Talven Text: research and design
Research date: 2 October 2026. Release: 0.100, an independent prototype.
What the evidence supports
There is no sound basis to call Talven Text the world's most readable font. Readability depends on the reader, task, language, size, spacing, and device. The goal of this project is attractive, clear body text with useful reader controls. This release has not undergone a controlled reader study.
Controlled studies of OpenDyslexic and Dyslexie found no consistent general advantage over conventional fonts. Spacing can help some readers, but greater spacing is not always better. Personal preference and reading speed can also differ. These findings support an adjustable specimen and careful testing, rather than a promise attached to a font name. See the study table below.
Legibility means how well a reader identifies letters and words. Reading performance also includes speed, errors, and comprehension across passages. Comfort and preference are separate outcomes. This project does not treat them as interchangeable.
What was studied
This focused literature review prioritizes controlled experiments and separates study results from design proposals. It covers children and adults, readers with and without dyslexia, print and screen tasks, letterforms, spacing, size, and reader differences. It is not a registered systematic review or a new meta-analysis.
| Study | Sample and task | Main finding | Important limit |
|---|---|---|---|
| Zorzi et al., 2012 | 74 French and Italian children with dyslexia read sentences aloud. | Extra spacing improved speed and accuracy. | The main condition changed letter, word, and line spacing together. Printed sentences do not establish a web font optimum. |
| Wery and Diliberto, 2017 | 12 pupils with dyslexia completed letter, word, and nonsense-word tasks. | OpenDyslexic did not improve speed or accuracy over Arial and Times New Roman. | Small sample from one specialist school. These were fluency tasks, not extended reading. |
| Kuster et al., 2018 | One experiment included 170 children with dyslexia. A second included 102 with dyslexia and 45 without. | Dyslexie did not improve reading speed or accuracy. Font preference did not predict performance. | Dutch children, specific fonts, and imperfect size and spacing controls. Null results do not mean identical comfort for every reader. |
| Marinus et al., 2016 | 39 low-progress readers read four texts with size and spacing controls. | An initial difference favoring Dyslexie was no longer statistically significant when both letter and word spacing were matched. | Low-progress readers are not necessarily diagnosed with dyslexia. This result is verified from the publisher abstract. |
| Galliussi et al., 2020 | 64 children with dyslexia and 64 controls read eight texts. Letterform, letter spacing, and word spacing varied separately. | Specialist letterforms did not improve speed. Wider letter spacing without enough word spacing reduced speed. | Two spacing levels, Italian readers, and an oral task. This does not identify an optimum for all readers. |
| Rello and Baeza-Yates, 2013 | 48 Spanish readers with dyslexia read texts in 12 fonts with eye tracking. | OpenDyslexic did not improve reading time. Some fonts differed in eye-fixation measures. | Fixation duration is not reading speed. The discussion does not establish that all sans serif or monospaced fonts read faster. |
| Rello, Pielot, and Marcos, 2016 | 104 participants read Arial texts on a desktop monitor at six sizes. | Larger type reduced fixation duration and improved some comprehension results. | Size also changed characters per line. The paper's points and line-spacing units must not be relabeled as CSS pixels or CSS line-height. |
| Joo et al., 2018 | Adult analysis included 38 people. A separate child sample tested the crowding correlation. | Individual crowding measures predicted benefit from wider spacing, especially in pseudoword tasks. | The child study did not replicate the spacing intervention. The result does not imply one visual deficit or one spacing prescription for all dyslexic readers. |
| Wallace et al., 2022 | Main analysis of 352 online adults. Each read five fonts drawn from a larger set. | The fastest observed font differed across readers. Preferred fonts were not reliably fastest. | Participants who reported reading or learning disabilities were excluded. The fastest-versus-slowest contrast can include measurement noise and is not a promised gain from this font. |
Publication years refer to journal or conference publication. Some papers appeared online earlier.
Recent evidence
Machado, Elmore, and Hall, 2025 tested five adults with dyslexia in repeated nonsense-word tasks. The authors found no significant accuracy or speed differences among Dyslexie, Arial, Times New Roman, and spaced versions of the conventional fonts. Three participants preferred spaced Arial; two preferred Dyslexie. The small sample, remote testing, and artificial task limit transfer to web reading. Reporting inconsistencies in the grid size and statistical degrees of freedom also warrant caution. This study supports separating preference from measured performance.
Azzarello et al., 2026, published on 31 July, reports a meta-analysis of 15 studies, 91 effect sizes, and 688 participants. The abstract gives a pooled effect of g = -0.04, with a 95% confidence interval from -0.15 to 0.07. It reports no consistent speed or accuracy benefit from specialist fonts. Only the abstract, metadata, and references were accessible for this review. The full methods, treatment of dependent effect sizes, and bias analyses were not assessed. This is secondary evidence, not an additional participant trial. Its reported result agrees with the cautious design position above.
