Transcribing Dialogs, Including Non-words

Transcribing Dialogs, Including Non-words
from "The Trains 93 Dialogues: Trains Technical Note 94-2

J. Word Annotations

The words spoken are entered in ESPS annotation files that end in .words, which are associated with each utterance file. Each word is indicated at the ending point of the word in the speech signal using its orthographic spelling. This should be the point that best separates the current word from the next so that when listening to each word individualy, there is a minimum of sound from the neighboring words. This will usually be at a trough in the amplitude of the speech signal.

J.1 Silences and Breaths

In order to capture word durations, we also mark other phenomena in the word layer. This includes breaths within the utterance, and noticeable silences, which disrupt the flow of speech. As with words, these are labelled at the end of the breath or silence, with a <brth> or <sil>, respectively. Only silences that are perceptible should be transcribed.

There is typically a silence or a breath at the beginning of an utterance file. These also need to be marked so that the duration of the first word can be computed.

J.2 Disfluencies and Partial Words

When possible, words should be transcribed using their standard spelling; do not try to capture the way a word was spoken by modifying its normal spelling. Disfluencies that are not identifiable words should be transcribed as they sound, with a dash, `-', at the end of the word. These include words that are so misspoken that the intended word is not recognizable. In cases when the word is recognizable but cut off, it should be labelled by spelling as much of the word as can be heard, followed by a dash. For example "oran-" would label the word "orange" that was interrupted before the "g" sound.

If the word that was cut off midstream, but the word that the speaker intended to say is obvious, then the rest of the word should be included in parentheses. For instance, in the above, it would be transcribed as "oran(ge)-".

Sometimes when a speaker is trying to interrupt the other person, she will start to say a word and then give up. The resulting sound should be transcribed as a partial word.

When multiple sounds are emitting in a row, they should be transcribed as a single partial word, unless there is a perceptible silence in between the two sounds.

J.3 Capital Letters and Proper Names

Capital letters should only be used to begin proper names, such as "Corning" and for the pronoun "I". Engine names, such as "E2" are problematic, since they could be viewed as a proper name. However, we will transcribe them as two words "E" and "two'.

A partial word should only be capitalized if its completion is clear, and that completion is a proper name. In this case, the completion will be put in parentheses. So, for the city "Corning", "Corn-" is not exceptable, but either "Corn(ing)-" or "corn-" is (the second case is where it is not clear to the transcriber that the speaker was going to say "Corning".

J.4 Mispronounced Words

Mispronounced words should in general be spelt correctly, unless the misspelling is needed in order to understand the subsequent dialogue. For instance, if the speaker says "Dansvull", and later corrects herself by saying "Dansville", then the first occurance should be spelt in a way that reflects the mispronunciation.

J.5 Numbers

Numbers should be spelled out in full, as they are said. For instance, if a speaker says "104" as "one oh four", this is what should be transcribed, as three words.

J.6 Times

Each component of a time, should be spell out in full. The words "a.m." and "p.m." should be spelled as such. For instance "four p.m." should be annotated as two words.

J.7 Common Words in Spoken Dialogues

There are a number of words that are common in spoken dialogues that do not occur in written forms. This section discusses how such words should be transcribed. Where possible, we use the spelling from Quirk et al. (1985).

Filled Pauses Filled pauses are very common in natural dialogue. There seem to be two types, ones that sound like "uh" and ones that sound like "um". The endings of these words are often prolonged, thus tempting transcribers to label it is "ummm". Rather, these words should be classified as either "um" or "uh", and transcribed as such. We also include "er", which is more common in British accents. Note that the filled-pauses should never be transcribed as partial words.

um Filled pause.
uh Filled pause.
er Filled pause. More common in British English.

The following is a list of commonly occurring acknowledgments, and how they should be spelt.

okayAgreement. Speakers will often produce variants of this, such as "kay", "mkay", "umkay". All of these variants should be spelt as "okay".
mm Agreement, stalling for time
huh Request for clarification. Puzzlement.
hm Stalling for time.
nah Informal version of "no".
nope Informal version of "no".
a-ha Interjection denoting surprise, as in "aha< I found it", rather than "uh-huh" as an acknowledgement ha Interjection, similar to "aha".
oh Surprise. ooh As in "ooh, that's gross."
yeah Informal version of "yes"
yep Informal version of "yes"

Contractions, that are common, should be written as one word. The following is a list of common contraction endings. Note that there can often be an ambiguity as to whether the speaker was saying the words as one or as two individual words, especially since words are often blurred together. If in doubt, annotate the word pair as two separate words, spelling out the second in full.

'llfor "will"
'vefor "have"
n'tfor "not"
'refor "are"
'sfor "is"

All other contractions are left to the transcriber's discretion as to whether they should be transcribed as one word or two.

Word Pairs There are some word pairs that are so altered (in pronounciation) that they seem to be one lexical item. Such pairs can be transcribed as single words. Below, we give some common word pairs.

lemmefor "let me"
wannafor "want to"
gonnafor "going to"
gottafor "going to"

J.8 Other Phenomena

Laughter, tongue clicking, and other phenomenon can be transcribed in the word tier if it does not overlap with the speaker's speech. Such phenomena should be enclosed in angle brackets. Common transcriptions are given below.

<click>Obvious tongue-clicking
<clicks> Tongue-clickings. Usually made to fill a silence.
<clear-throat> Throat clearing, and coughs.
<mumble> Incomprehensible word or words. Perhaps not even intended for the other conversant to hear.
<noise> Noises that the speaker unintentionally makes, as well as other noises.
<filled-pause> Speaker-specific sounds that the speaker makes to fill a silences.

Extracted from PostScript by R. Paul McCarty 2001/07/30