both additional steps could potentially add evidence speaking directly to development, rather than examining it solely through the writers’ perspectives.

6. A note about research ethics

Any research project that involves human participants must be planned carefully to take their needs into account. In many jurisdictions, guidelines, including research ethics approval, are in place. The researcher's responsibility includes compliance, but extends beyond it to fulfilling a duty of care to participants.

Because plagiarism is a stigmatising act, and because the consequences can be severe, investigating it can be considered sensitive-topic research, and both the stakes and the researcher's responsibility are particularly high, and in each of the projects described above – and in others investigating the topic – risks are present.

There is no blueprint for determining the appropriateness of a project or the measures needed to protect against risk. However, researchers can look to three guiding points. One is the extent of the duty of care that the researcher has for participants. For example, research task 7 could potentially uncover plagiarism in the work of currently-enrolled students, while research task 3 proposes an investigation of doctoral theses, which are usually available to anyone who wishes to examine them. It would be reasonable to consider that the researcher's responsibility extends further to students who have been recruited for a study than it does to writers who have made a free choice, outside of the parameters of the project, to publish their work.

The researcher can also consider measures to protect participants from negative consequences. Source-dependent writing, if discovered in a text produced within a research project, is unlikely to have consequences, but if identified in the work students have produced for academic credit, the potential for harm is greater. Anonymity can be promised, but various factors, such as the size of the group of participants, affect the likelihood of participant identity being inferred.

A third point for reflection is the balance between beneficence and maleficence; in other words, does the potential benefit of the project outweigh the potential for harm? In the case of inappropriate intertextuality, a small risk to an individual participant might be tolerable if the findings will bring light to bear on a pedagogical problem that sees many students sanctioned at universities around the world.

The sorts of questions that these considerations give rise to have no single, easy answer. However, with sensitivity, reflection, consultation (with supervisors, colleagues, etc.) and adequate reading in the literature on research ethics (e.g., Farrimond, Reference Farrimond2013), answers appropriate to the individual case can be identified.