Covered. Nineteen. At pain medicine

« previous post | next post »

Google Fi screens my calls, so that my phone doesn't even ring unless the caller is in my contacts, or passes some kind of quasi-Turing Test. This is a Good Thing, since I get half a dozen spam calls a day, often at inconvenient times. As a result, robocalls generally end up as voicemail, which Google Fi helpfully turns into a convenient text message — which is often amusing. For example, a couple of days before my second vaccine shot last month, a robocall from Penn Medicine got transcribed like this:

Hello, this is pain medicine reaching out to you regarding covered. Nineteen. We've implemented a short sentence screening survey before coming into your appointment. All pain medicine patients are being asked to complete this brief electronic symptom checker to answer the questions, please call 215-NNN-NNNN. If your appointment has been canceled or rescheduled, please disregard this message patients and visitors. I'm presenting two pain medicine locations for inpatient outpatient or emergency department care should be wearing a cloth face covering in accordance with current CDC and state guidance. Thank you.

[Callback number obscured]

What that should have been:

Hello, this is Penn Medicine reaching out to you regarding Covid 19. We've implemented a short symptom screening survey. Before coming in to your appointment, all Penn Medicine patients are being asked to complete this brief electronic symptom checker. To answer the questions, please call 215-NNN-NNNN. If your appointment has been canceled or rescheduled, please disregard this message. Patients and visitors presenting to Penn Medicine locations for inpatient, outpatient, or emergency department care should be wearing a cloth face covering in accordance with current CDC and State guidance. Thank you.

Curiously, the system knew that the source of the call was "Penn Medicine", and reported the message that way to me. But it still led the transcription with "Hello, this is pain medicine". And given that it thinks Covid 19 is "covered. Nineteen.", it can't have been monitoring the news very closely for the past year.

The Word Error Rate is pretty good — about 7-8% depending on how we count the numbers (which it got right).  But the sentence division and punctuation are surprisingly bad.
 



6 Comments »

  1. mg said,

    May 2, 2021 @ 5:48 pm

    When I try to use my Android phone's voice recognition, it still turns "Covid" in "covet" after more than a year of me correcting it each time.

  2. Rick Rubenstein said,

    May 2, 2021 @ 9:19 pm

    Presumably it got "Penn Medicine" from a simple lookup of the originating phone number. I would actually have been quite surprised (and impressed!) if it had leveraged this knowledge in its subsequent transcription attempt, since the aforementioned "it" is almost certainly two utterly disconnected subsystems. (I'll start to worry about the robot apocalypse when such subsystems can be integrated by simply telling one of them "Hey, there's this thing you should know about.")

  3. Orin Hargraves said,

    May 3, 2021 @ 7:47 am

    These sorts of errors seem to me of a pattern; I see it a lot in Google speech-to-text on my phone, which quite happily converts my utterances into words, but does not string them together in a way that could ever occur in natural speech, for grammatical and/or semantic reasons. I expect it's due to what Rick Rubenstein points out, an inability to leverage some kinds of knowledge and apply it across systems. But the priors that win the day often seem to be the unlikeliest from a real-world POV.

  4. Seth said,

    May 3, 2021 @ 8:21 pm

    Just a speculation, but I suspect the "covid"/"covered" issue is because it doesn't do backtracking from the 19. It has "covered" as more common or a better match than "covid" for that sound, while 19 is, well, 19. But it's not putting those together as e.g. "covered nineteen patients" being worse than "covid19 patients"

  5. Chris Button said,

    May 3, 2021 @ 9:38 pm

    “Sentence” for “symptom” makes pretty good sense too. We have the elision of the “p” in “mpt” such that the bilabial “m” is liable to interference from the coronal “t”, and then we have the “s” in “screening” fusing itself on the end to allow the coronal articulation to continue throughout.

  6. AntC said,

    May 6, 2021 @ 6:20 am

    At risk of reprising Geoff's highly amusing voice-activated lift: Alexa Can't Understand Scottish Accent

RSS feed for comments on this post · TrackBack URI

Leave a Comment