Frequently Asked Questions
Converser™ for Healthcare 3.0 FAQ
- How will Converser™ be used by illiterate patients?
- What about patients who are unfamiliar or uncomfortable with computers and keyboards?
- Can't patients use Converser's speech interface to respond?
- Do you have to use Converser™ on a Tablet PC?
- How can Converser™ be used if the English and Spanish speakers don't take turns on a single computer?
- What's the best computing environment for working with Converser™?
- What's the best usage case for Converser™?
- Before using dictation for Converser™ input, do you have to carry out voice training?
- What is automatic dictation?
- Which input mode should I use: typing, handwriting, pointing, or speech?
- Is Converser™ too complicated for my patients?
- Do you claim that Converser™ is better than human translators? Wouldn't an interpreter or a bilingual staff member translate better than Converser™?
- Does the system produce "medicalese" which patients may fail to understand even in their own language?
- If the patient responds through gestures or yes/no answers, how can I record the response on Converser's transcript?
- How can I teach patients to use Converser™, or help them if they're having a hard time?
- Converser™ takes a minute or two to launch. Should we keep it loaded?
- Can I adjust the font size?
- Can I adjust the size of the onscreen keyboard in the Windows Input Panel?
- Can I make and save my own Translation Shortcuts for later re-use?
- The Meaning Cue definitions don't include the meaning that I want.
- The Back-Translation looks bad, but the Meaning Cues look correct (or vice versa). What should I do?
- My input included a proper name, but the name was translated as if it were a common noun, verb, etc. What should I do?
- How does Converser™ handle cultural issues?
- How accurate is Converser™ really? What is the percentage of successful translations?
How will Converser™ be used by illiterate patients?
For now, patients who can't read or write Spanish at all will need to use Converser™ passively: English-speaking healthcare workers will be able to speak to them in natural spoken Spanish, using the system's text-to-speech features; but responses from the patients will be conveyed through gestures, by saying "sí" and "no", etc.
However, patients who can at least read simple Spanish can respond through Converser's Translation Shortcuts facility. They can select a response phrase by pointing at it in a short list. In this response mode, no written input is necessary.
Patients who can write simple Spanish can additionally use Translation Shortcuts in another way – by entering even a few letters or words into the Converser™ Input Window. Converser's Shortcut Search facility will then instantly fetch a short list of complete possible responses beginning with the entered letters (or containing the entered keywords), from which a selection can easily be made by pointing.
What about patients who are unfamiliar or uncomfortable with computers and keyboards?
Few of the literate patients we've tested were reluctant to at least try out computer-based translation. However, there are indeed many for whom typing and keyboards are unfamiliar. These patients can make good use of Converser's Spanish handwriting interface, available on Tablet PCs. The software can recognize both cursive and block letters with high accuracy. Patients (and staff members, too) can write directly on the screen using a stylus. Corrections are made by simply scratching out errors and rewriting.
Can't patients use Converser's speech interface to respond?
Not just yet. To help obtain its unprecedented coverage of vocabulary and linguistic structures, Converser™ uses dictation software, capable of recognizing essentially any input words. (By comparison, the speech recognition encountered in telephone applications is much more limited in range.) To date, dictation software must be set up for specific individuals. Currently, this one-time process takes less than two minutes – no problem for staff members, but still impractical for many patients.
In the near future, even this individual setup will become unnecessary. At that point, speech input by Spanish-speaking patients and other temporary users will become practical.
Do you have to use Converser™ on a Tablet PC?
No, Converser™ can be used on any standard PC. However, only on Tablet PCs can users enter text using handwriting. On other PCs, text entry via typing, pointing, and speech remain available.
How can Converser™ be used if the English and Spanish speakers don't take turns on a single computer?
A Spanish-speaking patient can understand Converser's spoken and/or written output without attempting to respond via Converser. In this case, there's no need to pass the computer back and forth. If a transcript is desired in this usage case, the staff member will have to record the patient's responses by taking a turn on his or her behalf and entering e.g. "Sí," "No," or "Yo no sé" ("I don't know").
It's also possible to use Converser™ on two separate computers, e.g. on two sides of a counter, desk, or window. In this case, it becomes unnecessary to switch between the English and Spanish interfaces: the staff member keeps the English interface in view, while the patient keeps the Spanish one. However, if transcripts of the Conversations are requested, they will be saved as two files on two machines. These files can later be correlated by referring to their respective login users and timestamps, as indicated in the filenames and file headers.
