As the Associate Director of the Berkeley Language Center, most of my time has been devoted to running the day-to-day operations of the BLC, working with faculty, designing new spaces, planning budgets – in short, all the activities involved in running a language center. But in addition to those responsibilities, the Berkeley administration and the two directors I have had the pleasure to work for afforded me time to pursue my particular interest: how we might best use the cultural and linguistic resources found in filmic texts in foreign language education. This freedom eventually led to the release of the Library of Foreign Language Film Clips (LFLFC) in 2009, and for the past three years, we have been diligently working on Lumière (i.e., LFLFC 2.0), which should be operational by the time you are reading this article.
Rather than write another piece on the affordances of film, how filmic texts compensate for the deficiencies in textbooks (and in particular for the canned videos accompanying many textbooks), or how a clip might be used to teach linguistic, cultural, intercultural, and symbolic competences, as well as visual literacy, this article will take a look back at the lessons I have learned in managing a large software project and its redesign, and in the process I will reveal the changes we have made in Lumière. Those readers who were intrigued by the title will be disappointed to learn that there is little if any of Hegel in this article, but rather a view of software design as one of problem-solving, with the solution to one problem generating a new problem, which generates a new solution causing a new problem, and so on.
Creating a structure
Initially, the LFLFC was created for use in language instruction, and as such, one of the early decisions was to make the database language specific, i.e., a Spanish instructor, upon entering the application, sees Spanish films, Spanish clips, and Spanish vocabulary tags. With 200,000+ vocabulary items in the database, it would not make sense for the Spanish instructor to see anything but the 10,700 tagged Spanish vocabulary items. If she happens to also teach another Romance language, say French, she could switch languages to see the films, clips and vocabulary tags in that language. From there it followed that each film would be assigned to a language, which works for the overwhelming majority of films, where any language use beyond the principal language is incidental. However, there are films that cannot be assigned to a single language: is Incendies a French, Arabic, or English film? Is Gegen die Wand German, Turkish, or English? Indochine French or Vietnamese? We thought we had a reasonable workaround when we decided to insert the film into the database in each language where there are sizeable chunks of dialogue in that language. As a result, a film such as Indochine was listed twice, once as a French film, and once as a Vietnamese film, so that when you searched Vietnamese films, Indochine would appear, and the clips where Vietnamese was the primary language would be found there; when you switched to French, Indochine would appear again with any predominantly French clips listed there.
This worked for a while, but by 2015, the LFLFC was being used not only by foreign language faculty, but also by faculty in American cultures, Film Studies, English, Education, etc. They wanted to search across the database and not limit themselves to search first in one language, then another. In Lumière, we enabled searches across languages, but then the films that we had accessioned in multiple languages (such as Indochine) appear in multiple places in the search results – an unnecessarily confusing result.
In an upcoming release of Lumière, we will change the underlying structure such that a film will be entered once with all principal languages spoken in the film identified (Indochine, languages: French, Vietnamese) so that a list of French films and a list of Vietnamese films will both indicate Indochine, and then each clip in the film will identify the dominant language of the clip, and any incidental languages will be tagged. For example, clips in Indochine where French is dominant would be identified as French language clips, and if in that clip there was a phrase of Vietnamese or English, then we add a tag for Vietnamese or English. A clip rarely has two languages both dominating, i.e., one character speaking in one language, and a second character in another – although it sometimes happens. In such cases, we will create two clips, one for each dominant language.
The interest in the LFLFC by faculty outside foreign languages had another impact on the development of Lumière. Faculty in film studies, art history, and other disciplines wanted to be able to create clips but didn’t necessarily want to spend the time to tag the clip. Even faculty in foreign languages who wanted a clip to illustrate cultural practices weren’t interested in spending the time to tag the vocabulary or learn our arcane tagging system. On the other hand, I resisted the impulse to let them create clips and skip the tagging requirement because this would result in a database where some clips are fully tagged, and other clips have no content (no vocabulary tags, no description, no year portrayed, no descriptive tags).
Our solution was to create the notion of the private clip: clips that faculty create for themselves, but which do not become part of the Lumière universe – they can be seen only by the instructor who created the clip, until such time that the clip is fully tagged and “published,” i.e., made available to everyone using Lumière.
