Research at the Lacito

Abstract:

The COCOON platform hosts many oral corpora. It allows to archive documents and make them publicly available, and also to add annotations to audio files. During this talk, I will show how the platform works et how I used it for my deposit. Data archiving and diffusion raises several questions: Which platform(s) to use? Must all oral data be made available? If not, what are the criteria for making data available? What is the purpose of archiving annotations? Which public(s) can be interested in the archived corpus? The meeting will be the occasion of discussing these topics.

Abstract:

ELAN is one of the most frequently used tools for linguistic annotation of audio and video recordings. A high number of linguists are familiar with the process of annotating files in ELAN, but surprisingly few are aware of its remarkable potential as a tool for searching said files. In our tutorial, we will show how the structural search multiple Elan files function works. We will begin with simple queries for when you need to find a specific word / gloss / translation in multiple Elan files and then step by step we will complexify our queries to ones with several layers and ones with regular expressions.

Target audience of this tutorial are linguists who already have files annotated in Elan and who intend to learn how to use Elan as a corpus query tool. However, if you just start to use Elan and do not have many annotated files or just thinking of starting using it, the tutorial will show the potential of the software.

Abstract:

Imagine that you recorded some very important data in the field and you later discovered that they cannot be opened by your working softwares. Imagine that you have short audio files and you want to merge them into one single long audio file.  Imagine that the sound quality of your videos is better than the one of your audios and you need to extract that sound from the video Imagine that the sound of the video is bad and you need to replace that sound by a better quality audio file. Imagine that you have heavy videos and you want to compress them without losing the original quality. Imagine that you saw some pictures or graphics in a research paper that you need to use in your study, but you cannot because they are embedded in the PDF. Imagine that you have lot of PDF files that you need to join in one file. Imagine that you have a secret document that you want a restricted person to be able to access it.  If you are in one of the above situations, then format factory, multimedia files converter is what you need. In this session you will learn how to:

1. Convert your multimedia data in various format depending on your need
2. Merge several audio files in one
3. Extract an audio file from a video
4. Separate an audio and video 
5. Extract some pictures from a pdf file
6. Encrypt a document
7. Download some videos online

Abstract:

The linguistic community regularly updates our tools for language description and documentation. Sometimes, an upgrade is so big that a new tool is set to replace an important tool that has been used for decades. FLEx (FieldWorks Language Explorer), for instance, is steadily replacing Toolbox as the most commonly used dictionary building application. In these cases, it is good to get used to the newer tool sooner rather than later, as developers and support teams reduce the maintenance of older tools and put more energy towards developing and supporting newer ones.

One major challenge is to migrate our data from one environment to the next. We want to keep our data and our analyses, and we want to avoid duplicating work by manually typing entries in our FLEx database that we already have in Toolbox. Unfortunately, it is sometimes difficult to find tutorials on how to do this.

In this session, we will demonstrate how to migrate a lexical database from Toolbox to FLEx. We will illustrate this with a lexicon of Dijim [cfa, diji1241]. The demonstration will take about 30 minutes. If time allows it, we can help others who are struggling to transfer their lexical data to FLEx and find solutions to any questions the participants may have.

You can download FLEx here: https://software.sil.org/fieldworks/ . Some tutorials are available here: https://software.sil.org/fieldworks/resources/tutorial/ .

Abstract:

Resumes the first session

Abstract:

No abstract

Abstract:
We will present a method to upload your recording sound files into FLEx and align them with your texts. You don’t need to know a lot about FLEx, but this meeting will be especially useful for you if you annotate (or intend to annotate) your texts in FLEX and want to be able to easily play any sentence while you’re working. This does not exist as a feature in FLEx, but it can be made possible.

Abstract:

The LLACAN and LACITO laboratories host a number projects exploring the potential of computational methods to facilitate (endangered) language documentation. Machine learning-based tools can effectively assist with linguistic annotation tasks including transcription, glossing and translation. Still, automatic processing remains little used, especially because the technology is still new (and evolving rapidly), and there is a lack of simple and user-friendly interfaces. Our laboratories aim to co-construct models and tools between field linguists and computer scientists.

In this context, explorations in Automatic Speech Recognition on field data are ongoing. After a general presentation of the “Elpis” project, current experiments will be presented (which use the Transformers from Huggingface and wav2vec Unsupervised from Facebook AI). The goal is to introduce the workings of the tools to a linguistic audience and to discuss the challenges of interdisciplinary collaboration with computer scientists.

Some references :

• Providing “field” linguists with access to automatic transcription
• https://hal.archives-ouvertes.fr/hal-03047148
• User-friendly automatic transcription of low-resource languages: Plugging ESPnet into Elpis
• https://halshs.archives-ouvertes.fr/halshs-03030529
• Integrating automatic transcription into the language documentation workflow: Experiments with Na data and the Persephone toolkit
• https://halshs.archives-ouvertes.fr/halshs-01841979

Abstract:

Maxime will present an essential tool for submitting texts or word lists to the Pangloss collection (https://pangloss.cnrs.fr/). The tool is a Python script that converts textgrid files (produced in Praat or ELAN) into XML files supported by Pangloss. The tool greatly simplifies the depositing process. Using this script is very simple but requires some configuration and a very basic understanding of how Python works.

Maxime will demonstrate the script with a guinea pig that already has Python installed (Neige). All participants are welcome to try the tool during the presentation, but there may not be time for technical support. Consequently, the session will essentially be a presentation, not a full tutorial. If you have never worked with Python, the presentation will allow you to judge whether it could be useful for you to use the tool and to learn the basics of Python (which is simple but will not be the subject of this session).

If, at the end of the session, enough participants are interested in this script or in learning Python, we will consider organizing a session (perhaps beyond the usual times of the seminar) devoted to getting started with Python and using pre-made scripts. This may also be relevant if you wish to use the script for identifying minimal pairs in a word list. That script will be the subject of a later session.

No abstract

No abstract