Wednesday, July 1, 2020

Transcription Tips

Transcription of oral texts is difficult and time consuming. It is made more difficult when the words are spoken quickly and softly, and when the audio is not optimal (e.g., too much wind, background noise). Adding to the difficulty, one may have several hours of oral texts to transcribe.

In the following post, I give some tips for transcribing, based on my own work. My work involves elderly illiterate speakers with no knowledge of English. Also, I translate the Sasi oral text into both English and Setswana. In other circumstances, there may be easier ways to transcribe efficiently. For example, if the transcriber were a native speaker linguist, or if native speakers were trained to do the transcriptions, it would greatly facilitate the process. But even for those cases, some of the tips below might be helpful.

For simple clearly spoken sentences with a high-quality recording, I assume that you will be able to recognize what was said, and transcribe it with no problem. The tips below are meant to cover more difficult cases.

1.
Divide and Conquer

The longer the sentence is, the harder it is to transcribe. Also, sometimes transcribing part of a sentence really well will give you information on how the rest of the sentence should be transcribed. If you have a long sentence, sometimes you can break it down into parts, such as a subordinate and main clause. Or a sentence may begin with a topic noun phrase, which you can break off from the rest of the sentence. Then you can work on each part separately.

Concretely, you can break up text like this using the functions in ELAN’s Annotation Mode (such as “New Annotation Here”). Then if you wish, you can join the pieces of the sentence together later using “Merge with Next Annotation”.

2.
Building a skeleton

Listen to the line a couple of times, and transcribe any words you recognize in the order in which they appear (leaving space between the words to fill in later). This skeleton forms the basis of your work on the sentence.

3.
Recognize and transcribe common filler words.

In Sasi, many of the filler words are from Setswana, such as mme ‘but’, ga kere ‘isn’t it’, kana ‘or’, gore ‘that’, fa e le gore ‘it it is that’. These words are extremely common, but they are usually spoken very quickly in reduced form. If you know that they occur frequently, it is easier to recognize them.

4.
Recognize and transcribe words from another language.

In the context of my work, it is usually pretty easy to recognize a Setswana word, since the phonological structure of Setswana and Sasi are so different. Since the translator is literate and a native speaker of Setswana, he is able to write down the Setswana word and I can use that in the transcription. There is a large amount of code-switching in Sasi oral texts, so this technique is useful

5.
Recognize and transcribe common phrases.

In Sasi, there are many common phrases like ŋǀuu ma ǃʔaã-a ǁaʔã ʘui ‘I don’t know anything’. These are usually spoken at a very fast rate, sometimes mumbled very faintly. If you did not know the expressions ahead of time, they would be very difficult to transcribe.

6.
Recognize common reductions

There are many morphemes that have slightly reduced versions. For example, in Sasi, kaki (‘with’) is often pronounced as kai. In fact, ki in general (in all its uses, as a plural prefix, causative prefix and linker) is often reduced to i. Also, the word ʘui other is often pronounced kui. Knowing about these kinds of reductions can greatly aid the transcription process.

You will get a feel for filler words, common phrases and common reductions by transcribing texts. So, the more you transcribe texts, the easier the process becomes.

7.
Ask the consultant to repeat the sentence slowly.

If your consultant is able to do this, it is a very valuable transcription tool. With this powerful technique, it is much easier to transcribe quickly or softly spoken words. In my experience, not all consultants are able to do this even with extensive training. You can also ask the consultant to pronounced phrases and words slowly to help in transcription.

8.
Use the grammar to help you.

If you can recognize a particular construction in a line of text, that can help you transcribe the other words. For example, if you recognize that a relative clause is being used, then you know that you need relative clause morphology, and a predicate in the right position. By having a good knowledge of the grammar of a language, you can leverage a very small amount of information to help you with the transcriptions.

9.
Get the translation.

Getting the Setswana and English translation of a Sasi line of text can yield all kinds of useful information for the transcription. For example, if there is a relative clause in the Setswana translation, there will probably be a relative clause in the Sasi sentence. Also, any content words (e.g., the word for “dog”) in Setswana will probably be translations for Sasi content words. Similarly, if there is a smaller phrase that is difficult to transcribe, you could play that phrase for the consultants and ask for a translation. Even a really hopeless and difficult passage can be unraveled with this technique.

The flow of information also goes the other way (from transcription to translation). Once you have an accurate transcription you can then check the translation, and see if it is accurate. In my work, I have a Sasi transcription, a Setswana translation and an English translation. All three have to be consistent, so information gained about one, can provide constraints on the others.

10.
Ask the consultant to define particular words.

If you hear a word in a sentence that you do not recognize, ask your consultants to define it. You can either play the word from the sound file, or say it to them (if your pronunciation of the language is accurate). In many cases, these words will be in the dictionary, where a precise transcription can be found. In some cases, there will be a new word. In that case, I try to get a rough transcription and gloss, which I follow up on at a later date with more systematic lexical elicitation.

11.
Transcribe the next line.

Sometimes if you are having problems with transcribing a particular line, the issues will be resolved in the next few lines. For example, if some word is not spoken clearly, in the next line it might be repeated more clearly. Also, what is being expressed in particular line might become clearer in the next few lines.

12.
Read your transcription out loud.

Once you feel that you are starting to have a good transcription, read the line out loud to the consultants and see if they agree. Sometimes they will find errors, and correct them right away. But if they agree, do not think that the transcription is perfect. The consultants may simply think that what you have read conveys approximately the same information.

13.
Repetition

A brute force method of transcribing is to listen to the whole line over and over again, picking up on a new word each time you listen to it. Once you transcribe a word, you play the sound file again, and your mind can pick up on a new word to transcribe. If you have a difficult sentence of ten words, you might end up playing the sound file ten times or more, which is jarring for everybody.

14.
Word-by-word from a point.

Sometimes when sentences are spoken very quickly, and the transcription is not immediately obvious, it is necessary to play the line word by word, and transcribe each word phonetically. This process can take place from any point: from the beginning or end of the sentence, or from some word in the middle of the sentence.

From the beginning of the sentence, the process works as follows. Suppose the line of text contains the words w-x-y-z. First listen to x and try to transcribe it. Then listen to x-y. Then to x-y-z, and lastly to x-y-z-w. This is laborious process, since each word must be isolated, played and transcribed. But it usually yields a valid phonetic transcription.

From the end of the sentence, suppose the line of text contains the words w-x-y-z, start with z, then y-z, then x-y-z, then w-x-y-z. Transcribing from the end of the sentence is sometimes easier (especially if the end of the sentence is spoken more clearly than the beginning of the sentence).

15.
Be aware of speech errors.

Consultants often start to say one phrase, then back up and say another. Such speech errors occur rather frequently in oral texts. In other words, not every word and morpheme uttered by the consultant is going to fit grammatically into the sentence that they utter. And not every word and morpheme is going to make sense. You can choose to transcribe the speech errors, or choose to omit them. If you are ready for the occurrence of such errors, they are easier to deal with.

16.
Come back later.

You can leave a difficult line not completely transcribed, and return to it at the end of the oral text (or even on another day). Sometimes after the whole oral text is transcribed, the particular issues in a line will be resolved.

17.
Ask a different consultant.

In the worst-case scenario, you can mark the line as (???) and return to it later with a different consultant. They might bring a different perspective. Surprisingly, this technique has worked many times for me. You might even find a consultant who is particularly strong at this task.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.