Bibliography and references

Table of contents

  1. Basics of caption typography
  2. Existing caption and subtitle fonts
  3. Basics of Line 21 character rendering
  4. Basics of teletext rendering
  5. Character shapes
  6. Font preferences
  7. Caption reading
    1. Blurred vision

Basics of caption typography

Typography and appearance” (from Best Practices in Online Captioning)
“Typography and TV Captioning,” Print, January/February 1989
Sequel: “Reading the Tube,” Print, March/April 2002

Existing caption and subtitle fonts

Southall, Richard, “Character generator systems for broadcast television.” Information Design Journal, 2(1):44–57 (1981)
Silver, Janet, John Gill, Christopher Sharville, James Slater and Michael Martin, “A new font for digital Television subtitles” (Tiresias)
Gill, J., J. Silver, C. Sharville, J. Slater, and M. Martin, “The design of a typeface for digital television
And here is the extent of the research adduced in this paper: “Within the time available, thorough testing to determine the optimal design of the typeface was not possible. However[,] limited testing was done with patients in a low-vision clinic to identify the features of a typeface which they found helpful.” And there you go: The entire basis upon which Tiresias is claimed to be a superior font for “subtitling”: Asking a few people with low vision what they liked.
Gill, J.M., and S, Perera, “The Tiresias family of fonts

Vernon... found confusable letters include f and t, l and t, c and e, n and a, i and j, [and] I and J. Burt... maintains that [the ability to distinguish] is necessary [for] Il1!i and [for] h and b, C and G, Q and O, J and F, [and] R and Q. For numerals, Tinker... found 3582 to have low legibility.

Basics of Line 21 character rendering

Keith Jack, Video Demystified (LLH Technology Publishing, 2001), p. 337

Characters are displayed using a dot-matrix format. Each character cell is typically 16 samples wide by 26 samples high (16 × 26).... Dot rows 2–19 are usually used for actual character outlines. Dot rows 0, 1, 20, 21, 24, and 25 are usually blanked to provide vertical spacing between characters, and underlining is typically done on dot rows 22 and 23. Dot columns 0, 1, 14, and 15 are blanked to provide horizontal spacing between characters, except on dot rows 22 and 23 when the underline is displayed. This results in 12 × 18 characters stored in character ROM....

Some caption decoders support multiple character sizes within the 16 × 26 region, including 13 × 16, 13 × 24, 12 × 20, and 12 × 26. Not all combinations generate a sensible result....

Basics of teletext rendering

Dennis N. Pim, Television and Teletext (Macmillan, 1988)

[I]t was decided to use 40 characters per row.... [T]he numer of rows was determined, and a value of 24 was specified – although... an extra 25th row is used in the new enhancements to the specification.

This source also lists the basic teletext character set as follows (excluding letters, numerals, blob, and space):


Numerous supplementary character sets are available.
Other sources list the dot matrix used for small characters (as typically found in teletext captions) as 5 × 9 pixels. However, there’s conflicting information, some of it coming, curiously enough, from our friends at Tiresias:

Effectively, putting different information on the two adjacent interlaced lines doubles the character resolution from 7 × 5 to 14 × 10, making for much smoother-looking characters.... Although the basic alphanumeric information for teletext is transmitted as a 7 × 5 dot matrix, most modern receivers use a character-generator ROM which can provide a 9 × 10 matrix display, in order to allow for character ascenders and descenders to be more attractively displayed.

Character shapes

Old Line 21 decoder fonts did not use descenders on the lower-case g·y·p·q·j. Some new decoders, to this day, do not provide descenders.

Fonts with no descenders are hard to read off a screen.

Treurniet, William C., “Spacing of characters on a television display.” In Processing of Visible Language 2, Kolers, Wrolstad, Bouma, eds. (Plenum Press, 1980)
The test involved looking at a screen filled with random letters. One line was indicated. The subject had to spot the first letter on that indicated line, find the next occurrence of it on the same line, and state the letter to the right of that.

Analysis of frequency of misses showed... an effect due to length of descender.... Analysis of scanning rate in characters [also pixels] per sec showed a significant effect due to...length of descender.... Descender lengths of zero resulted in significantly more misses than descender lengths of one or two pixels. There was, however, no significant difference between the latter two descender lengths. Similarly, the lack of a descender was related to a significantly slower scanning rate than were descender lengths of one or two pixels. Again, there was no significant difference between the latter two descender lengths. These findings suggest that the descender should extend below the line by at least one pixel if it is to be readily noticed.

The reason for the effect of descender length is probably straightforward: A descender that extends below the line by even one pixel is a cue distinctive enough to aid detecting that character accurately.

Font preferences

Almost no research on this topic exists. A severely limited study on font preferences for captioning for HDTV produced results that are open to serious dispute, given that the candidate fonts chosen were Times, Helvetica, and Monaco and the high-definition signal was simulated.

(Navoy, Judith,) Advanced Television Closed Captioning Features (WGBH, 1997)

We tested three different fonts: Helvetica, Times, and Monaco. We included Helvetica because it is a popular sansserif font, and Times because it is a popular serif font. We included Monaco because we wanted something similar to today’s captions for which no “font” exists. Monaco is similar to current captioning because it is a sans-serif, monospaced font. Although most of the captions we tested were in mixed case, we elected to use Monaco in all upper case so that it would look more like current captions.

