Speech recognition for newly documented languages: highly encouraging tests using automatically generated phonemic transcription of Yongning Na audio recordings

Automatic speech recognition tools have strong potential for facilitating language documentation. This blog note reports on highly encouraging tests using automatic transcription in the documentation of Yongning Na, a Sino-Tibetan language of Southwest China. After twenty months of fieldwork (spread over twelve years, from 2006 to 2017), 14 hours of speech had been recorded, of which 5.5 hours were transcribed (200 minutes of narratives and 130 minutes of morphotonology elicitation sessions). Oliver Adams, the author of mam, an open-source software tool for developing multilingual acoustic models, volunteered to experiment with these data. He trained a single-speaker automatic speech transcription tool over the transcribed materials and applied it to untranscribed audio files. The error rate is low: on the order of 11% of errors in phoneme identification. This makes the automatic transcriptions useful as a canvas for the linguist, who corrects mistakes and produces the translation in collaboration with language consultants.

Today I am posting my initial report to Oliver Adams, written in the field in Yunnan, on May 12th, 2017. I consider this date as a landmark in my work documenting Yongning Na! This document remained confidential until today because its online availability would have de-anonymized a submission by Oliver Adams & collaborators that was under double-blind review for a conference in Australia. (This paper, which presents Oliver’s work to an audience of language technology specialists, is now accepted, and will be finalized in the next few days; in the meantime, the submitted version is available here.)

Initial report to Oliver Adams
about the usefulness of the automatically generated phonemic transcription of Yongning Na audio recordings by speaker F4 (Mrs. lɑ˧tʰɑ˧mi˥ ʈæ˧ʂɯ˧lɑ˩mv̩˩)


I am delighted that you took the time to develop a statistical model for this particular speaker of this particular language. I was initially skeptical about the usefulness of having a first-pass automatic transcription to improve manually. My expectation was that having to correct mistakes generated by the algorithm would add to the workload (wading through an imperfect transcription), and that this would not be faster than the old method (inputting from zero, without an automatically generated transcription to start with), and less straightforward. But as soon as I began using the transcription, I found it pleasant and useful. I am keen to continue the exchange and participate in plans such as the deployment of this tool for all the languages in the Pangloss Collection.

Detailed report

After receiving your sending on May 3rd, a quick preliminary operation was performed : removing all spaces (automatically), so as not to have to remove spaces manually inside each syllable, thus : l ɑ ˩ to lɑ˩. Then I set out to use it as a basis to produce a full annotation of the classical type: transcription plus translation and comments.

  1. My experience of listening to sentences and adjusting the transcription

When revising the automatically generated transcription, I add tone group breaks (the  |  signs), punctuation, and nonlinguistic gap-fillers: əəə... and mmm... (oral filled pauses and nasal filled pauses). I adjust (correct) the transcription and add comments (lines starting with %) and a sentence-level translation in French. (I add Chinese later on, and English even later.)

I tried two methods : (i) improving the transcription on my own, listening to the recording, and making a note of passages where there is something that I do not understand or am not fully sure about, and which I need to ask the consultant later; (ii) working with the consultant, advancing through the automatically generated transcription together with the consultant, and asking her questions as they arise. Both methods work. The former is good for situations where the consultant’s availability is limited: in a short work session, the work focuses on the points that require the consultant’s expertise. The latter is safer and more luxurious for the linguist: as I listen to the audio and amend the transcription, I read aloud each sentence of the corrected transcription to the consultant, and she can verify if it’s OK, which is always good. Listening sentence after sentence is also more pleasant for the consultant (and, again, safer) because the context is fully clear thanks to the linear progression through the audio; by contrast, ‘jumping’ from one spot to another for verification requires the consultant to remember the context, which is feasible but requires more effort than linear progression through a document.

