• No results found

Transcribing and annotating audio and video: Jeff Good MPI EVA and the Rosetta Project

N/A
N/A
Protected

Academic year: 2021

Share "Transcribing and annotating audio and video: Jeff Good MPI EVA and the Rosetta Project"

Copied!
36
0
0

Loading.... (view fulltext now)

Full text

(1)

Transcribing and annotating

audio and video:

Jeff Good MPI EVA and the Rosetta Project [email protected]

(2)

Goals of presentation

Discuss basic concepts of audio and video transcription and annotation

Illustrate process of transcription and annotation using Elan
(3)

3

What is annotation?

In recent years, a new conception of language documentation has been emerging (see, e.g., Himmelmann (1998), Woodbury (2003))

This view takes primary sources of data (e.g., audio and video) to be the foundational materials for language documentation
(4)

What is annotation?

“Traditional” linguistics is then conceptualized as annotations on primary data, including

Transcription of audio or video

Annotation for grammatical
(5)

5

Text annotation example

Cicko,

[

ch’aara ’a goj,

]

’i

bu’u.

cat.

erg

fish

&

see.

cvpan 3s.abs b

.eat.

prs

‘The cat sees a fish and eats it.’

1

Extra layer of annotation

(6)

Why does this matter?

It’s a pretty different way of doing “language documentation” than before

It forms the conceptual underpinnings of the functionality of annotation tools

It can be a lot more work at first...

...with (hopefully) a worthwhile payoff
(7)

7

Good annotation

Under present thinking, good annotations should have the following properties

Archival format

Time-aligned to primary data

Transparent, documented terminology
(8)

Archival format

How do you make sure your

annotations are in an archival format?

Short answer: Use a tool designed for research purposes (e.g., Elan, Shoebox)

What not to do?

Use FileMaker, Microsoft Word, etc.,
(9)

9

Archival format

Simple tip: If the annotation file isn’t designed to be easily opened in a

plain text editor (e.g., Notepad, TextEdit), it’s not archival

The biggest mistake people make isn’t deliberately choosing a program that uses a “bad” format—it’s not even

thinking about formats before using some program

(10)

Time-aligned annotation

When you’re annotating audio and video, ideally, you want the

annotations to be time-aligned

That is, you want them to be “linked” to appropriate sections of the audio

and video recording

This allows you or another researcher to have access to the primary data on which an annotation is based
(11)
(12)

Terminology

When doing annotation of linguistic data, there will always be a need for specialized terminology

For example, transcription systems, like IPA, are a type of specialized

terminology

Interlinear glossing also uses

specialized terminology (e.g., “sg” for “Singular”—itself a specialized term)

(13)

13

Terminology

When possible, use existing standard term sets (and document that you’ve done this)

For example, IPA with notes on any modifications/interpretation

Leipzig glossing rules for interlinear abbreviations
(14)

Terminology

Document the use of any special

conventions you devise for your data

Develop controlled vocabularies and make use of any features of your tools supporting their use

Controlled vocabulary: A

standardized list of terms used for annotating data

(15)
(16)

Terminology

Possible controlled vocabularies

Yes/No

Speaker identifiers

Left/Center/Right (for eye gaze)

Grammatical phenomena of particular interest
(17)

17

Elan

Elan is a time-aligned annotation tool available at:

http://www.mpi.nl/tools

Supports annotation of

Audio (in WAV format)

Video (in MPEG and Quicktime format)
(18)

Elan

Noteworthy features of Elan

Designed in the context of language documentation

Supports Unicode

Export/import of Shoebox files
(19)

19

Annotation tiers

In Elan, tiers are where annotations are located

Tiers can be thought of as a “line” in an analyzed text. For example:

Transcription tier

Morpheme-analysis tier

Interlinear tier
(20)

Anatomy of an Elan window

Annotation viewer

Wave form

(21)

21

Elan in action

A brief demonstration of Elan, including

The tiers I’ve been using

Making a new annotation
(22)

Elan’s tier types

Time-aligned

The “foundational” annotation,

directly aligned to audio or video.

Typical example: sentence transcription
(23)

23

Elan’s tier types

Time subdivision

Must be linked to a basic time-aligned tier

Allows you to make subdivisions of that tier with their own timestamps

Typical example: Words in a sentence
(24)

Elan’s tier types

Symbolic subdivision

Must be linked to another tier

Allows that tier to be subdivided without times associated with the subdivision

Typical example: Morpheme subdivision
(25)

25

Elan’s tier types

Symbolic association

Must be linked to another tier

Cannot be subdivided further
(26)

Schematic example of tier types

Puer puellam amat.

puer puellam

amat

puer puell am am

a

t

boy girl ACC love PRS 3s

The boy loves the girl.

Sentence-level transcription time-aligned with wave form

Time subdivision of sentence into words Symbolic subdivision of words into morphemes

Symbolic association of morphemes with glosses

Symbolic association of sentence transcription with free translation

(27)

27

Tier types

These tier types weren’t invented out of the blue for Elan

They correspond to the “meanings” of different kinds of linguistic parsing

The tool designers allow for flexibility of tier types (a good thing)

It’s up to the linguist to understand

their data well enough to use the right tier type

(28)

Tier templates

It is likely that for a given project

you’ll have some tier sets you’ll use often

Elan provides the ability to save a set of tiers as a “template” that can be
(29)

29

Aside: .eaf files

How does Elan store annotations?

Inside XML files using an .eaf extension

What does that mean?

Your archivist will be happy

Hopefully, you’ll never need to know more than that
(30)

Elan conclusion

This is just an introduction

Elan has more features than I have made use of or can describe here

My wish list for features

Integration with a phonetic analysis tool (e.g., Praat)
(31)

31

Other tools

There are a number of annotation tools out there

Two that seem to also be popular among linguists

Transcriber (http://trans.sourceforge.net/) (apparently good for conversational recordings)

Praat (http://praat.org) (primarily known as phonetic analysis software, but also has facilities for time-aligned annotation)
(32)

Why annotate?

Time-aligned annotation is a lot of work

For me, it’s much more time

consuming than just jotting things down in a notebook

(33)

33

Why annotate?

Intangible reasons

It’s currently considered good documentary practice

It facilitates wider use of resources by other people

Time-aligned annotations combine linguistic analysis with the
(34)

Why annotate?

Tangible reasons

It allows you (and others) to double-check your analysis more easily

The ability to do searches across structured annotations facilitates analysis

Makes creation of sound “clips” much easier
(35)

35

Conclusion

If you’re going to go through the trouble to make good recordings...

...it’s worth going through the trouble of annotating them well.

Unsure of how to proceed?

Consult the E-MELD School of Best Practices
(36)

References

E-MELD School of Best Practices

http://emeld.org/school /

Leipzig glossing rules

http://www.eva.mpg.de/lingua/files/morpheme.html

Himmelmann, Nikolaus P. 1998. Documentary and descriptive

linguistics. Linguistics 36:161–195.

Woodbury, Tony. 2003. Defining documentary linguistics. In P.

Austin (Ed.) Language documentation and description, volume 1,

33–51. London: Hans Rausing Endangered

References

Related documents