What's the best computing environment for working with Converser™?
A basic choice is which platform to use. Should Converser™ be run on a Tablet PC or a on a standard PC? And, if a Tablet is chosen, should it be full sized (e.g. the Motion Computing LE1600) or smaller (e.g. the Motion Computing LS800 or Samsung Q1b ultra-mobile)? Finally, if a Tablet is chosen, should it be used with or without its detachable keyboard?
These choices obviously affect the mobility of the system. In general, stationary uses of the system are the easiest, since both hands are free; but mobile uses are often necessary, and Converser™ is designed to enable them as well. Large screens are advantageous for visibility; but in some applications patients won't need to see the screen, and can take advantage of voice output instead.
Choice of platform will also affect the availability of the handwriting and typing interfaces. Generally, the more alternatives available for input mode, the better; so arrangements offering both typing and handwriting are ideal. However, for some applications, especially mobile ones, handwriting alone will be sufficient.
Speech input for staff members will be available on any platform. The choice and arrangement of the microphone then becomes relevant. The Motion Computing LE1600 and Motion Computing LS800 Tablet PCs have built-in microphones which give good speech recognition results in relatively quiet environments. Desk microphones can also be convenient for fairly quiet stationary applications. For noisier environments, headsets are best. Both wired and unwired (Bluetooth) mics are available. (Please see NaturallySpeaking documentation for Nuance's recommendations concerning models.)
What's the best usage case for Converser™?
Converser™ is designed to be flexibly applied to many healthcare situations, so there's no one best usage case. However, several factors define a range of more and less optimal situations for Converser™ use.
- How critical? Converser™ includes powerful facilities for checking the accuracy of translations, so it's more reliable than most automatic translation systems. Nevertheless, for critical situations, such as life-threatening emergencies, human interpreters are obviously preferable if they're available. Converser™ is best used in non-critical situations.
- How noisy? If you're using speech input, the quieter the surroundings, the better. However, considerable noise can be tolerated if a headset is used (as opposed to a built-in mic on a Tablet PC, a table mic, etc.).
- How mobile? Converser™ can be used on any standard PC or Tablet PC running Windows XP. Thus it's as mobile as the platform that carries it. It also contains several features which facilitate hands-free and eyes-free use (such as voice commands and automatic pronunciation of Back-Translations). Even so, the situations which are intrinsically the easiest to handle are those which leave both hands free and allow easy viewing of the interface.
- How literate? Converser™ can be used by illiterate and semi-literate patients, but the quality of the interaction will obviously be relatively constrained. The freest Conversations will be those in which patients can read and write at least simple Spanish.
- Any barriers or obstructions? If patients are separated from healthcare staff by counters, window, or barriers, it may become impractical to pass the Converser™ machine back and forth. There are two main solutions: (1) Patients can use Converser™ passively only: they can hear its output, but can't use it to respond; or (2) Converser™ runs on two machines, one for the staff side, and one for the patient side. (A third, compromise solution would be to arrange a monitor for patients which mirrors the staff member's view. Patients could then observe the Converser™ interface, with its Spanish text and transcripts.) In any case, the ideal situation for access to Converser™ by both English and Spanish speakers is one without obstructions.
To sum up, the ideal usage case would be one which is (1) non-critical, (2) relatively quiet, (3) stationary rather than mobile, (4) serving at least minimally literate patients, and (5) without obstructions between patients and staff. However, the system can be useful even if the usage situation is non-optimal in several respects.
Before using dictation for Converser™ input, do you have to carry out voice training?
Not for the latest dictation software, but you do need to register, which takes less than two minutes. Dragon NaturallySpeaking 9.0 does ask each user to register individually, so that the program can learn from corrections and gradually adjust to each specific voice. However, registration no longer requires reading aloud for five or six minutes, as was necessary until fall, 2006. Instead, DNS now uses the roughly 30 seconds of speech which is recorded during microphone adjustment as its initial speech sample. Once DNS has digested this sample (which takes another 30 seconds), you're good to go. (Users of earlier DNS software versions will still need to read aloud.)
What is automatic dictation?
Automatic dictation is one kind of automatic speech recognition. You're dictating when you talk and a computer program types out your words as text. In Converser™, you have to enter text into the Input Window before you can translate it. Dictation is one way to perform this input function; the other ways are typing, handwriting, and pointing.