Using the automatically generated transcription, I completed transcription of the HOUSEBUILDING2 narrative, a document of 29’59’’. Transcription of this recording had been slow, extending over several field trips whose main focus was on other topics (study of tone, and creation of a dictionary): I had transcribed 2 minutes in August 2014, 1’20 in June 2015, and 17’ in May 2016. The last 8 minutes were transcribed in May 2017 (completion : May 9th 2017).

I also transcribed 3’30’’ of BENEVOLENCE. So on this field trip, it’s a total of 9’30’’ using your transcriptions, plus 3’30’’ done before your transcriptions came in. That’s 13 minutes of verified transcriptions, in about 12 hours of work sessions devoted to transcription. Seen from this point of view, the difference with earlier methods is not huge. But it is hard (in fact: impossible) for me to quantify the acceleration gained by using the automatic transcription: as the years go by I am more and more familiar with the language, and the consultant is ever better at explaining, and repeating to confirm the exact pronunciation. This makes it hard to compare transcription speed. Also, about half of transcription sessions is spent discussing with the consultant about all sorts of topics, adding new example sentences to the dictionary, and writing comments of all sorts (the lines that start with % in the transcription document, later converted to a XML <NOTE> tag). What matters is the high quality of the data, and I find that the quality of the HOUSEBUILDING2 transcription is high. So I am, by now, a fully convinced user.

  1. Comments on recognition rate and on errors

The recognition rate is (in my impression) spectacular. Technical successes include the treatment of overlong vowels: phonologically (=for phonemic transcription purposes), lengthening needs to be overlooked, and the software overlooks it. The 500 ms long ɤ in /dʑɤ/ (S1 of BENEVOLENCE) is well recognized, for instance.

Occasional cases where the output is not phonologically well-formed (a syllable without a tone, for instance, or sequences such as kgv̩ in S15 of BENEVOLENCE) are easy for the linguist to spot in the transcription. This did not prove any trouble.

Some phonetic and phonological facts excuse several of the algorithm’s « mistakes » :

  • differences between u in your notation and o in mine (or the other way round) should not count as errors, as there are almost no contrasts in Yongning Na (and hence instances of /u/ are few). Phonetically, the transcription of [o] and [u] sounds by your system is good. A higher-level filter operating on syllables could adjust to ‘my’ phonemic analysis with 100% confidence: all instances of /jo/ and /ju/ and /ʝo/ and /ʝu/ in ‘your’ output can be translated to /jo/ in ‘my’ system, as the other combinations don’t exist (under ‘my’ phonemic analysis). Differences between M and H tones in initial and final position inside tone groups should not count as mis-categorizations, either, since the contrasts between H and M are neutralized in tone-group-initial position, and neutralized in tone-group-final position following a L tone. (Sorry if this sounds awfully complex; in fact, this neutralization rule is not so hard.)
  • There are a few loanwords in this text, such as « Taiwan » and « Japan », borrowed through Southwestern Mandarin. Loanwords contain a few otherwise unattested combinations : « Japan », /ʐɯ˩pe˧/, has a /pe/ syllable otherwise unattested in Yongning Na, which only has /pi/, realized as [pe]~[pi]. The automatic transcription, / pi /, is acoustically fine.
  • ʐ instead of ʈʂʰ, in S9 : this is a phonetically unsurprising confusion. In the training corpus, /ʈʂʰ/ appears all over the place in the demonstrative (also serving as 3rd person pronoun), a morpheme in which it is strongly hypoarticulated, losing both aspiration and affrication. This is an artefact of the training corpus.

To sum up, errors are few, and they make linguistic sense : they are not (as I feared) random added noise for the transcriber.

