Writing a Literature Review: Three Tasks, Not One

If you supervise student research projects, you might recognise the feeling: a few weeks into the supervision cycle, you receive drafts in your inbox, and you can tell that the student has spent much time reading; you can see that the referencing works most of the time; and (if you are lucky in this GenAI era), the sources may even be real, recent, and relevant. And yet, somehow the text just does not work: You read on and on, and several pages into the text, you still cannot tell what any of it is for. Each paragraph reports something that somebody has written about the topic, but the paragraphs could be shuffled into a different order or even deleted without much loss.

I used to struggle with such chapters because I didn’t always know how to best advise the students who wrote them. What follows is an attempt to address this.

Three types of text

When I was less experienced, I would respond to these drafts by asking students to focus more. This, I realise now, cannot have been terribly helpful, because the student had already focused as hard as they knew how. What they did not do effectively was to reflect on the purpose of the text.

The mistake was that they approached the literature review as a single task, perhaps that of proving familiarity with the field. A more useful perspective, I think, would be to distinguish between three tasks, with three different purposes, and as a consequence three different relationships to the study that follows.¹

These are: contextualising information, which tells the reader where they are; the theoretical framework, which tells the reader how you propose to see whatever it is that interests you; and the survey of recent research, which establishes what is known, and therefore what is not. I will try to separate these in the paragraphs that follow, in the way I wish I was able to explain to these students in the past.

Contextualising information

The first move in a literature review (whether located in a chapter named ‘Literature review’ or in the introductory chapter) is to orient readers by answering the questions where are we? and how is this place different from the ones we already know about? In this move, the task is one of describing the setting, the population, the relevant policy or institutional arrangements, and the scale of the phenomenon that we want to study. The purpose of this move is entry. That is to say, a competent reader from an adjacent subfield should be able to follow the rest of the study because of this section, without having look elsewhere for information.

Let’s suppose you’re writing a thesis on, say, the use of AI in English Language Education for refugees and migrants in Greece. Even before you start writing, typically, you have a lot of knowledge that helps you make sense of what you observe. This includes, among other things, the structure of the education system, what classes refugees attend, the usual teachers’ background, and the infrastructure one might encounter in a normal Greek school. Your reader² doesn’t have all this background. This is especially true for readers who engage with your work years after you’ve written it, or readers in other countries. It is therefore your job to orient them.

What to look for when contextualising your work

The kinds of sources on which you should draw here are the ones that many literature reviews do not value: ministry circulars, curriculum documents, statistical yearbooks, census data, institutional reports, and even occasional newspaper pieces can all be valuable sources for this move.³ Recency matters more than scholarly pedigree here, because what you are describing is a changing state of affairs. A 1980 article is a perfectly acceptable source for a theoretical claim, but it’s a poor source for, say, average class size, or teacher perceptions.

One test I propose to students is subtraction. Look at every sentence, claim, and paragraph you’ve written. Now try deleting it, and ask yourself whether the paper has become harder to enter or merely thinner. Padding makes this section weaker, because –just like a dense forest– the context becomes harder to navigate. If removing a sentence costs the reader nothing, it was never doing contextual work, whatever its subject matter.

What contextualising sections look like

When writing this section, aim for declarative and unhurried prose. Short sentences in simple present and present perfect work well. Crucially, the citations here are non-integral: the source sits in brackets at the end, because it functions as evidence that something is the case, not as a voice in a conversation.

Compare the two following examples:

Since 2016, reception classes for pupils with migrant backgrounds have operated in parallel with mainstream provision (Ministry of Education, 2023).

and

Kostoulas (2023) argues that reception classes have operated in parallel with mainstream provision since 2016.

The first version reports on a fact. Whatever your position might be on Greek statistics, it’s part of our shared knowledge background and probably the most authoritative source on a topic. The second version transforms this fact into a contested claim, and in doing so it invites readers to wonder who disagrees.

