Three methods of text entry have been used for real time subtitling. The first two employ fast-keying techniques. Such systems are designed to produce verbatim transcripts, this can be inappropriate for television subtitling if the word rate is very fast. The operators must therefore be retrained to edit the soundtrack, or to work in conjunction with an editing interpreter. A more serious disadvantage is that deaf viewers can be confused by inaccuracies in the spelling of the subtitle output. This problem is likely to reduce as the development of transcription systems continues. It is desirable to integrate the machine-shorthand writer into the pre-production process, for example by making scripts available for preview. Lastly, of course, the technique relies on the availability of trained operators.4.2 Advance Preparation
i) Phonetic Keyboards
The first involves a special phonetic keyboard designed for verbatim transcription, such as the Palantype system or the Stenograph system. A trained operator uses the keyboard to enter a series of phonetic codes representing speech, and a computer decodes this information to produce, so far as possible, a conventional transcription. Due to ambiguities in the phonetic coding, operator errors and spelling complexities, the spelling and word-boundary identification in the transcribed output is not always accurate. Depending on the size of the dictionary in the transcription computer, and on the error correction techniques available, an average output accuracy of between 75 per cent and 95 per cent is generally achieved, at speeds of up to about 200 words per minute.
ii) Velotype Keyboards
The second transcription method uses the Velotype syllabic chord keyboard, which can attain a speed of around 100-140 wpm with a trained operator. It is lower cost than the phonetic machine shorthand systems and does not require extensive dictionaries, although shortforms are essential where unusual spellings are likely to occur.
iii) Qwerty Keyboards
The third method of text entry for real time subtitling uses an ordinary Qwerty keyboard as an alternative to machine shorthand. The problems of non-standard spellings are largely overcome (except for occasional typing errors), but the maximum rate of text input is much reduced. A maximum subtitling rate of about 80 wpm is typical. The use of shortforms provides a valuable means of speeding keyboard input and reducing the likelihood of spelling errors.
Having decided that a particular live programme is to be subtitled, it is necessary to choose an appropriate strategy.4.3 Choosing Shortforms
The choice is based on an assessment of the likely format of the programme, the availability and reliability of scripts and the expected presentation speed. In most cases, a hybrid approach is probably necessary. This involves switching between manual cueing of prepared subtitles during scripted portions and live inputting.
Flexibility is an important feature of live subtitling equipment, since switching between different transmission modes must be achieved rapidly and straightforwardly.
It is helpful to draw up a running-order or programme plan based on information available in advance about the likely content of the broadcast. Liaison between the subtitling service and the programme production team is of value during the planning phase. Scripts available in advance can be edited and typed into the subtitling system for storage in memory and/or on disk. Two points are worth noting here:
i) Pre-stored subtitles should be accessible in groups chosen to distinguish subject boundaries. This makes random access easier should the running order change during the broadcast.
ii) Pre-stored subtitles should be limited to two lines, since three-line texts may obscure foreground detail and there is little time to reposition subtitles during subsequent live cueing.
In addition to pre-stored subtitles produced from scripts, background research should enable a number of general 'fallback subtitles' to be prepared relating to the expected programme material. This approach involves making available a subtitled commentary which can be used independently of the soundtrack. Editorial discretion is required when integrating these standby subtitles with conventional commentary-based material.
An additional and important aspect of advance preparation concerns a method for speeding keyboard input by using abbreviations, or shortforms, in place of words or phrases expected to occur in the broadcast. The technique was first developed to reduce the burden on the operator of a conventional qwerty keyboard when working under pressure.4.4 Subtitle Composition
Prior to transmission, the subtitler assigns two or three character shortform abbreviations or mnemonics to selected words or phrases. Experience indicates that proper nouns such as the names of people, places, buildings, bridges, boats etc can usefully be abbreviated in this way. When subtitling sport, terms relating to the particular game can also be stored.
Then, when a shortform is typed in an ordinary sentence, the subtitling computer automatically detects and expands it to its full form, using a predefined shortform dictionary. Such dictionaries can be stored on disk and recalled for later use.