3. Benefits for the linguist

  • A benefit which I had not expected is that correcting the transcription allows me to be more alert : in a sense, I am in « proofreading mode », instead of when I type the transcription from scratch. In earlier transcriptions (« pre-mam days »), I made errors due to mis-typing : for instance, I clearly heard a L.L.H.L tone sequence, but somehow typed it as L.L.L.L. Mis-typing is pretty frequent. I keep track of all corrections to texts over the years, so I know about those mistakes. They can be really hard to correct: during fieldwork I make efforts to produce the most accurate possible transcription, in consultation with the language consultant, so I don’t want to change transcriptions later just on the basis of auditory impression, and I must trust the transcriptions. So one mis-typed symbol becomes a problem, if the issue is not spotted at once. If I never realize that something is wrong, it can create trouble later for other people using the data (worst scenario). If I realize later that something sounds wrong, I make a note, and ask the consultant on the next field trip : that takes up some of the (scarce) time of the field trips, and pretty soon (in a few years’ time ?) the consultant won’t be available anymore, at which point I shall simply cease to make corrections for want of being able to double-check with her. So precision of transcription is essential in « my » workflow. Errors can creep in anywhere, anytime, but I feel that the process of seeing the transcription in front of my eyes and correcting it allows me to correct the mistakes more systematically: in a sense, the task is already a « proofreading » one, so (I think) my attention is sharper and the result is better.
  • A benefit that belongs in the same category as those noted in the 2014 paper: having my attention drawn to phonetic facts that I have learnt to overlook because they are not phonemically contrastive. ‘Weak’ words (grammatical morphemes) undergo influence from their environment and get hypo-articulated. For instance, the negation gets transcribed as mõ˧ in HOUSEBUILDING2 (23m50s) instead of mɤ˧. This is interesting for me: now that I think of it, the vowel in that syllable is probably nasalized, and acoustically unlike the average ɤ vowel of ‘full’ words (lexical words). This is the stuff that phonetic evolutions are made of, explaining why and how grammatical morphemes follow phonetic evolutionary paths that are different from the rest of the lexicon.
  • The automatically generated transcription removes the risk of overlooking a repetition of a word. In casual oral speech, there are repetitions, hesitations etc, which hearers normally process in a ‘nonlinear’ way, factoring out dysfluencies etc. As I am committed to transcribing everything on the recording (not producing a ‘groomed’ text with the audio as a mere prompt to the consultant), I need to make efforts to ‘linearize’ the recording in the transcription: for instance, if a word is repeated five times, in “pre-Oliver Adams times” I used to pause, count the repetitions on my fingers, and carefully input all of those. This requires attention, and sounds a little ridiculous to the consultant, as she feels (rightly, I must say!) that this hyperbolic emphasis on form detracts from the “core” linguistic work in which she participates: explaining the meaning to me, explaining difficult words, etc. Now, with the automatically generated transcription, repetitions are all “linearized” in the first place, and I need to devote less attention to producing the transcription. Less attention devoted to typing letters and checking that I have not forgotten a word (or a longer passage) makes me more confident about the progression along the narrative, and hence more ‘available’ for the consultant for meaningful discussions about this and that. I less frequently need to tell the consultant “Wait for me a minute” while I go forward / backward in my notes. A couple of times I lost track of where I was in the automatically generated transcription, but thanks to the indication of time provided inside the automatically generated transcription it was not hard to get back on track.
  1. Ideas for future refinement
  • Add an indication where the algorithm is ‘skating on thin ice’ for recognition : such as when several candidates are about equally likely. This could for instance be added as a tag inside the transcription, and, in turn, be displayed with a special colour in an editor (an editing interface for the linguist). So when the linguist/transcriber looks at a full screen of automatically generated transcription, the passages that are unproblematic would be in green, and those where « confidence » is lowest would be in orange. This is not indispensible, of course : fussy people like myself want to produce « haute couture » pieces (linguistic annotation as a fine art !), and so they’ll want to check every word no matter how obvious it is that the automatic translation is already fine. On the other hand, if « your » tool is intended for rapid transcription of large volumes of data, it would be great to direct the user’s attention to places where verification seems particularly necessary.
  • A big piece of work : improving tone recognition. How is tone recognition conducted now : is it independently computed for each syllable ? In view of the importance of syntagmatic effects on tone in Na (=effect of the position of the tone inside the sequence of tones, and relative to the boundaries of tone groups), the model could be improved based on a « window model » relying on linguistic hypotheses on (i) possible sequences of phonological tones and (ii) expected phonetic implementation for these tone sequences. This would suppose arriving at a division into breath groups and smaller units (roughly corresponding to « sentences » and, inside sentences, to tone groups). If someone is interested to work with me on this, there are explanations and hypotheses in Chapter 8 of the book ; I think of these as a basis for future experimental phonetic studies, and dialogue with speech recognition would be super. Currently, tone recognition is not bad at all, but there is a lot of room for improvement. Given the nature of the tone system, I think that approaching 100% would be much, much easier than in languages such as Mandarin and many (most ?) other Chinese dialects.
  • A phonetician’s question : Is there any way to look under the hood and see which acoustic properties are used to identify the various phonemes? That is, the (relatively) stable properties that allow for the successful identification of a given consonant or vowel. I understand that the statistical models are not based on properties that can be straightforwardly identified with acoustic cues ; and I guess that if retrieving acoustic ‘features’ from recognition models were easy to do, it would have been done a long time ago (for the languages for which good recognition systems exist). But your achievements raise my expectations about what is possible !