Dictation is different from the kind of speech recognition increasingly encountered on the telephone, from both the functional and technical viewpoints.
- Functionally, it's different because the dictation vocabulary is comparatively so large and broad – usually 60,000 general-purpose words and up – as compared with a few hundred or thousand words, limited to specific topics, for telephone systems.
- Technically, it's different because, to guess which words it's hearing, a dictation system uses statistics about common word sequences. Telephone-class systems, by contrast, generally use specific grammars, or sets of embedded word lists, for these predictions. Also, dictation systems have until now always been user-dependent: they need to know who is talking in order to learn the characteristics of each particular voice. Telephone-class system, however, are always speaker-independent: anyone can call in.
Converser™ uses dictation software rather than telephone-class speech recognition for speech input, in order to obtain its unique breadth of coverage, a breadth unmatched by any other speech-to-speech translation system: because dictation can type out almost any words you might say, you can use your voice to enter almost any text for translation.
Which input mode should I use: typing, handwriting, pointing, or speech?
All of the above.
If your application is stationary, e.g. at a reception desk, feel free to type. (You may not have to type the entire input, though: Converser™ can search for Translation Shortcuts based on only partial input.) If you prefer, or if your application is mobile, handwriting is a handy but somewhat slower alternative. Pointing with the mouse or stylus, e.g. in order to choose Translation Shortcuts from the Shortcuts Browser, is always available.
As for speech input, if your environment isn't too noisy, it can considerably speed your text entry. It takes a little practice, but it can definitely be worth it. (Tip: To avoid leaving the microphone on by mistake, be sure to enable push-to-talk functionality: the Converser™ User Manual gives detailed instructions.)
In Converser™ 3.0, patients can use three of these input modes: typing (with Spanish keyboard values – transparent stick-on labels are supplied), handwriting, or pointing. Spanish speech input is not supported for this version.
You and your patients should feel free to mix the available input modes: type, write by hand, point, or speak, as convenient, even within a given sentence or phrase. (In particular, staff members should feel free to correct speech recognition errors by typing or handwriting. Voice corrections commands are available and useful in many situations; but there's no law which says that they must be used exclusively.)
Is Converser™ too complicated for my patients?
In general, no. Like most sophisticated modern software, Converser™ offers powerful features which beginning or less capable users can and should ignore. (Compare for instance Microsoft Word, in which many advanced features are reserved for power users.) This is the case for Converser's powerful verification and correction facilities.
Converser™ is unique in enabling users, both English and Spanish speakers, to verify input and translation and, when necessary, to correct errors at both stages. This enablement is crucial for some applications in that it makes careful use of the system possible when appropriate. However, not every situation requires this level of care, and some patients are unable to apply it.
When verification is unnecessary or too difficult, patients can click Aceptar (OK) immediately after clicking Traducír (Translate), without bothering to verify or correct the current translation. It's true that translation quality will be reduced in such cases, but the output will often be sufficient for communication.
Another way to avoid verification and correction is to use Translation Shortcuts – pre-translated sentences which require no checking. Shortcuts can be used by pointing at them in the Translation Shortcut Browser, or by selecting them from the Shortcut Search Drop-down Menu, following text entry of just a few letters or keywords.
Finally, passive use of Converser™ by Spanish speakers is appropriate for many applications. In this case, patients will be able to understand what staff members say, but will respond through gestures or yes/no replies. (If transcripts are desired, staff members can enter short replies on users' behalf.)
Finally, remember that Converser™, while powerful and versatile, is not for everyone and every situation. If a patient is struggling or stressed and an alternative translation solution is available, by all means use it.
Do you claim that Converser™ is better than human translators? Wouldn't an interpreter or a bilingual staff member translate better than Converser?
Converser™ is not intended to replace human translators, with human hearts and minds. However, it will reduce translating costs, and will be invaluable when access to qualified interpreters is limited. By providing reliable translation instantly and 24/7, it fills a vital niche in the range of translation options.
We do not claim that Converser™ is in any sense superior to human translators. All the same, the system does offer some benefits – beyond affordability and 24/7 availability – which human interpreters do not offer. These include verifiability (with a human translator, how can you judge the translation accuracy, and what could you do if you found an error?); privacy (no third party need be present); consistency (the same translation can be produced for a given expression each time); and ease of record-keeping (since a transcript can be kept for each conversation).
Does the system produce "medicalese" which patients may fail to understand even in their own language?