What to avoid when contextualising

A common mistake here is the dreaded funnel that begins at the dawn of time. “Language has always been central to human societies” and “The first debates on the nature of language can be traced in Socrates’ thought as reported in Cratylus (Plato, 360 B.C.E., 383a–384c)” are not context. At best, they are throat-clearing, and frankly, readers have seen such claims so many times that they are just annoying.

You should also avoid introducing contested claims into this section. If a point needs defending, then that’s not context, and it belongs further down.

Theoretical framework

The second, and perhaps most challenging, part of a literature review is the presentation of the theoretical framework that informs the study. This is where you stop describing and start committing. Choosing to frame a study through, say, sociocultural theory rather than cognitivism, is a positioning move that tells readers what you will consider data, how you will connect them, and what is just background variation or ‘noise’. This is the section where you make this commitment visible, and present to the readers the specific conceptual apparatus (i.e., the terms and constructs) that your analysis will later use to interpret what it finds.

Many students approach me with research ideas about language learning motivation. Motivation, however, is not something you can describe in the same way you draw a glass of water on a table (and even then, an impressionist would see the same glass differently from a realist or a modernist). You may choose to view it from a socio-educational perspective (contrasting instrumental and integrative motives), or from the lens of self-determination theory, or through the L2 Motivational Self-System. Or, if you are ambitious enough, you may want to create a new perspective that picks the best parts of all of the above and weaves them into a new theoretical lens. Whatever you choose, you need to articulate this early on and share it with readers.

In my experience, this is the section students struggle with the most, among other reasons, because it asks them to take a position before they feel ready. Feeling uncertain is normal, and taking your time to get things right is often a good strategy. What is less defensible is committing to a theoretical path that you are not yet ready to describe or defend.

What to look for in a theoretical framework

Counterintuitively, your hardest thinking will be informed by your most straightforward reading. Rather than reading many sources (often in summary), look for few sources, and read deeply.

When making your theoretical commitment, read few sources and read deeply.

Wherever possible, go to the originating text, and read enough of it to know what problem the concept exists to solve. If you want to use Complex Systems Theory in Applied Linguistics, go to Larsen-Freeman and Cameron (2008). For ecological theory, read Bateson (or van Lier, if you’re brave enough). If you want to try Intentional Dynamics, get my book with Juup Stelma. Concepts drift as they travel from study to study, and picking up a concept third-hand from someone else’s literature review risks missing the qualifications and nuance that made it useful.

This section is challenging and rewarding. The starting point is to decide on the central concept that you want to understand, describe or explain. Let’s assume it’s ‘integration’. Then think about the elements and processes that directly connect to it.⁴ What makes integration easier or harder? What happens to people when they have integrated into a society? Then make sure that you have properly introduced every concept and explained the relationship among them. That’s your framework.

You need to imagine Sisyphus happy

When you’ve finished the study, you will almost certainly have to return to this section and streamline it. I often use ‘Chekhov’s gun’ as an analogy here. Pavel Checkov used to say that if you plan to write a gunshot in Act III, you should show the audience a gun in Act I, and (conversely) introducing a gun in Act I commits you to a gunshot by the end of the play. The same applies here: every element in your theoretical framework must appear again in your analysis and discussion. A concept that is defined beautifully and never returns is not framework. It is a side dish.⁵

What the theoretical framework looks like

When writing your theoretical framework, integral language works well. This involves putting the author’s name in the sentence itself, rather than tucking the citation into a bracket at the end. This helps because it keeps the attention focused on the argument you’re developing and its connections to the informing theory.

Some people will disagree,⁶ but I would also suggest first-person writing for two reasons: first, because this is you committing to a theory, and hiding behind passive and third-person constructions seems disingenuous. And secondly, because the harder the concepts you describe, the simpler the language should be.