In the coming months (after your PhD defense), I hope you will be willing to continue the dialogue and continue improving the system. If there were a way for me to ‘feed’ new transcriptions into the system to improve it, or to ‘re-train’ it anew with improved versions of the transcriptions, that would be terrific. The system is currently good enough to be useful ; I (for one) am interested to improve it further, to the point where Yongning Na has a good recognition system for that one speaker, and perhaps even a translation module – now is a time to dream ! 🙂

  1. Beyond consultant F4 and myself : Expectations of the Pangloss Collection ‘community’

Receiving the high-quality automatic transcriptions was extremely exciting for me. I talked about this to Guillaume Jacques & his students Lai Yunfan and Gong Xun. They are revolutionizing Sino-Tibetan studies by their work on Rgyalrongic languages (a group of dialects/languages that are as important to Sino-Tibetan as Sanskrit to Indo-European). Guillaume’s one-line answer was « C’est vraiment génial, si ce monsieur a du temps à consacrer au japhug, je lui saute dessus. » :: “This is great. If this gentleman has some time to devote to Japhug, I shall pounce on him.” You may remember my mentioning Japhug Rgyalrong in our earlier correspondence : it’s the language on which Guillaume does the most work. He transcribes reams of recordings by himself: he is so advanced that occasional verifications with his consultant are enough. Your tool could greatly accelerate his work. You could build on the availability of a much larger amount of transcribed audio than for Yongning Na. Guillaume does some transcriptions every day. Given his role as a leader in the field, and the importance of the language, the availability of a tool akin to the one you did for Na could do wonders. The Japhug corpus could also be a fantastic ‘playground’ for work on automatic translation. (For tasks beyond phonemic transcription: Guillaume wrote a dictionary of 7,000 entries.)

Of course everything takes time; doing a Japhug recognition tool would require understanding how he works and a couple of special conventions he uses. He is less finicky than me concerning phonetic detail, so I think he sometimes just leaves aside repetitions, hesitations and dysfluencies (whereas I try to transcribe everything), so perhaps that could make the training procedure slightly harder (due to occasional mismatch between transcribed string and audio signal).

Other colleagues, such as Alex François, also have large data sets that could be appropriate (dictionary+abundant texts). Guillaume and Alex are both interested to invest the time to make the most of brilliant technology like the one you developed for Na.

Let me stop here for now ; you can sense that I don’t feel as if this were the end of the story. Félicitations !

Yunnan, May 12th, 2017


Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

“Tone in Yongning Na” now published!

Book cover: Tone in Yongning Na

The volume Tone in Yongning Na is now published.

It is open-access: the books is freely available as a PDF from the publisher’s website.