This is up to the staff member. Converser™ contains a very large medical vocabulary, so it can translate sophisticated medical language when appropriate; but it also contains an even larger general vocabulary (of some 300,000 entries!), so that most everyday expressions can be translated as well.
In fact, because of this large general-purpose foundation, staff members and patients can feel free to range beyond strictly medical conversation to talk about family matters, hobbies, etc. No other speech-enabled translation system offers this breadth of linguistic and topical coverage.
If the patient responds through gestures or yes/no answers, how can I record the response on Converser's transcript?
You can answer for the patient through Converser.
- Just take a turn on the patient's behalf.
- Then click on "Sí", "No", or "Yo no sé" ("I don't know") in the Translation Shortcuts Browser. These short answers are available in every Shortcut Category and sub-category.
Alternatively, you can type the first character or two and then use Shortcut Search to fill in the rest.
- Select By initial characters (Por las Letras Iniciales) search mode via the Translation Shortcuts (Traducciones Favoritas) menu or the equivalent toolbar icons.
- Then begin to type "Sí", "No", or "Yo no sé" into the Input Window. One character will suffice. The completion will appear in the Shortcut Search Drop-down Menu (along with other expressions beginning with the character).
- Click on the completion (or select it and press Enter) to activate the Shortcut and put it on the record.
- Of course, you can type the entire expressions "Sí", "No", or "Yo no sé" if you like; but in this case remember the accents! "Si" and "se" are different words than "sí" and "sé," just as "to," "two," and "too" are different in English! (To type an accented letter, type the accent mark first – it's on the "[" key, as shown by your stick-on transparent key labels. Then type the letter.)
How can I teach patients to use Converser, or help them if they're having a hard time?
Several training aids for patients are built into Converser:
- Spanish-language animated tutorials (Quickstart, Text Entry, Translation Shortcuts, and Speech), available on the Ayuda (Help) menu.
- the How to Use Converser™ category of Translation Shortcuts, which contains a wide range of phrases explaining the use of system features, e.g. "Click on the sentence that you want Converser™ to translate".
- the Spanish-language version of the online Converser™ for Healthcare 3.0 User Manual, available on the Ayuda (Help) menu.
Patients will also learn by watching what you do. Don't hesitate to handle the interface for them, especially for the first few turns. Stick at first to the minimum actions:
- double clicking to activate Translation Shortcuts in the Translation Shortcut Browser;
- entering basic text in the Input Window by handwriting, typing, or using the onscreen keyboard;
- and clicking the Traducír (Translate) and Aceptar (OK) buttons.
Verification and correction of input and translation can be considered intermediate-level actions. Beginners can bypass them in many situations.
Finally, remember that Converser, while powerful and versatile, is not for everyone and every situation. If a patient is struggling or stressed and an alternative translation solution is available, by all means use it.
Converser™ takes a minute or two to launch. Should we keep it loaded?
Yes, that's a good policy for many applications.
However, the minute or so required to launch the system will generally be much less than the time needed to summon or call an interpreter.
Can I adjust the font size?
Yes. Use the View > Fonts menu, which gives a choice of Small, Medium, Large, and Extra Large.
Can I adjust the size of the onscreen keyboard in the Windows Input Panel?
Unfortunately, not in the present version of the Windows software (XP Tablet PC Edition SP2). Size adjustment was possible in earlier versions, but has been phased out. Please address your requests and suggestions on this topic to Microsoft.
Can I make and save my own Translation Shortcuts for later re-use?
Not yet. This powerful feature will be available in future versions of Converser™ for Healthcare.
The Meaning Cue definitions don't include the meaning that I want.
Please send email on this and other problems or requests to email@example.com. You'll receive a prompt reply and timely information on system fixes and updates.
Meanwhile, try substituting a different word and re-translating. Remember that it's not necessary to start from scratch: you can simply
- replace the relevant word in the Input Window on the Translation Screen
- click the Retranslate button.
The Back-Translation looks bad, but the Meaning Cues look correct (or vice versa). What should I do?
Converser's tools for verifying translations, Back-Translation and Meaning Cues, should be used together for maximum confidence. When both tools agree that a translation is acceptable and understandable, it usually is (if not always perfectly idiomatic or smooth).
If one or the other looks bad and the current translation is an important one, it's generally best to (1) fix word meanings and/or (2) rephrase or simplify the input until both clues show acceptable results.