The complexity of tags
In the LFLFC we have two types of tags: vocabulary tags and descriptors. Vocabulary tags are the dictionary form of the language spoken in the clip and used primarily as a heuristic for student comprehension (each clip lists for students all the vocabulary spoken in the clip in alphabetical order). On the surface, this seems fairly straightforward, but problems arise when dealing with non-standard forms that are not found in most dictionaries. Common truncations such as “gonna,” “gotta,” and “wanna” are now lexicalized and in most dictionaries, but forms like “wouldja” (would you) are not; furthermore, do we tag “sewn” as “sewn”, or as “sew”? Is there a difference if it occurs as part of verbal construction or as an adjective (“She has sewn …” vs. “the sewn garments”)? Should idioms be tagged as a single unit or broken into their constituent parts?
Vocabulary tags present some difficult choices, but they pale in comparison to devising a system of descriptive tags, which are meant to capture the speech acts (e.g., greetings, persuasion, disagreement), linguistic features (e.g., idiom, metaphor, slang), and cultural content (e.g., health, law, alcohol) in the clip and to help faculty find an appropriate clip for whatever they might be trying to illustrate through the clip, e.g., a Spanish faculty member teaching a unit on the Argentine coup d’etat in 1976 might search on the tag “politics”, or a Chinese instructor interested in traditional medicine might search on the tag “health”.
The tagging system has gone through many iterations. In a pilot project conducted in 2007-8, we employed several Russian and French graduate students and asked them to tag a few films and let them choose the tags that they would use. Clips were submitted, but for most clips, a variety of grammatical terms dominated: cases, verb conjugation types, syntactical features, etc. This may be one way of looking at a segment of film dialogue, and to this day many language instructors ask us to create 10-second clips to illustrate this or that grammatical feature, but in most cases that grammatical feature flies by in a fraction of a second. Is this the best use of a medium rich in culture and meaning-making?
What was being tagged in this pilot were isolated grammatical forms. We threw out this initial batch of clips and started from scratch, focusing on culture and language function. We created a set of tags and invited taggers to suggest additions, but with the understanding that those suggestions would have to be approved. For example, the narration of a sequence of events found in a French and Russian film that generated in the pilot the tag “passé composé” in French and “perfective aspect” in Russian, would now be tagged as “narration”, focusing on function rather than grammatical form.
A second problem with tagging has been the plethora of tags. We started with “car” and “train”, and kept adding more and more vehicle types, eventually ending with “rickshaw”, “bicycle”, “skateboard”; in another semantic field we ended up with “doctor”, “dentist”, “hospital”, “nurse”, “diseases”, “injuries”, “medicine” – one can see that the finer the differentiation between tags, the more tags one needs. A detailed tagging system might be advantageous if you have a language such as French or Japanese with 2000+ tagged clips, but not particularly helpful for a language such as Zulu with only 10 clips. A second problem with a finely tuned tagging system is that it makes searching much more difficult: imagine a scene where a boy runs into his home with a scraped knee and his mother cleans the wound, puts a bandage on, and kisses the knee “to make it feel better” – the clip is tagged for “injury”; a second clip has a doctor talking to a dying cancer patient about his disease and gets tagged for “doctor”, “hospital”, “disease”. How does a faculty member know what to search for? Isn’t looking through a list of 600-700 potential tags a bit much to ask? For the past year, we have been culling through our tag list and reducing the number of tags: the various vehicles are now subsumed under the tag “transportation”, and all of the health-related tags are now subsumed under “health”.
This creates a new problem: in Russian, there are 102 clips tagged for “health”; in Spanish, 65 clips are tagged for “politics”, in Japanese, 338 clips are tagged for “food”. An instructor has no time to sift through even dozens of clips, much less hundreds, to find the right clip to meet her pedagogical goals. How might we facilitate that process?
In the LFLFC we encouraged faculty to leave comments for other instructors, breadcrumbs if you will, that would allow them to point to aspects of a clip that worked (or didn’t) for them. We also encouraged faculty to create lesson plans around clips so that we are not reinventing the wheel from one institution to another. Unfortunately, few instructors make use of these capabilities. We also encourage taggers to be detailed when writing the clip description, but it is hit and miss: would the tagger use the word “injury” in her description of the clip so that a search on injury would generate the clip described above in the results?
