Back in the old days when some of us were younger and starting out on our language documentation and description careers (for me in 1972, as described in this blog post) the world was pretty much analogue and we didn’t have digital hardware or software to think about.
Back then recordings were made with reel-to-reel tape recorders, like the Uher Report, or if you had really fancy kit a Nagra. For those of us working in Australia on Aboriginal languages you could archive your tapes at the Australian Institute of Aboriginal Studies (AIAS), as it then was, later the Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS). They would copy your tapes onto their archive masters and return the originals to you and all you, as a depositor, had to do was fill in tape deposit sheets. You were supplied with a book of these, alternately white and green, with a sheet of carbon paper to be placed between them. For each tape you had to complete a white sheet listing basic metadata and a summary of the contents of the tape, tear off the white copies (keeping the green carbon copy) and submit them to the AIAS archive. In addition, the Institute encouraged the preparation of tape audition sheets where the content of the tapes was summarised alongside time codes (in minutes and seconds) starting from the beginning of the tape. Sometimes these were created by the depositor and sometimes by the resident linguist (at that time Peter Sutton).
So, if you wanted to find out where in your stack of tapes you could find Story X by Speaker Y you simply had to look at the deposit sheets and/or the audition sheets.
Alas, those days are gone and we are in the digital world, where our experience is mediated via software interfaces that can fool us into seeing the world the way the interface presents it. For language documenters Toolbox is often the software tool of analytical choice (along with ELAN)1 for the processing and value adding analysis and annotation of recordings. As I claimed in a previous post, the existence of Toolbox means that for many documenters annotational value adding only means interlinear glossing, and alternatives such as overview or summary annotation (like the old tape audition sheets) are not part of their tool set. I have two pieces of evidence for this:
- the Endangered Languages Archive (ELAR) at SOAS has so far received around 100 deposits comprising roughly 800,000 files. Among these deposits there are many that are made up entirely of media files (plus basic cataloguing metadata) with no textual representation of the content of the files beyond a short description in the cataloguing metadata. When asked about annotations, depositors typically respond that they “are working on transcription and glossing” but because of the time needed they cannot provide anything now. They do not seem to consider an alternative, namely time-coded overview annotation which can (and probably should) be done for all the media files, only some of which would then be selected and given priority for interlinear glossing. Why? One reason might be because there is no dedicated software tool designed and set up to do this in an easy and simple manner (interestingly a tool that can be so used, and that produces nice time-coded XML output is Transcriber, though it is generally thought of as a tool for transcription annotation only — it also does not have a “reader mode” that would allow for easy viewing and searching across a set of overview annotations created with it);
- during training courses and public presentations over the past couple of years I have been warning that current approaches to language documentation risk the creation of “data dumps” (which I have also called “data middens”) because researchers are not well trained in corpus and workflow management and additionally suffer from ILGB or “interlinear gloss blindness” which drives them to see textual value adding annotation in terms of the interlinear glossing paradigm2 The most recent example of such a presentation was during last months grantee training course at SOAS (the Powerpoint slides from my presentation are available on Slideshare). All but one of the grantees attending the training had never heard of, or considered creating, overview summary annotation before launching (selectively) into transcription and interlinear glossing of their recordings.
I may be wrong about the source of the current ILGB and perhaps Toolbox is not (solely) to blame, but I do believe that it plays a part in a narrowing of conceptual thinking about annotation in language documentation, and hence the behaviour of language documenters.
NB: Thanks to Andrew Garrett for his comments on my previous post that caused me to think more deeply about these issues and attempt to explicate and exemplify them more clearly here.
FollowNotes
- ELAN is a tool designed for time-aligned transcription and annotation of media files, and is also widely used by language docunenters, bringing with it its own kind of window on the world that I do not discuss here
- There may be a separate further dimension to be concerned about that results from the shift from analogue to digital hardware, rather than being a software issue. In the old days tapes were expensive and junior researchers in particular only had access to a rationed supply and therefore had to think seriously about what and how much to record. Today, with digital storage being so cheap and easy to use (especially for copying and file transfer), there is a temptation to “record everything” on multiple machines (one or more video cameras plus one or more audio recorders) and not write much down because “you can always listen to it later”. This can easily and quickly give rise to megabytes of files to be managed and processed. I saw this temptation among the students taking my Fieldmethods course this year — they learned after a few sessions of working with the consultant this way about the pain that then comes from the need to search through hours of digital recordings for which they had few fieldnotes or metadata annotations.