It is open-source: the LaTeX code is freely available on GitHub (from the same page as the PDF).

Last but not least, the analyses are based on data hosted in an open archive (the Pangloss Collection) that allows free download of original sound files and transcriptions (no login required).

I think I should now let the book speak for itself 🙂 and thank once again the many colleagues who helped all along the way.

One of the aims of this blog post is to spread the word about the publishing house. Publishing a book at Language Science Press was a sparkling experience all along. If you do not know about Language Science Press yet, I strongly encourage you to read the description of the “Motivation”, “Setup” and “Vision” of this wonderfully forward-looking publisher. And if you like the project, you can help with the ongoing fundraising drive of Language Science Press (described here) by talking to your librarian about financial support to the publisher. (Feel free to get in touch with me if you would like to discuss this topic.)

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Project completed! Looking back at the past 48 months, and forward to online grammars and more connected data

The project completion date is December 31st, 2016. This is the time to look back at the past 48 months, and forward to more and more connected data: online grammars with one-click links to data, and other improvements for the greater benefit of linguistic research.

The aim of this project was to build several types of corpora (traditional stories and dictionaries) for unwritten and endangered Sino-Tibetan languages spoken in Nepal and in Tibetan areas of China (Khaling, Thulung, Koyi, Japhug, Zbu, Khroskyabs, Na), as well as to produce tools to analyze and format these corpora. We were able to develop a dictionary interface on the Pangloss collection website (http://lacito.vjf.cnrs.fr/pangloss/dictionaries/), and an online tool of parallel corpus comparison (http://himalco.huma-num.fr/corpus/comparable/index.htm and  http://lacito.vjf.cnrs.fr/pangloss/dictionaries/). The Khaling, Thulung, Japhug and Na corpora of the Pangloss collection were also significantly enriched.

The software engineer Céline Buret collaborated with us during 24 months. She developed a python library called pylmflib (https://pypi.python.org/pypi/pylmflib/1.0) which implements in XML the lexicographical norm LMF, with conversion tools from the MDF format (used by many fieldwork linguists to make their dictionaries) to LaTeX, HTML and DOC. PDF versions of the dictionaries (available on the website http://himalco.huma-num.fr/dictionaries/index.htm and on HAL-SHS) have been generated using LaTeX, and a paper version of the Khaling dictionary has been derived from it.

In the wake of the HimalCo project, new tools are being developed for the pylmflib library (conversion of the LMF dictionaries to Android format, graphical interface) to allow the production of multimedia dictionaries in other languages studied by linguists from CNRS and other institutions, in France and abroad.

For the Khaling dictionary, a set of PERL scripts have been written to (i) automatically generate the conjugation paradigms (ii) automatically convert from API to Devanagari script.

To sum up, this project has made possible to realize three types of corpora :

  • Three dictionaries of unwritten and endangered languages: Khaling, Na and Japhug. In the case of Japhug and Na, these dictionaries are the first to be available for these languages. As for Khaling, it is the first dictionary in a reliable transcription, which provides in addition the complete conjugation paradigms, of crucial importance in this language.
  • A corpus of 64 hours of audio recordings (and in part video) transcribed and annoted in four languages (Japhug, Na, Khaling, Thulung) has been archived in the Pangloss collection, which is thus significally enriched by 555 new documents.
  • A pilot interface of parallel corpus comparator, using some of the traditional stories from the corpora of Kiranti languages.

Scientific publications based on these data can be looked up here (full-text, no login required).

We now look forward to further experiments and realizations in the publication and exploitation of the results of linguistic analysis.  The idea is to put together a complete set of linked electronic resources comprising grammars, dictionaries, text corpora, and multimedia documents. This work gradually consolidates the empirical basis of the formulations and models proposed in typology and language theory. We hope you will want to continue to follow us, now on other blogs and websites (see in particular Panchronica — in French — and of course the Pangloss Collection).

