Transcription, transliteration

✎ This article/section is a stub — some half-sorted notes, not necessarily checked, not necessarily correct. Feel free to ignore, or tell me about it.

(Outside of linguistics, these are confused with some regularity)

Transcription

Transcribing usually means taking sound and writing its contents down.

Regularly to that same language's writing system.

In linguistics, particularly in the context of phonetics or dialectology, it mat transcribes to a phonetic script, such as IPA.

Speech segmentation, forced alignment

Speech segmentation is figuring out where meaningful units begin and end.

It can refer to the cognitive process, as well as the artificial imitation used in e.g. speech recognition.

Forced alignment, or force-aligned transcription, means automatically aligning audio files with existing transcripts.

In other words, "Assuming that this was what was spoken, fit the audio to this"

Transliteration

Transliteration is the process of transcribing from one writing system to another.

Transcription often strives to be simple to understand and apply in the target language, so phonetic accuracy is often, while a concern, may be secondary.

It often seems done to make it easier to learn.

For example, japanese has 23 phonetic syllables, which are easier to learn than the 46 symbols of hiragana and katakana, and certainly much simpler than the thousands of ideograms, that all map to the same sounds somehow. With transliteration, the sounds can be learned somewhat separately from the 46 symbols.

Romanization is transliteration into the latin alphabet. Japanese is strictly phonetic and the syllables so lends itself well, and even alphabets like Cyrillic are quite phonetic so Romanized well enough.

Converting Latin-alphabet words to systems like Japanese and Russian is harder, mostly because not all sounds have obvious counterparts. For example, Japanese has the problem that the only thing that can be written must adhere to its syllables, so you have to insert sounds and/or use a similar consonant to get to a close syllable.

Other problems include the fact that you're basing phonetic conversion on written characters may mess up digraph cases and such.

Transliteration as an input method

One use is input methods on computers. Languages like Japanese, Chinese, Russian and many others can be typed phonetically with Latin characters, which usually works out as a pronunciation source for respective characters, as you type. This allows western keyboards to be used for these languages.

It will require you to only use characters that will convert, and may force you to choose between alternatives. Neither is a practical problem.

Overlap

Transliteration and transcription are easily confused.

Romanization can indicate transcription as well as transliteration, because it only indicates the target alphabet - in this case only the source makes a difference. When you want to write Russian and Chinese in Latin characters, you would transcribe Chinese and transliterate Russian.

You need to convert Chinese via its pronunciation; the logographs are themselves not phonetic.

Starting with Russian, a phonetic alphabet, you can transliterate it into English and into various other languages (slightly differently, since you keep pronunciation and writing system in mind), while writing it as IPA would be transcription.

-->