We took a slightly different approach in Lumière, where we introduce the concept of clip ensembles. In Lumière, faculty can gather clips on a particular topic under a rubric that they define: this might be an ensemble named “Dealing with authoritarian governments” or “Doctor/patient interactions”, i.e., something much narrower than the broadly defined tags (“politics” or “health”) in the current system. This gives instructors some control over how material is organized in Lumière. Clip ensembles may be “published”, which makes them available to other instructors.
Film as text
When I began using film clips in my own classes, I focused on the film dialogue as a model of conversation that was more “natural” than the scripted dialogues found in our textbooks (despite the fact that those film dialogues were also scripted, rehearsed, and shot multiple times before arriving in the form viewed on the big screen – such is the power of good acting). Following Sherman (2003), I saw film as both a model for language use and as source material for student production (narrating plot, describing characters and scenes, hypothesizing about what might come next, etc.). As I have argued in Kaiser (2018) and a webinar for CERCLL cited in footnote #2, this approach fails to allow students to explore the representations of culture and the meaning-making devices employed in film. Just as language instructors should not ignore the literary devices in printed texts even when their focus is on other matters, I would argue that it is essential that we consider filmic devices when discussing a scene from a film. After all, the meaning of the words in a dialogue is colored by the setting, by the way the scene is shot, the position, facial expressions and gestures of the actors, the lighting, and any accompanying music.
In order to facilitate an awareness of filmic devices, Lumière includes a glossary of 70+ film terms with definitions in English and sample clips of the term in use. We give faculty the ability to add the translation of the term in the language being taught. When ordering a film or clip for student viewing, the instructor may choose to include up to 10 terms with definitions (in English) and sample videos, or all 70+ terms without video, and these are then made available to students on a separate tab when the student is viewing the film or clip.
Other changes forthcoming in Lumière
In addition to a much-improved interface, better browsing filters, a robust search engine, better tools for managing the collection at other institutions (each institution indicates which films in Lumière it owns), more information on how the faculty at the institution are using Lumière, we also added a new feature for students. In the LFLFC, instructors have had the ability to add annotations to the clips they send to students. In Lumière this feature has been extended to whole films. In addition, students will be able to make comments on clips or films that have been assigned to them, and instructors can restrict viewing of the comments to the instructor only, or make them available to all students in the class. Student comments are retained in the Lumière database for the duration of time that the clip/film has been made available to students, but are then destroyed 48 hours after access to the clip/film has been removed. This will enable instructors to follow student reaction to a clip/film as it is being viewed, as well as enable student-to-student interaction while watching the clip/film.
We look forward to continuing to improve Lumière in future years. Inevitably, the changes we have made here will create new problems requiring new solutions. Please do get in touch with me if you have any suggestions for or questions about Lumière.
Kaiser, M. (2011). New Approaches to Exploiting Film in the Foreign Language Classroom. L2 Journal, 3(2). http://dx.doi.org/10.5070/L23210005. Retrieved from https://escholarship.org/uc/item/6568p4f4.
Sherman, J. (2003). Using authentic video in the language classroom. Cambridge, England: Cambridge University Press.
 For readers unfamiliar with the LFLFC, please see Kaiser (2011) or the webinar I did for IALLT in 2016. In short, the LFLFC/Lumière is an online database of 5000+ films and 20,000+ clips cut from approximately 500 of those films. Each clip is tagged for the vocabulary spoken in the clip in dictionary form and for descriptors of the linguistic, cultural and discursive content of the clip. When delivered to students over the Internet, the clip is accompanied with an audio track slowed down by 50%, along with information about the film. The LFLFC/Lumière is provided to other non-profit educational institutions offering courses for credit and for teaching purposes only, but institutions have access only to the films and clips from the films that they have indicated they own. Please go to http://lumiere.berkeley.edu for more information on how your institution can become involved.
 See the bibliography for Kaiser (2018) and the webinar I recently did for CERLLT and which is available on the “About” tab at http://lumiere.berkeley.edu.