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

“Tone in Yongning Na” is the first book to enter Open Review stage at Language Science Press!

The HimalCo project allowed for the completion of the linguist’s “three treasures” for the Yongning Na language: a dictionary; a collection of texts; and a monograph. This monograph was submitted in June 2016 to Language Science Press.

Language Science Press publishes high quality, peer-reviewed open-access books in linguistics. All publications are free for both authors and readers. General Editors are Stefan Müller (FU Berlin) and Martin Haspelmath (MPI for the Science of Human History). They are supported by a high-profile Advisory Board.

The book, entitled “Tone in Yongning Na: Lexical tones and morphotonology”, was accepted in October 2016. Anonymous reviewers provided a wealth of comments and suggestions, which were gratefully implemented. Now “Tone in Yongning Na” is the first book to enter Open Review stage at Language Science Press. I hope you will want to have a look!

Two quick notes for readers:
(i) The draft that is currently open for you to review has a few unsightly page breaks and line breaks: for instance, the word “construction” begins at the end of page 144 (“construc-“) and the ending (“tion”) is orphaned on page 147. Some examples are also split over two pages. And a few lines spill into the margins. Please kindly disregard these issues concerning page breaks and line breaks: we’re aware of those, but page breaks and line breaks will only be finalized at the very end of the typesetting stage.
(ii) It seems that the PDF made available on the reviewing interface, PaperHive, is not searchable. It makes it hard to navigate the volume. In case you struggle with this, you can download the (searchable) PDF at the following address:
You can then EITHER provide your comments by logging in to PaperHive and locating the places you have comments about, OR send me comments by e-mail (alexis.michaud@cnrs.fr). All comments will be gratefully acknowledged whether transmitted through PaperHive or by e-mail.

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

8 months to completion: quick progress note

This is a {quick [progress note]}
(i.e. a quick note about the project’s progress)
rather than a {[quick progress] note}
(i.e. a note reporting quick progress in the project)! 🙂

Dictionaries are big pieces, which raise many (most?) issues that linguists can confront when tagging and encoding data.

The dictionaries are currently here:

To come in the next 8 months: an Android interface for the dictionaries; improvements in parallel texts; and stable embedding of the dictionaries in the interface of the Pangloss Collection. Plans for later include full lemmatization of texts; phoneme-level alignment of transcriptions of the Na texts; and ‘crowdsourcing’ for the improvement of the dictionaries and collections of texts: allowing easy deposit of new resources and offering tools for a widened range of contributors (including the communities of speakers) to enrich the annotation through added translations, comments, and feedback generally.

The project’s funding was entirely consumed by the end of last year (2015), as originally planned. The ANR agreed to allow us to extend the project’s duration for 1 year, i.e. until December 2016 (without extra funding). The work now continues thanks to other sources of funding: the “Empirical Foundations of Linguistics” project (funded by ANR-CGI from 2011 to 2019); financial help from the participants’ home laboratories; and when necessary, our own money (e.g. for my next field trip to Yongning, in May 2016, which will help finalize a book-length study of the tone system of Yongning Na).

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Phonetics in the Field: topics of data acquisition, e.g. means to avoid reverberation when recording audio, and use of electroglottography

How to avoid reverberation when recording
How to avoid reverberation when recording

To study less-documented languages where they are spoken, know-how of various kinds is required, including the essentials of sound recording (along with some knowledge of botany and medicine, if possible). From a technical point of view, if one is going to spend weeks and months transcribing and studying a recording that will remain as one of the only traces of a language, it is well worth taking a few precautions to achieve good audio quality. Whenever possible, I pad up my workplace with quits, to avoid having too much reverberation. The result is acoustically OK, though visually unattractive! (Follow this link to get to the recordings) The cardboard boxes piled up in the room contribute to covering up the hard, smooth surfaces (cement walls and floorboards).

