South African research abstracts hedge 25% more than the rest of the world’s, not less — and they make evidence claims 8% less often. The one measure on which they run ahead is unverifiable importance adjectives, which are 35% above the world rate.
That is the finding from 7 433 abstracts of journal articles published in 2025, retrieved from the OpenAlex API on 24 August 2026: 3 873 with at least one South African institutional affiliation and 3 560 drawn from the same population without a country filter. Every figure below is computed from those abstracts, and the word lists and method are published so that you can disagree with them.
The headline numbers
| Measure | South African-affiliated | World sample | Source and year |
|---|---|---|---|
| Abstracts analysed | 3 873 | 3 560 | OpenAlex, articles published 2025, retrieved 24 Aug 2026 |
| Words analysed | 886 814 | 751 809 | Same |
| Median abstract length (words) | 220 | 200 | Same |
| Hedges per 1 000 words | 5,13 | 4,09 | Same |
| Boosters per 1 000 words | 3,55 | 3,56 | Same |
| Hedge-to-booster ratio | 1,45 | 1,15 | Same |
| Abstracts containing at least one hedge | 57,3% | 45,6% | Same |
| Abstracts containing at least one booster | 49,4% | 44,8% | Same |
| Abstracts containing at least one importance word | 27,8% | 19,9% | Same |
| Significant, significantly, significance per 1 000 words | 2,26 | 2,51 | Same |
The first thing to note is what the data does not show. The common assumption — that researchers writing in an additional language state findings too flatly and hedge too little — is not visible here at all. It is contradicted. South African abstracts hedge substantially more than the comparison sample, and a considerably higher share of them contain at least one hedge.
The second is that total boosting is essentially identical: 3,55 against 3,56 per 1 000 words, a difference of a third of one percent. If you stopped there you would conclude that there is no difference in how strongly South African research writing makes claims. That conclusion would be wrong, because the total is hiding two movements in opposite directions.
Where the difference actually sits
Splitting the boosting vocabulary into families locates it. These family lists are defined separately from the headline booster list and overlap it only in part, so their rates do not sum to it.
| Booster family | South Africa, per 1 000 words | World, per 1 000 words | Difference |
|---|---|---|---|
| Importance — essential, crucial, vital, critical, imperative, indispensable, paramount | 1,70 | 1,26 | +35% |
| Evidential — demonstrate, prove, confirm, establish, clearly, conclusively, evidently | 1,44 | 1,57 | −8% |
| Essential, crucial and vital only (excluding the ambiguous word critical) | 0,92 | 0,72 | +28% |
Read those first two rows against each other, because the whole story is in the contrast.
Evidential boosters are lower in South African writing. Abstracts with a South African affiliation say demonstrated, confirmed and established at 1,44 per 1 000 words against 1,57 in the world sample. On the claims that attach directly to data, South African research writing is measurably more cautious than the international average, not less. The significant line points the same way: 2,26 against 2,51, about 10% lower.
Importance boosters are higher, and by a wide margin. They run 35% above the world rate, and the proportion of abstracts containing at least one importance word is 27,8% against 19,9%. Because critical is genuinely ambiguous in scientific writing — critical temperature, critical care, critical path, critical theory — the figures were recomputed with it removed. The gap narrows but does not close: essential, crucial and vital alone run 0,92 per 1 000 words against 0,72, a difference of 28%.
The individual words
The three most frequent hedges are the same in both samples and in the same order: potential, may and could. In the South African sample potential alone accounted for 1 044 of the hedge tokens across the two runs, making it by a wide margin the most common single hedging word in South African research English.
Among importance words, the ranking differs. The South African sample runs critical, then essential, then crucial, then vital. The world sample runs critical, then essential and crucial at roughly half the South African rate, then vital at well under half. Rarer words behave the same way: imperative and indispensable are uncommon in both samples but consistently more common in the South African one.
What was counted, and how
Transparency matters more than usual here, because a hedging count is only as good as its word list.
Hedges (43 forms counted): may, might, could, appear(s/ed), seem(s/ed), suggest(s/ed), possible, possibly, probable, probably, likely, unlikely, potential, potentially, perhaps, relatively, somewhat, apparently, presumably, assume(d), tend(s/ed), approximately, largely, generally, typically, usually, often, arguably, plausible, plausibly, indicative, broadly, partially, seemingly.