Here’s an example:

Following Larsen-Freeman (1997), I treat classroom interaction as a dynamic system, which is to say that I do not assume that its outcomes are proportionate to its inputs. This has a specific consequence for the analysis: it means that atypical cases will be treated as evidence of a process that is inherently non-linear.

Note the second sentence. The whole point of presenting a theoretical framework is to commit to its consequences. Whether you signal these explicitly with phrases like “this means that” or not, every claim in this section must do this kind of theoretical work. If you cannot answer the “what does this mean for my analysis” question, your framework is not ready yet — it’s just a reading list.

What to avoid when writing a theoretical framework

A common pattern among weak theoretical frameworks is the definitions parade (“AI has been defined as x (Author A, 1998), Author B 2004) defines it as y, and most recently Author C and D (2011) defined it as z”) followed by nothing. Surveying definitions is useful only if it ends in a choice, and the choice is the whole point of the survey.

Consider the following extract from Wittgenstein’s Lecture on Ethics. Notice that he begins with a parade of definitions (‘ethics is the enquiry into what’s vaulable/good/important’), but crucially, he develops out of this a new point, which I have bolded. If you want to present readers with many definitions of the same thing, use this diversity to make a point — otherwise you just sound indecisive.

Synthesising different perspectives

Now instead of saying “Ethics is the enquiry into what is good” I could have said Ethics is the enquiry into what is valuable, or, into what is really important, or I could have said Ethics is the enquiry into the meaning of life, or into what makes life worth living, or into the right way of living. I believe if you look at all these phrases you will get a rough idea as to what it is that Ethics is concerned with.

Now the first thing that strikes one about all these expressions is that each of them is actually used in two very different senses. I will call them the trivial or relative sense on the one hand and the ethical or absolute sense on the other.

Survey of recent research

The third major move in a literature review is the survey. This is a section where you map what we collectively know about whatever it is you want to research. Paradoxically, however, the point of this section is not to establish what is known, but to trace the contours of the unknown.

A literature survey exists to trace the contours of the unknown

If you are familiar with Doctor Who, the following analogy might work. In Clara’s Diner, the Doctor narrates his experience in Gallifrey to an attentive waiter. Clara has been wiped out of his mind, but by narrating the story in detail, he is able to reconstruct her from the contours of the space she occupied.

Your job here is similar: by outlining the borders separating known from unknown, you need to convince readers that something remains undone, and that it is worth doing. This is the ‘warrant’ of your research, i.e., the reason why you invest time doing it, and the reason why readers should invest time in reading about it.

What to look for in a survey of recent research

The selection test here is positional overlap. A source belongs in this section if it occupies (or nearly occupies) the space you are claiming. You need to consider if it asks a version of your question, if it studies a comparable population, or if it uses a design whose limitations provide the starting point for your design. If a source merely shares a keyword with yours, then it belongs (at best) in the context section.

When thinking about what to include in this section, you may find it helpful to read the limitations and future-research sections of the studies closest to yours. These sections, which you will find at the end of most articles, are where researchers declare what they were unable to do. What you will also want to do is group information as you read. That is, try to sort studies not by author or date of publication, but by what they found and by where they disagree. This grouping will eventually become the structure of your survey.

Defining a gap and claiming it

When thinking about the warrant of your research, you may want to remember that there are three kinds of ‘gaps’ in knowledge. The first type is an empirical gap, where no research exists about a setting or population, and there is theoretical reason to expect something different there. Then there are theoretical gaps, produced when existing findings conflict and no available framework reconciles them. And finally, there are methodological gaps, where findings look as though they might be artefacts of design decisions in past studies. The strongest reviews name which kind they are claiming, and go on to explain how they will address it.