Placing the electroglottographic collar
Placing the electroglottographic collar

Ms. Latami Dashilame had not participated in linguistic fieldwork before she began to teach me her language. As we became more familiar, she willingly accepted to take part in such unfamiliar tasks as phonological investigations and phonetic studies, using somewhat impressive (but fully innocuous) techniques including electroglottography: a measurement of vocal fold contact during speech. (Click here to access a web page presenting electroglottography, including some scripts for analysis of EGG signals.)

The traditional Na costume, which she had put on that day for the visit of a journalist who wanted to see us at work, includes an elaborate head-dress which is incompatible with a head-worn microphone. On the other hand, it combines easily with the collar of the electroglottograph.

Linguists who do fieldwork have ample opportunity to realize the paramount importance of mutual understanding between the investigator and the consultant. I think that ‘laboratory phoneticians’ stand to gain by learning from fieldworkers’ experience of the data collection process. This is one of the beliefs that inspired a joint paper with Oliver Niebuhr about “Speech Data Acquisition: The Underestimated Challenge“.

To me, growing acquaintance with my teacher Ms. Latami Dashilame is a wonderful learning experience. In her childhood, she was one of the actors in a documentary-fiction film about the Na and their unusual family structure (‘A-zhu’ marriage among the Naxi of Yongning《永宁纳西族的阿注婚》). Later, her son became an anthropologist, specializing in Na (Moso) society, and a number of his colleagues came to her home. She witnessed how the Na of Yongning became an object of curiosity and fantasy, and how Na culture became folklorized for the promotion of the tourist industry, with its pleasant sides and its less lovely ones. These experiences shook her belief in some of the traditional beliefs that had been passed on to her by her elders, such as Buddhist beliefs. While she conscientiously goes through all the prescribed rituals on a day-to-day basis, her own belief in Buddhist teachings such as reincarnation was faltering. The narratives that she recorded show how she suspends her belief in the traditional lore of a waning cultural universe which nevertheless remains hers.

Photos: Qin Qing (秦晴), 2012

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Thulung in Paris, followed by a visit to Nepal

Following the earthquakes in Nepal this spring, it was difficult to get the authorizations to travel to Nepal on official French government business, and so the process was begun to try to bring Chandra Kala Thulung to Paris.

While French citizens can buy visas upon arrival in Nepal, the reverse is not true, and thus began a long process of attempting to secure a Schengen visa for Chandra. This process involved many months of waiting for visa-related documents sent to Nepal to arrive (they never did) and eventually resorting to getting more local help: CNRS has a network of CNRS offices around the world, tasked with facilitating international cooperation. The Delhi office came to the rescue, agreeing to receive Chandra’s visa documents and hand them directly to the French consulate in Delhi (there is no longer a French consulate in Nepal, and visa applications must be sent to India). The French scientific consular officer in Delhi signed off on the visa request, vastly speeding up the process. The result was that Chandra’s visa was ready in Delhi, one week before her planned departure, the “only” remaining issue being how to get it back to Kathmandu in time (thus bringing us back to our original problem which was circumventing a not-very-reliable Nepali postal service). The CNRS office in Delhi tried to use DHL (which apparently does not service “remote areas” such as Kathmandu!) and eventually found a local carrier. There are no street addresses in Nepal, adding another layer of complication, but Chandra eventually got her passport and visa, 36 hours before her flight!

Once she arrived in Paris, we got to work quickly and efficiently, and the three weeks sped by, with many epiphanies about Thulung for me along the way: among them, clearly (for the first time) contrastive vowel length, and a fairly clear understanding, finally, of where to look for the applicative -t suffix which is such a fundamental feature of Kiranti verbal morphology but is much more elusive in Thulung (because of the merging in many cases of CVC and CVCt transitive verb types as well as the much more productive use than in other languages of an analytical causative).

I then followed Chandra to Nepal, where we continued work on the verb dictionary we are compiling (480 verbs so far, with clear verb category assignation and recorded examples for each verb).