Boosters (41 forms counted): demonstrate(s/d), prove(s/d/n), clearly, strongly, definitely, undoubtedly, certainly, obviously, evidently, must, always, never, highly, substantially, considerably, remarkable, remarkably, novel, unprecedented, superior, excellent, outstanding, establish(es/ed), confirm(s/ed), conclusively, undeniably, markedly, greatly, tremendous, vital, crucial, essential.
Matching is on whole lowercase word forms after stripping punctuation, so there is no substring noise. Rates are expressed per 1 000 words of abstract text, which controls for the difference in abstract length between the two samples — South African abstracts run about 10% longer at the median, and an unadjusted count would have reported a spurious gap on every measure.
Abstracts shorter than 40 words were excluded as truncated. Two independent random samples were drawn with different seeds and pooled; the samples were not matched on discipline.
“Significant” was deliberately excluded from the booster list. In a results abstract it usually carries a precise statistical meaning rather than a rhetorical one, and counting it as a booster would inflate every scientific corpus at once. It is reported separately in the headline table.
How stable are these figures?
Every headline measure was computed twice, on independent random samples, before pooling. The agreement is close:
| Measure | Run 1 | Run 2 | Source |
|---|---|---|---|
| South African hedges per 1 000 words | 5,07 | 5,18 | OpenAlex, retrieved 24 Aug 2026 |
| World hedges per 1 000 words | 4,09 | 4,09 | Same |
| South African importance boosters per 1 000 words | 1,72 | 1,68 | Same |
| World importance boosters per 1 000 words | 1,22 | 1,30 | Same |
The direction and rough size of every reported difference reproduced across both runs. That is a statement about sampling stability only. It says nothing about the four limitations below, which are not reduced by drawing a larger sample.
What this measurement does not establish
- Affiliation is not authorship nationality. The South African sample is defined by at least one South African institutional affiliation on the record, which includes internationally co-authored papers and excludes South African researchers working abroad. It is a reasonable proxy and not the same thing.
- Abstracts are not dissertations. An abstract is the most compressed and most heavily edited text a researcher produces, and much of the South African sample will have passed through journal editing. The pattern may be stronger or weaker in a dissertation chapter, and this study cannot say which.
- Field composition differs between the samples. No attempt was made to match the two samples by discipline, and disciplines differ in their baseline use of both families. South Africa’s research output is weighted differently from the world’s — more health, education and social science, less of some engineering literatures — and part of the measured gap may be composition rather than national style. A field-matched comparison would be a stronger design and is not what was run.
- A rate is not a judgement. Nothing here shows that the South African rate is wrong and the world rate right. What it shows is a measurable difference in one family of words, which matters mainly because international reviewers and external examiners read against the second distribution.
The wider caution about counting South African research output — and why two honest sources give different totals for the same year — applies here as it does to the publication counts in our analysis of DHET accredited journals and research output units.
What it means for your own dissertation
The actionable version inverts the advice most students are given. Your hedging is probably fine. Your importance adjectives probably are not.
An importance claim is unfalsifiable. Nobody can check whether a topic is crucial, which is exactly why an examiner discounts the sentence — and why a paragraph that opens on one starts from zero credibility. The replacement is not a weaker word. It is a fact.
| Instead of | Write |
|---|---|
| Employee retention is a crucial issue for South African municipalities. | The municipality recorded a vacancy rate of X% in critical skills posts in 2024, against Y% nationally. |
| It is essential to study learner performance in rural schools. | Two-thirds of the district’s learners are taught in schools without a subject specialist, and no study has measured the effect. |
| This study is of vital importance to policymakers. | The programme this study evaluates reaches 4,2 million beneficiaries and has not been evaluated since 2019. |
| The results clearly demonstrate a strong relationship. | The results indicate a moderate association (r = 0,41). |
The pattern is the same in every row: a number, a comparison or a documented absence replaces an adjective. That is also the sentence an examiner cannot argue with.
The opposite failure is real too
Because these data show South African writing hedging above the international rate, the mirror-image error deserves naming. A finding buried under three hedges — “the results may possibly suggest a potential association” — reads as though the writer does not believe it, and an examiner will discount it exactly as the sentence invites them to.
One hedge that accurately marks the uncertainty is the target. Not none, and not three. Where a result is genuinely robust, say so once in plain language and let it stand; the strength of the verb you attach to it is a separate decision, set out in our verb bank for a South African dissertation.
A ten-minute check on your own abstract
- Search for seven words: essential, crucial, vital, critical, imperative, indispensable, paramount. In an abstract of about 220 words — the South African median here — expect roughly zero to one.