(It's true that false negatives sometimes occur – translations that look bad via Back-Translation or Meaning Cues but are actually good. These cases are bugs, and should be reported to firstname.lastname@example.org. You'll receive a prompt reply and timely information on system fixes and updates.)
Note that Meaning Cues can sometimes catch ambiguities that Back-Translation does not reveal. For example, the word "file" might be translated incorrectly as "polisher" when you want the "data file" meaning. This inappropriate meaning would show up clearly in the Meaning Cues; but the Back-Translation might still contain the ambiguous word "file". This Back-Translation wouldn't be wrong, but because it would still be ambiguous it would not reveal the problem when used alone.
My input included a proper name, but the name was translated as if it were a common noun, verb, etc. What should I do?
Make sure that the name is capitalized. "Mark," "Sue," and "Mike" will be treated as a proper names, but "mark" (with lower case) will mean "spot, blotch, stain, etc.," "sue" will mean "take legal action," and "mike" will mean "microphone".
If necessary, you can also use the DO NOT TRANSLATE option, which is always available in the Change Meaning Window.
How does Converser™ handle cultural issues?
Converser™ is a translation tool for humans – but is not a human itself! It should be used by people with sensitivity, understanding, and warmth, but cannot itself supply these human qualities. However, English-speaking staff members and Spanish-speaking patients will find it much easier to express these qualities when the linguistic barrier to expression of ideas is lowered.
In general, we stress that Converser™ can't fully replace human translators, with human hearts, human minds, and human cultural awareness. The system is instead intended to provide translation instantly and 24/7when access to qualified interpreters is limited, thus filling a vital niche in the range of translation options.
When cultural issues are raised, often the reference is actually to linguistic differences among Spanish-speaking groups. Interpreters try to stay aware, for instance, that a bus is called a "guagua" in Puerto Rico, Cuba, the Dominican Republic, and the Canary Islands, but is called a "camión" in Mexico and Central America, "autobús" or "coche línea" in Spain, and so on – just as a "pancake" may be called a "flapjack" or "griddle cake" in different parts of the US. (Even so, a neutral or standard term like "autobús" or "pancake" will generally be understandable.)
Converser™ has the ability to capture these dialect items in its translation lexicons (dictionaries), which are continually updated. When ambiguities result from the vocabulary items (as for instance with "camión," which might mean either "bus" or "truck," depending on the dialect), they can be indicated and resolved via Meaning Cues.
Of course, the effort to capture these dialect elements will remain a work in progress. Please communicate about them via email@example.com. You'll receive a prompt reply and timely information on system fixes and updates.
How accurate is Converser™ really? What is the percentage of successful translations?
Measurements are complicated by Converser's multiple options for input (speech, typing, pointing, and handwriting), and by the wide range of healthcare situations – from intake interviews to rehabilitation.
Further, methodology for evaluation of speech translation, and of machine translation more generally, is a matter of ongoing debate in the research community: there can be degrees of "correctness," and criteria can vary from "all vocabulary items and structures are perfect" to "the utterance is understandable".
Given this variability, it's impossible to provide a scientific measurement of Converser's overall success rate.
However, it's crucial that the system can be monitored and corrected at two levels – input and translation. Thus users can raise the accuracy as much as necessary for the sentence at hand by interactively repairing errors. The time spent on corrections can depend flexibly on the importance of the current input.
At the input level, all text is subject to automatic spell checking: words not recognized by the spellchecker turn red, and right-clicking on these will bring up a list of correction candidates.
The input methods themselves are highly accurate: Nuance claims that speech recognition with Dragon NaturallySpeaking is above 90% accurate, and this is in line with our own experience; Microsoft's handwriting recognition is excellent, though no figures on error rate are available; and pointing and typing are fully under the user's control.
At the translation level, while no percentage figures can be offered, Converser™ users have a unique opportunity to monitor and correct translations.
- When both Back-Translation and Meaning Cues give positive indications, we believe that the resulting translations are understandable in the great majority of cases.
- When either tool gives a negative indication and the current translation is an important one, it's generally best to (1) fix word meanings and/or (2) rephrase or simplify the input until both clues show acceptable results.
- It's true that false negatives sometimes occur – translations that look bad via Back-Translation or Meaning Cues but are actually good. However, it's better to be safe than sorry, and better to err on the side of caution. (False negatives are bugs, and should be reported to firstname.lastname@example.org. You'll receive a prompt reply and timely information on system fixes and updates.)