Four guidelines have been found to be valuable when using shortforms:
i) The shortforms should be chosen by the person who is eventually going to use them at the keyboard.
ii) Where possible, some consistent abbreviation technique should be developed, eg taking the first three letters of single words, and the initial letters of multiple-word sequences. This assists in recalling the shortform, and may enable it to be deduced if forgotten.
iii) It is important to ensure that no shortform can also be a valid word, otherwise there can be an erroneous expansion when shortforms and ordinary words are mixed. iv) An easily visible list of shortforms and their expansions should be available to the subtitler during the broadcast - for example posted on the wall. This makes it possible to look up quickly and check an item should it be necessary. The advantage of the shortform technique is in its flexibility, since it is entirely up to the operator how the abbreviations are chosen and used. Considerable typing time is saved, and the possibilities of spelling error are reduced. For example:
thol = The House of Lords wbr = Westminster Bridge tbr = Tower Bridge ve = Victoria Embankment gm = Greenwich Meridian sou = Southwark
The construction of subtitles for informative subjects such as news should convey the whole meaning of the material. This need not mean using the same amount of words. Research into this area has described the concept of 'idea units'; that is where a proposition or key information is given. These units should be distinct with minimal repeats, and relate to the original information.4.5 Subtitle Presentation
If the programme speech is too rapid either for the viewer or for the chosen means of text entry, the speech must be edited 'on the move' before entering it as a subtitle.
Such editing must be performed very rapidly to avoid long delays between speech onset and the appearance of a subtitle. This is a skilled task, and its degree of success depends on the type of programme and the editor.
Narrative-style programme commentaries given during major live outside broadcast events are relatively easy to edit in real time. The commentator is not visible, thus reducing synchronisation problems, and the salient points of what may be a leisurely commentary pace can readily be picked out and subtitled. In contrast, the subtitling of a live news broadcast presents severe difficulties. Information is presented in compact form, and the rate of delivery is usually rapid. In addition, it may be almost impossible to edit politically sensitive material without distorting it. Between these extremes are situations in which a trained editing interpreter can work with varying degrees of success.
Each real-time subtitling system suffers from the problem that both composition and text entry impose a delay between the start of an utterance and the appearance of the corresponding subtitle. The delay varies from one to three seconds for verbatim phonetic machine shorthand, to around five seconds for edited input using qwerty. Delays can be reduced by 'scrolling' presentation methods, but in teletext this can be difficult to follow. This can cause difficulties to the viewer, especially if the programme soundtrack is in obvious synchrony with the visual material as, for example, during an on-screen interview.4.6 Guidelines for Real-time Subtitling
i) Word-by-Word Display
Two methods of word-by-word displays are currently available: a screen which overwrites when it reaches the bottom line or a screen which scrolls, ie jumps up, pushing the top line of text out of the text window.
Although of value for live subtitling, the use of a word-by-word display can create problems for the reader because of the speed of speech output and possible confusion in eye-movement. Its advantage, however, is in the provision of near-verbatim text.
ii) Standard Format or Block Text
The subtitles are presented in complete phrases or sentences similar to those prepared subtitles associated with recorded programmes. Whilst the 'on-screen' appearance of this form is often slower because of the longer wait for complete syntactical sentences, the ability of the reader to flit from pictures to words assists certain deaf viewers in understanding the programme. Current research indicates that both methods of live subtitling are accepted by approximately equal proportions of deaf viewers.
Early research indicates that first attempts were considered to be too fast. Although the rate of subtitling is driven at the rate of the presenter/journalist, there is still a need to focus carefully on reading variables. In such situations, although preparation time is limited, efforts must be made to adhere to at least the following:
The following are offered as more detailed guidelines during the preparation of subtitles in real time:
- Subtitles should contain a reasonable percentage of the words spoken.
- 'Idea units' or key facts should appear as a good percentage of the spoken message (see Section 4.4).
- Avoid 'idea units' which are unnecessary or different from the original.
- Where possible, avoid non-linguistic line breaks (splitting verbs etc).
- Attempt to avoid overrunning shot changes (synchronisation).
- Where possible avoid dynamic displays; that is to say blocks are considered more acceptable than the scrolling/word-by-word format.
When cueing prepared texts for scripted parts of the programme:
- Maintain a regular subtitle output with no long gaps (unless it is obvious from the picture that there is no commentary) even if this means subtitling the picture or providing background information rather than subtitling the commentary.
- Aim for continuity in subtitles by following through a train of thought where possible, rather than sampling the commentary at intervals.
- Produce complete sentences even for short comments because this makes the result look less staccato and hurried.
- Bear in mind that a subtitle specific to a particular scene can often be phrased sufficiently broadly to 'survive' a sudden camera cut without having to be abandoned. If pre-stored specific subtitles are used, ensure that they are cued at appropriate times.
- Send an apology caption following any serious mistake or a garbled subtitle; and, if possible, repeat the subtitle with the error corrected.
- Do not subtitle over existing video captions where avoidable (in news, this is often unavoidable, in which case a speaker's name can be included in the subtitle if available).
- Do not start subtitling 'cold'. A short rehearsal should be conducted just prior to transmission.
- Try to cue the texts so that they closely match the spoken words in terms of start time.
- Try to include speakers' names if available where in-vision captions have been obliterated.
- Do not cue texts out rapidly to catch up if you get left behind - skip some and continue from the correct place.