Joe Clark: Accessibility | Design | Writing

Response to report on captioning on French CBC channels

In 2004, retired Canadian Senator Jean-Robert Gauthier, a hard-of-hearing person, filed a complaint against Société Radio-Canada, the French arm of the CBC, concerning captioning. As part of the settlement process, SRC agreed to submit a report on the state of captioning, particularly real-time captioning, on Radio-Canada (the general-interest public broadcaster) and Réseau de l’information (the all-news network).

The report, which you have to E-mail the Canadian Human Rights Commission to get, carries the date of 2005.04.28, though I received it three months later.

I read the French original, entitled “Comité de travail concernant le sous-titrage codé en français à la télévision de Radio-Canada,” and have the following remarks. All quotations are my own translation.

Online version

As of autumn 2005, there is an online version of the report in English.

Caption quantity

There is, as ever, an obsession with quantity of captioning rather than quantity and quality. Senator Gauthier’s complaint, which I haven’t seen, presumably alleges that captioning anything short of 100% of programming constitutes discrimination, or unequal treatment, on the basis of disability. I believe he was particularly concerned about live programming.

Right there in the executive summary, Radio-Canada plans to reach 100% captioning on Radio Canada “all day” and 90% on RDI, again “all day,” with 100% during the most-watched times of day, 0600–1000 and 1600–0000 hours.

There is, however, an ambiguity. CRTC rules state that the broadcast day runs from 0600 hours to 0100 hours each day. The broadcast day is up to 18 hours long by policy. Officially, overnight programming is unregulated; it barely exists.

A footnote at the end of the report states that “the network timetable broadcasts about 18 hours of programming per day, 126 hours per week, 6,552 hours per year. RDI broadcasts 24 hours a day, that is, 168 hours per week, 8,736 hours per year.”

As it stands now:

This human-rights case may result in a requirement for SRC to caption every minute that Radio-Canada and RDI are on the air, or it might require 100% captioning of the broadcast day. The report leans toward the former, but does not even discuss the mismatch between calendar days and broadcast days. I recommend that the Commission specifically and unambiguously address this discrepancy in its decision.


Reliance on voice recognition

As ever and yet again, we’ve got a report on captioning that spends too much time on the topic of voice recognition. That failed vapourware technology, which will not be producing accurate captions at any point in our lifetimes, is held up as a possible solution to the problem of captioning at RDI; the report suggests that voice recognition will push RDI from 90% captioning to 100%. (They mention “technological advances, particularly in the field of voice recognition.” I’m not sure what other “technological advances” are on the horizon. They really mean voice recognition.)

The report also mentions that “to our knowledge, the technology of voice recognition has yet to attain the quality levels of stenography for real-time captioning on television.” Indeed it hasn’t, and it’s not going to, either, anytime soon. SRC’s in-house captioning department uses speaker-dependent voice recognition as a transcription method for offline captioning, which is a different matter entirely. (Actually, they also say they use “a regular keyboard.”)

SRC would be well advised to stick with stenography for that extra 10% captioning on RDI, and I encourage the Commission to disregard altogether any appeals to a technology that doesn’t exist as a solution to a present-day captioning problem.


SRC proposes to work with La Cité collégiale, Ottawa, to develop a twelve-month training course for French-language stenography. (Incredibly, to this day there is no such course; though la Cité collégiale had been trying to get one off the ground since 2003, there weren’t enough students, the report states.) SRC will then provide internships for twelve months “at its cost” (does that mean the interns will be paid?) for graduates of the program.

The details, however, are a problem:

  1. If fewer than 15 students sign up, the course will be put off for a full year. The report does not explain how SRC will increase its captioning quantity if that happens.
  2. Radio-Canada will run a public-service announcement for the course, but only in Ontario and points east. (The report lists a budget of $12,000 to produce the PSA and $150,000 to $200,000 to broadcast it. That’s a lot of money, and only the actual production costs represent real expenditures. The PSA allegedly aired from April to June 2005.)
  3. The report lists a few other ways in which SRC will promote the training course, including passé methods like press conferences, but never mentions a Web site.
  4. SRC will indeed accept interns who graduated from the program, but only two of them. How will that be sufficient to caption live portions of a 24-hour news network and of a general-interest network? (The report states that 75% of RDI programming and 40% of Radio-Canada programming is broadcast live.) What will the other 13 students do? Why is there a 15-student minimum if only two are needed?


The training program will be of dubious benefit anyway since the report states that interns will have to transcribe a mere 110 words per minute in order to be hired by Radio-Canada. This is an outright joke. English-language real-time captioners aren’t considered competent below 180 wam, with 220 wam preferred – and that’s for a language with fewer words of shorter average length than French! (I’m actually giving you a conservative estimate. Gary D. Robson, in The Closed Captioning Handbook, p. 119: “Real-time stenocaptioners must regularly work at sustained speeds of over 225 words per minute with accuracy of 99% or better.)