Where findings disagree
Zorzi reported a benefit from substantial spacing changes. Galliussi found no general benefit from its combined spacing condition and found a penalty when extra letter spacing weakened word boundaries. The studies used different amounts of spacing, baseline fonts, and tasks. Joo adds evidence that reader differences matter. These results do not support a fixed "dyslexia spacing" value. Zorzi, Galliussi, Joo.
Marinus is a useful warning about attribution. A difference between font packages can come from their spacing, not just the shapes of the letters. A fair test must compare intended defaults and also control apparent letter size and spacing. Marinus.
How the evidence shaped this font
The choices below are design hypotheses. None has a measured performance result for Talven Text.
| Design choice | Implementation | Reason and limit |
|---|---|---|
| Familiar body-text structure | Proportional sans serif, double-storey a, single-storey g, upright default. | Supports the intended web use and a quiet paragraph texture. The research does not establish that this combination is universally best. |
| Distinct similar characters | New barred I, curved-foot l, flagged and footed 1, and dotted 0. | Makes these outlines visibly different. This is not proof of fewer reading errors. |
| Open shapes | New c and e contours, with a raised e crossbar and angled terminals. | Keeps their mouths visible in the inspected weights. The best aperture size remains untested. |
| Larger lowercase | X-height changes from 496 to 520 units in a 1,000-unit em. Cap height stays 668. | A modest size choice for body text. It also affects the middle proportions of capitals. Larger visible letters can confound comparisons at the same CSS size. |
| Moderate built-in spacing | Add 10 units to positive advances, split across sidebearings. Space and nonbreaking space gain 24 units in total. | Avoids imposing extreme tracking. Word spaces grow more than ordinary advances. The exact values are optical design choices. |
| Adjustable weight | One variable axis, 300 through 700, with default 400. | Lets the reader and designer tune stroke weight. It does not diagnose or treat a reading difficulty. |
| Reader controls | Size, weight, line height, tracking, word spacing, line length, and light or dark paper. | Makes different settings available. The specimen does not rank readers or claim a preferred setting is the fastest. |
The spacing control adds equal extra word spacing when letter spacing rises. The separate word-spacing slider adds to that amount. This preserves a stronger word boundary as tracking changes. It is a conservative interface rule, not a validated optimum.
What is new, and what is inherited
Talven Text is a modified OFL edition of Atkinson Hyperlegible Next, not a wholly original drawing of every glyph. The pinned source repository supplies the broad Latin character set, most base shapes, punctuation, accent system, kerning, and the initial weight model. Upstream designers include Elliott Scott, Megan Eiswerth, Linus Boman, Theodore Petrosky, and Letters from Sweden.
This release adds seven custom base outlines, updates their composed derivatives, changes lowercase proportions and spacing, and rebuilds the variable family with new names. The source, exact modifications, and build script are included. See FONTLOG.md and font-manifest.json.
The SIL Open Font License 1.1 applies to the font and font sources. The source copyright remains intact. The original authors and Braille Institute have not endorsed this derivative. We used the pinned OFL repository, not a separately licensed download.
Website choices
GitHub's Mona Sans and Hubot Sans page and Monaspace page informed the specimen format: large type, direct controls, glyph inspection, and visible explanation of the font. The site uses its own layout, copy, colors, and font assets.
The reader starts at 20 CSS pixels, weight 400, line height 1.65, and a maximum measure of 60 ch. These are design defaults, not scientific thresholds. A ch is based on the width of the font's zero. It is not an exact character count for proportional text. Narrow screens reduce the line length.
Comparison fonts use the same CSS size. Their x-heights and shapes differ. The comparison is for exploration, not a controlled experiment. System sans depends on the operating system. Georgia can fall back to the available serif font if it is absent.
WCAG 2.2 does not prescribe one font family. Its AA text-spacing criterion requires content to survive user overrides at line height 1.5, paragraph spacing 2 em, letter spacing 0.12 em, and word spacing 0.16 em. Those values are override tests, not required defaults. The AAA visual-presentation criterion addresses further user controls and line length. W3C text spacing, W3C visual presentation.
What still needs testing
This release needs a reader study before any performance claim. A useful test would:
- Recruit readers with and without dyslexia and report each group's results.
- Compare Talven Text with its upstream base, a familiar sans serif, and a familiar serif.
- Test intended defaults and separate conditions matched for x-height, spacing, contrast, and measure.
- Randomize font and passage order. Use several equivalent passages without repeating text within a reader's comparison.
- Measure time, comprehension, errors where relevant, and comfort separately.
- Test any apparent personal winner on new passages and on another day.
- Record device, viewing distance, language, and prior font familiarity.
These are proposed methods, not a completed experiment or a power calculation. Sample size should follow a prespecified outcome and power analysis.
This first version covers Latin text. It has no designed italic, no claim of full global-script coverage, and no manual TrueType hinting. Technical font checks and browser tests support delivery, but they do not replace reader tests or production testing across operating systems.