The problem is compounded by the ease with which a researcher can collect more data than they can reasonably process (annotate, gloss, etc.) in a useful timeframe. The mindset these days is to record as much as you can, preferably in video as well, and then deal with the mountains of raw data later. My impression of the analogue days, feel free to disagree, Peter, is that researchers would reserve their recording capacity for the really useful interviews, since each tape was another weighty thing that had to be carried with them. Listening to Osborne’s recordings of Tiwi, I get the sense that for each hour, there were probably four or more hours of preparation, which for all intents and purposes would resemble today’s elicitation sessions; questions that go nowhere, false starts, failed attempts at speaking, and testing out analyses. The actual recording was more like a prepared interview; the researcher asking for forms as if reading off a script and the interviewee diligently providing each answer twice. As evidence, a few times I hear Osborne disagreeing with the interviewee, and telling him that it’s “Y, not X, X means such-and-such”.
Obviously we can do better than that these days, and I find the sessions where I’m getting things wrong, being corrected, making false starts and obviously trying out mistaken analyses, tend to be more useful. But surely recording everything, and coming back from the field with 100+ hours, a fifth of which is useful, is madness.
Also, I think it’s a good idea as a first pass when transcribing, to do as you suggest, and broadly annotate the recording with the rough contents. It also helps then when you’re properly transcribing, to start and stop at natural positions where the topic changes, rather than in the middle of a section. I just use another tier in Elan for this.
I don’t disagree with your ideas, but another factor does play a role, namely the career aspect. I’m not entirely sure people will want to give up interliner glossing because it would split language documentation and description even further. With it, you can still combine descriptive linguistics with language documentation, and a language documentation project you worked on will still work just fine for later work at a university writing a grammar or more analytical articles. The recordings and such don’t go away, but it does mean a lot more work later on if you want to do analysis.
@Don — I’m not suggesting giving up interlinear glossing. What I am saying is don’t be blinded into thinking that it is the only way to do annotation, and indeed it should be done selectively. To organise and manage a corpus it is better to use other techniques in combination with it, overview annotation of the whole corpus (as originally promoted by my colleague David Nathan) is one of those.
I also find your idea that this comes from Toolbox interesting, incidentally. I’ve had discussions with other linguists about Flex, many of whom have disliked it precisely because it pigeonholes you even more into certain ways of thinking. I don’t want to criticize SIL too much about this, as I do appreciate all that they do. But I’m also not aware of a single alternative program for either Toolbox or Flex developed by any academic or secular scholar (alternative, meaning used for the same purpose.. ELAN just has different functions in my opinion, complementary but not overlapping that much). Perhaps if our tools changed, so would our mentality?
@Aidan — thanks for your comment which does add another element (I was tempted to include it when I wrote this post last night but it was getting too long). If you go back even further, to Stephen Wurm’s tapes from NSW and Qld you will see total economy of recording because on his 1955 trip he only had £100 from Capell to cover his whole NSW-Southern Queensland trip and a handful of tapes. He recorded only samples and then only the speaker uttering language forms in a stop-start fashion (you can occasionally hear him shouting “Now!” just before someone speaks). When I started my Diyari fieldwork Bob Dixon told me to be sparing with my use of tapes, and not to use them like Dr XXXX who “used tapes like toilet paper”. There is a downside to this approach, of course, and one you mention, which is the value of recording “mistakes” that can later make sense once you have understood the language structure and use better (if you manage to go back to them, of course!). Tamsin Donaldson once told me she regretted Bob’s advice not to record elicitation sessions and to only record the Ngiyampaa bits of texts and NOT the English translations or commentaries. Now she can’t go back and hear what her consultants had to say in English about the materials.
I know of at least one DoBeS grant that was given money for a 1-year extension with the explicit instruction to NOT collect any more data but to work on processing the mountain of digital files that had already been collected in the previous 4 years. When I mentioned this in a recent workshop there were gasps of horror from some members of the audience (“But you might miss something” being one response I got).
I’m not sure how practical this would be on a large scale, but it’s possible to do summary annotation with Praat (by adding a tier to the bottom on which you only put boundaries for large blocks of discourse; you can type in as much description as you want, although it won’t display well at all zoom levels). Since the annotation files are plain-text with time stamps, you could make multifile “maps” out of them with some tinkering.
Yet another side-effect of excessive focus on interlinear glossing might be the marginalization of linguistic theory in some strands of language documentation. When I was a graduate student, studying mostly theoretical syntax and semantics at UCLA, although faculty were supportive of fieldwork on diverse languages, IGT was never pushed as an end goal. Sometimes it was useful, but the idea of systematically glossing texts was definitely seen as less important than the advances that could be made by analyzing those texts for specific features of interest, or discussing those features with native speaker consultants in elicitation. Peter’s point is so much about allocation of time and resources: is the attention paid by fieldworkers to glossing commensurate to its actual importance?