Type of gap	Definition	Warrant
Empirical gap	A setting or population hasn’t been examined, and there’s a theoretical reason to expect something different there	Justifies studying that specific setting or population, on the grounds that existing findings elsewhere may not transfer
Theoretical gap	Existing findings conflict, and no available framework reconciles them	Calls for work that resolves or explains the conflict, rather than just adding another data point
Methodological gap	A finding looks like it might be an artefact of research design	Calls for testing the finding with a different method to see if it holds

What literature surveys look like

The most effective literature surveys are synthetic and comparative. By this I mean that the information appears in sections organised by claim, with studies as evidence for the claim. Compare the following examples:

Reported gains in self-efficacy are consistent across secondary settings (Author A, 2019; Author B, 2021; but see Author C, 2023, for a rare example of negative effects ). However, the two studies conducted with adult learners found no comparable effect (Author E, 2020; Author F, 2022). Both, however, relied on self-report measures administered immediately after the intervention, which leaves open whether the difference lies in the learners or in the instrument.

Author A (2019) conducted a quasi-experimental study among secondary school students. By comparing pre- and post- test results, she found modest gains […] In a study conducted in 2020, Author C examined adult learners. Participants in this study used questionnaires to report their perceived self-efficacy levels of at the start and end of a school year. His results indicate […]

In the first example, the gap follows from the paragraph. The second one is just a list that leaves it to the reader to find the pattern and identify the gap from it.

What to avoid in literature surveys

As hinted above, you will want to avoid the annotated bibliography march: Author (2019) found… Author (2020) found… Author (2021) found… This is the single most common failure mode, and what makes it weak is that no paragraph has a subject of its own, only a sequence of other people’s.

Another common problem is claiming a gap by virtue of absence of evidence alone. “No study has investigated the wellbeing of CLIL teachers in Greece” may well be true, and is not by itself a reason for anyone to care.⁷ The claim has to be completed: We have reasons to believe that CLIL teachers in Greece are different from CLIL teachers in other settings, and other language teachers in Greece, for whom research exists, because… Absence of research is a fact about the literature. A gap is a claim about knowledge.

Concluding thoughts

Just to clarify, none of this requires three separate chapters, and I would gently discourage anyone from imposing the structure mechanically. Sources move between the categories, too: a study can supply a contextual fact early on and reappear later as the near-miss whose limitations generate the warrant. This is normal, provided you know which job it is doing each time and write the sentence accordingly.

What produces the unusable draft I described at the start is not poor reading, and not too few citations. It is writing every paragraph in the same register, so that describing a setting, committing to a framework and arguing for a gap all sound like the same activity, which is reporting that something exists in the literature.

So the question to hold while drafting is not ‘have I read enough?’, which has no meaningful answer, but a narrower one, asked paragraph by paragraph: what is this paragraph doing, and would anyone notice if I just deleted it?

---

Notes

1. I am describing functions, not headings. In some traditions these appear as three sections, in others as two chapters, and in a short article they may be four paragraphs. The distinction I am drawing is about what the writing is doing, not about where the subheadings go. ↩︎
2. Do not assume that the primary reader of your thesis is an examiner. The life of a paper extends long beyond the viva. ↩︎
3. This is also why the contextual section ages fastest. If there is a part of a dissertation that will need rewriting between submission and publication, it is usually this one. ↩︎
4. Different theories will have different names for these: behaviourists might think of ‘antecedents’ and ‘outcomes’; in CDST, we think of ‘shaping influences’ and so on. I will use ‘elements and processes’ as a sort of shorthand. ↩︎
5. I have nothing against side dishes, but I am told by my doctors that I should avoid them. And if I can’t enjoy them, neither will you. ↩︎
6. Because they are wrong :p ↩︎
7. Researchers in the psychology of language learning seem to be especially prone to this. In the attempt to carve a sub-discipline into being, lots of studies make claims to very questionable empirical gaps: Do French as a foreign language teachers experience stress in different ways from teachers in ELT? Do language teachers in one Chinese town have different wellbeing from teachers in another Chinese town? Maybe they do, but it’s the writers’ job to explain rather than assume agreement. ↩︎

---