- For each hit, ask what fact it is standing in for. Write the fact instead. If there is no fact, the sentence was doing no work.
- Then search the evidence family: demonstrates, proves, establishes, confirms, clearly. Keep the ones your design licenses and drop the rest a rung.
- Count your hedges, in both directions. Zero across a whole abstract usually means a finding has been stated more firmly than the design allows. Three stacked in one sentence means the opposite.
- Repeat on your conclusion only. Abstracts and conclusions carry the overwhelming majority of both families, and they are the two sections examiners and readers actually read in full.
Claim strength inside the discussion chapter is a related craft with its own rules — bounding each finding, handling results that contradict the literature, and writing limitations that strengthen rather than apologise — and that is worked through in our guide to writing the discussion chapter. If the abstract you are auditing is being written in an additional language, the register questions that sit alongside all of this are in our guide to writing a dissertation in a second language.
Writing at the strength your data licenses
Calibration is hard to apply to your own writing, because the sentence that overclaims felt true when you wrote it. It is much easier when the claim sits next to the evidence it rests on rather than in a different file.
Tesify builds your chapters with every claim attached to its source, so the distance between what a study found and how strongly you have stated it is visible while you write rather than in the examiners’ report. There is a free tier to start on. The judgement about how far to push a finding stays yours.
Write your dissertation with Tesify
Frequently asked questions
What is a hedge in academic writing?
A word or construction that limits how strongly a claim is made — may, might, suggests, likely, potentially, appears to. Hedging is not weakness; it is the mechanism by which a writer states exactly as much as the evidence supports and no more.
What is a booster?
The opposite move: a word that strengthens a claim beyond its plain statement — demonstrates, clearly, undoubtedly, crucial, essential. Boosters divide into evidence claims, which a study can support, and importance claims, which nothing can verify.
Do South African researchers hedge less than researchers elsewhere?
No. In this sample South African-affiliated abstracts carried 5,13 hedges per 1 000 words against 4,09 in the comparison sample, a rate 25% higher, and 57,3% of them contained at least one hedge against 45,6%. The common assumption that researchers writing in an additional language under-hedge is not visible in these data.
So where is the difference?
Overall boosting is level — 3,55 per 1 000 words against 3,56. But that total hides two movements in opposite directions. Evidential boosters run 8% lower in the South African sample, while importance boosters run 35% higher. South African abstracts make fewer claims about what was shown and more claims about why the topic matters.
Which words drive the importance gap?
Essential, crucial and vital. Counted together and excluding the ambiguous word critical, they occur at 0,92 per 1 000 words in the South African sample against 0,72 in the world sample, a difference of 28%. Counting critical as well widens the overall importance gap to 35%.
Why was “significant” left out of the booster count?
Because in a results abstract it usually carries a precise statistical meaning rather than a rhetorical one, and counting it would inflate every scientific corpus at once. It is reported on its own line instead, where the South African rate is 10% lower than the world rate.
Where does the data come from?
The OpenAlex API, filtered to journal articles published in 2025 with an abstract present, retrieved on 24 August 2026. The South African sample additionally required at least one South African institutional affiliation. Two independent random samples were drawn with different seeds and pooled.
How stable are these numbers?
They were computed twice on independent random samples before pooling, and the two runs agreed closely. South African hedging came out at 5,07 and 5,18 per 1 000 words in the two runs, and the world figure came out at 4,09 in both.
Does this mean my dissertation should hedge less?
Not as a general instruction. These data suggest the caution in South African research writing is real and mostly appropriate, and the more likely problem in a given dissertation is an unfalsifiable importance adjective standing in for a fact. Check your importance words first and your hedges second.
Can under-hedging and over-hedging both be problems?
Yes. An unhedged finding claims more than the design licenses and invites an examiner to test it. A finding buried under three hedges reads as though the writer does not believe it, and examiners discount it accordingly. The target is one hedge that accurately marks the uncertainty, not none and not three.
Is affiliation the same as the author being South African?
No, and this is the main limitation. The sample is defined by at least one South African institutional affiliation on the record, which includes internationally co-authored papers and excludes South African researchers working abroad. It is a reasonable proxy and not the same thing.
Can I cite these figures in my own literature review?
Yes, with attribution to this page and the retrieval date of 24 August 2026, and read the limitations section first — particularly that affiliation is not author nationality and that the two samples were not matched by discipline.