If a program has dialogue at an expected 180 to 220 words per minute, SRC promises to caption 50% to 61% of it. Will that be enough, Senator Gauthier?

Myths about the French language

Thankfully, the report spares us the oft-reiterated lie that French-language captioning is “more complex” than English-language, hence captioning targets should be lower. Offline captioning isn’t any harder than it is in English. Real-time captioning isn’t conceptually harder; we merely lack trained court reporters.

The report does head off in a new direction, though, when it states that “stenotypy was developed first in English-speaking countries, since the syntax of the language is simpler.” That’s a severely ethnocentric view that no professional linguist would back up. Certainly the morphology of French is more complex than English, with its gender and number agreement, but that’s simply part and parcel of captioning the French language.

Further, “the complexity of the French language does not facilitate phonetic transcription and has hindered development of the technology for the French language,” apparently. The statement is false. French real-time captioners’ keystrokes for suffixes (like masculine, feminine, and plural endings) are no different from English-language captioners’ keystrokes for suffixes. If the report’s authors were more familiar with the ways in which words are built up using machine shorthand, they would not have made a statement of this sort.

Historical errors

The report wastes a full page listing some of Radio-Canada’s extremely impressive captioning milestones, some of them incorrect or misleading.

January 1982
“Radio-Canada airs the first dramatic series in French with closed captioning, Terre humaine.” Dramatic series, perhaps, but Clown White aired in English and French on both CBC networks in 1982. (I watched it.)
“Radio-Canada captions Le Téléjournal at 10:00 PM. It’s the first time in the world that a television network offers hard-of-hearing people a daily newscast with closed captions.” The National was captioned in that era, too. World News Tonight began using real-time captioning in 1982, and I watched The Captioned ABC News every weeknight growing up in the 1970s. If anything, Société Radio-Canada was embarrassingly late to the game.

Elsewhere, we are told that “to our knowledge, Radio-Canada is the only French-language broadcaster in the world to caption all of its news programs, including live segments.” That may be true now, but in 1996, TFO captioned Panorama live, with cleaned-up captions on rebroadcast.

Replacement for Médiatex

The report states that SRC uses the Médiatex system for real-time captioning. It was codeveloped by IBM and uses the Grandjean keyboard.

The SRC is looking for a replacement for Médiatex. According to the report’s schedule, «un appel d’information (Request for Information)» will be published from April to June 2005. But Marc Cavanagh of Radio-Canada informs me via E-mail that the process was still outstanding at the end of July 2005, though he refused to state when the contract would actually be awarded. (He also refused to provide a copy of the “supply request,” and went so far as to forward my inquiry to SRC’s lawyers. Is there something to be afraid of?)

Quality and standards

As I suggested before, Senator Gauthier, the Commission, and the SRC are dealing with only half the problem. Yes, everything needs to be captioned (and nearly everything needs to be audio-described), but just any old captions won’t do. Quality is an issue, and anyone who thinks there are actual quality standards at work at SRC hasn’t watched their captions.

At the level of a human-rights complaint, any settlement with SRC must at the very least:

  1. Require mixed case at all times.
  2. Require the use of the full EIA-608 accented character set. (I’m sure you’ve noticed that SRC’s real-time captioning uses no accented character other than é, for no technical reason whatsoever.)
    1. If backward compatibility is proven to be an issue rather than merely asserted, the optional extended character set would not have to be used.
    2. Ban incorrect character glyphs like << and >> errantly used for « and », which in any event exist solely in the optional character set.
  3. Don’t reuse real-time captioning on programming and segments that were already real-time-captioned. If a one-hour program airs live at noon and is repeated at midnight, I don’t want to see the original airing’s incorrect and slow real-time captions. Do not just run the noontime tape; clean up, add to, and live-display the captions on the midnight airing. (The same applies to individual unchanged news items repeated throughout the day.) Reserve real-time captioning for live and near-live programs; a repeat broadcast is neither of those.
  4. Do not exempt subtitled programming. If Radio-Canada runs House of Flying Daggers with original Chinese dialogue and French subtitles, captions must also be present (for non-speech information, unsubtitled utterances, speaker identification, and the like). Note that captions and subtitles are different even if this is hard to get across in a language that uses the same word for both.
  5. No use of scrollup captioning on fictional programming at all.
  6. A blink rate of two frames on all pop-on captions (not zero and not four). Blink rate is the number of blank frames between captions. SRC’s low-competence suppliers tend to use blink rates of zero, making captions too fast, or four, making them too slow. (The CBC’s captioning software, Swift, defaults to zero, an error about which Softel has sent several alerts to customers.)
  7. Caption any English-language dialogue within predominantly-French-language programming.

Posted: 2005.07.29 ¶ Updated: 2006.08.07, 2007.03.01

Homepage: Joe Clark Homepage: Joe Clark Media access (captioning, Web accessibility, etc.) Graphic and industrial design Journalism, articles, book