Reading an address out loud

Dictation involves reading, saying, hearing and writing, and each of the four can go wrong independently. The reader can lose their place in a long unbroken string. The speaker can produce a sound that maps to more than one symbol. The listener can resolve an ambiguous sound the wrong way. The writer can transcribe correctly and still lose their place.

None of these is unusual. What makes an address different from a phone number is that all four failures are silent. Nothing in the process produces a signal that something went wrong, and the result surfaces much later as an address that does not work, with no way to tell which of the four steps failed.

The alphabet contains several letters that sound alike when said quickly, and it contains digits whose names collide with letter names. The traded symbol gauge counts the visual version of this problem, and the spoken version overlaps with it heavily without being identical.

That is worth noting carefully: a symbol that is easy to confuse by eye is not necessarily easy to confuse by ear, and the reverse is also true. Dictation therefore adds a second, different set of confusions on top of the ones a reader already has, rather than replacing them.

A 56 character dictation is not a short one repeated several times. It is long enough that the reader loses their place, and losing your place in a string with no repeated chunks means there is no way to work out where you were.

The vowel run measure is directly relevant here. A long run without a vowel cannot be grouped into syllables, so the reader has nothing to chunk with, and both parties are counting characters rather than following a rhythm. Counting to 56 out loud while also reading symbols is where most attempts collapse.

• Use a phonetic alphabet for every letter, without exception. Switching between plain letters and phonetic ones is where mistakes concentrate.
• Say digits with a distinguishing word. The digit names in this alphabet collide with letter names more than people expect.
• Group in fixed blocks and say the block number before each block. That gives both sides a place marker to recover from.
• Have the other person read it back completely. A read back is not optional here, because there is no other way to detect any of the four failures.
• Replace it with a copyable version as soon as one exists, and throw away the dictated copy.

Do not do this. There is almost always a way to move text that does not involve a human voice, and the difference in reliability is not marginal.

This panel exists because people do it anyway, usually in a hurry, and a list of the specific failure points is more useful than a general instruction not to. But the instruction stands: an address is data, and data should move as data.

The four failures above are all about the two people involved. There is a fifth thing worth naming, which is that saying an address aloud puts it into a room.

A dictated string is audible to anybody present, recoverable from any recording that happens to be running, and memorable to nobody but reconstructible from a partial hearing plus the shared opening. None of that is a technical problem and all of it is a practical one.

The board is not going to give advice about who is in your room. It is worth saying only that dictation moves an address out of the channel you chose and into the ambient one, and that this is a property of speaking rather than of the address.

Is a read back enough to catch errors?

It catches most of them, because the failures are usually not symmetric. It is the single highest value step and it is the one people skip when they are in a hurry.

Does spelling it phonetically remove the problem?

It removes most of the sound collisions. It does not help with losing your place, which is the other half of the difficulty.