The Helvetica font was the clear choice of participants. Participants praised its clarity and simplicity. Most participants felt that Times was too “busy” or “crowded.” This was due in large part to the serifs, although some felt that another serif font may have worked. Most of the participants felt that Monaco was too large. This may have been because the letters were all uppercase. The poor response to Monaco indicates that it was not an effective approximation of today’s captioning after all.

Print fonts used as simulations for onscreen caption reading will lead to misleading results. You don’t put bicycle tires on a car to test how well it corners.

Jordan, Amy B., et al., The State of Closed Captioning Services in the United States. Philadelphia: Annenberg Public Policy Center, 2003, pp. 28, 46–47 (scanned PDF)

[D]eaf respondents felt that captions are too small and too fast... Hard-of-hearing respondents... cited problems with spelling errors, size and speed of the captions, contrast, and placement of the captions on the screen. Some respondents suggested having a mechanism such as video selection [sic] that would allow individuals to select their own preferences for captioning style. [...] Many respondents expresed personal preferences for the captions – larger type, all caps, or slower pace, for example. [I]t is critical to consider creating a technology that can allow the expression of personal preferences with individual programming.

Caption reading

Rather than reinvent the wheel, we’ll just point you to a previous research summary on caption reading.

Blurred vision

Thorn, Sondra, and Frank Thorn, 1989: “Television and vision: reading captions when vision is blurred.” American Annals of the Deaf, 1989 March; 134(1):35–38

[S]tudies have shown that hearing-impaired peole are much more likely to need glasses than normally-hearing people, and if they wear glasses, they are often the wrong strength....

Six hearing-impaired and 14 hearing subjects participated in this study... [and] viewed a 19-inch colour television set from a distance of three metres.... The subjects read the captions aloud as the captions appeared on teh screen.... Subjects received eye exams before participating in the study. If necessary, their vision was corrected.... Subjects read the captions under five conditions of blur [–0.25 to +1.75 diopters]

Just +0.75 diopters reduced the ability of the hearing subjects and +1.75 dopters made caption-reading impossible for them. Hearing-impaired subjects could accept slightly more blur than the hearing subjects.... Although subjects read most captions correctly with +0.75 diopters of blur, they seemed to do this more slowly and more hesitantly than when vision was clear....

The ability of some people to read captions through +1.75 diopters of blur was at first surprising.... We believe that some of the subjects have learned to quickly interpret the shapes of words even when most letters are blurred by using the same contrast-detection mechanisms as those used in the detection of gratings.

Thorn, Frank, and Sondra Thorn, 1996: “Television captions for hearing-impaired people: a study of key factors that affect reading performance.” Human Factors, 1996 September; 38(3):452–63

[C]aption-reading is more difficult than many reading tasks. Because the text is divided into segments of one to seven words [sic] and displayed for a limited time, this leaves only a small margin for error.... [V]isual problems are more common in the hearing-impaired community than in the population at large. Studies have shown that children and young adults with congenital hearing loss are more likely to have visual problems than are normally-hearing people [mostly due to] refractive errors. To make matters worse, if deaf people have a refractive error, they are also more likely than [hearing] people to have the wrong eyeglass prescription....

Why does blur reduce caption-reading ability when vision is blurred only 1.0 or 1.5 [diopters]? ...[T]he most likely answer is simply that a person reading continuously-presented text cannot ponder or recheck his or her first impressions. Rather, a reader has one good look at each letter in its context and then must move on....

[W]e believe that the simultaneous presentation of a second captioning style that has a slower rate of presentation and larger text may respond to some of the special needs of the hearing-impaired population.... [C]aptioning agencies canot adopt such a policy until the parameters of the text needed for the congenitally deaf and elderly hearing-impaired populations are empirically determined through rigorous experimentation with these people.

Thorn, S., F. Thorn, and D. Malloy, “The elderly read TV captions as well as young adults.” Presented at the annual meeting of the Association for Research in Vision and Ophthalmology, 1995

As with young adults, small increases in blur and text speed decreased performance.... Surprisingly, elderly subjects’ performance improved in comparison to young adults under the most adverse conditions....

Despite the commitment of government, producers, and broadcasters to provide this expensive service [sic], there is almost no research on the legibility of captioned text or its appropriateness for the target population.

Melody F. Harrison, Barbara B. Braverman: “Legibility factors related to the captioning of video displays: A review of the literature.” In Captioning: Shared Perspectives, Barry J. Cronin, ed. (National Technical Institute for the Deaf, Rochester Institute of Technology, 1980)
It’s pretty old now, but this paper notes (where H represents the height of a TV screen):

Researchers... found that lower-case ascenders or descenders 1/4 to 3/8 inch allow 16H–24H viewing and provides legibility at a distance of 18–24 feet, 21–27 feet, and 23–30 feet, respectively, from a 17″, 21″, or 25″ picture tube. It has been found tha tthere is a significant decrease in the accurcy with which symbols can be identified as the line size of the symbols decreases.... Shurtleff recommended that the minimum acceptable vertical symbol resolution be 10 lines.... Millerson... recommended that minimum lettering size should be 1/10 to 1/25 picture height.

Version history

2008.02.22 14:48
Updated again, with more from Sondra Thorn.
Added Annenberg study.