The second week of my trip, I worked on Khaling ideophones in view of a presentation at the Himalayan Languages Symposium scheduled for November 26-28th in Kathmandu. We have yet to see whether the symposium will take place: Nepal is in a tough spot these days, with a politically-motivated blockade at the Indian border resulting in a fuel crisis bringing the country to its knees and making transportation and cooking very difficult, as supplies dwindle.

My Khaling hosts spent my last afternoon in Nepal building a wood-fire stove on their roof, ensuring that whatever happened they would still be able to cook meals. The construction was a cheerful event, with lots of laughing interspersed with discussions about the best design for a stove and how someone should take a picture and send it to (Indian PM) Modi to show how hard he was making Nepalis work.   Yet again, Nepalis face adversity in good cheer and with their usual resourcefulness.


photo 1

photo 2

photo 3

Au revoir to Céline Buret — with all our thanks!

Céline Buret brilliantly completed her two-year contract with CNRS as the main engineer in the HimalCo project. This post is to say a BIG Thank You and wish her all the best in the next steps in her career!

The Python library that Céline developed to implement the LMF ISO standard for dictionaries is now available online. For an example of what the XML code looks like, check out the Na dictionary in LMF format (machine-readable). This file is available for download from the same page as the (human-readable) PDF version. This deposit also provides the full database in its original format: as a TEXT file that can be opened with the Toolbox software (SIL MDF format).

Further development of the tools is being taken up by her colleagues Séverine Guillaume (permanent member at CNRS) and Rémy Bonnet (on a contract funded by LabEx “Empirical Foundations of Linguistics”).

Merci Céline !!!

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Nepal News


Like anyone with a strong attachment to Nepal–and many more besides–we were heartbroken to witness from afar the two major earthquakes which devastated Nepal on April 25th and May 12th. Thankfully, the individuals we work with most closely have all reported that they are safe. While there was serious damage in Solu district, where the Rai languages are spoken, it seems that the number of casualties was not as high as was initially feared.
The French government is for the time being not allowing French civil servants (a status covering French academics) to travel to Nepal. As a result, we are working to bring a Thulung speaker to Paris during the summer, which will allow us to continue to work on these amazing languages while giving Nepal time to recover from its wounds.  We still hope to return to Nepal in the fall–Nepal has expressed that the best way to help is to travel to the country, as it relies so much on tourism–and will find ways to spread the word about these languages either way.

Nepal, February-March 2015: Khaling dictionary work



Guillaume and I recently returned from fieldwork in Nepal. The main goal for the trip was to finalize–to the best of our abilities–the Khaling verb dictionary that we have been working on.

Verbs are of course traditionally presented in dictionaries in their infinitive form, a form which–for Kiranti language–is insufficiently informative morphophonologically to make it possible to conjugate verbs.

The Khaling dictionary is exclusively verbal, verbs being listed by their infinitive but also by their roots, allowing users to recover the information necessary to assign them to a verb class. The dictionary is trilingual–Khaling, Nepali, English–and includes multiple culturally relevant trilingual example sentences for each verb, along with recordings (accessible in the online version) of infinitive forms, relevant conjugated forms and examples. The dictionary will also contain verb conjugation tables, enabling users to generate all the forms of the verb in question (generating these verb forms is no simple task, considering the possible number of verb stems and agreement markers which index agent and patient. See our paper cited below). Ultimately, the dictionary will be accessible through an android app, as cell phones are ubiquitous in villages (and more convenient in some ways than paper dictionaries).

The trip was thus spent poring over definitions, translations, example sentences, as well as finding new verbs which had not yet been included. The work was exhausting but also enriching–the examples were rife with interesting data on valence and argument structure–and we shared many laughs over Dhana’s love of alliterative examples [such as phoŋkantsha po:lu pɵnɛ plôntɛsi, ‘Uncle readied himself for basket weaving’]and Dhan Maya’s favorite themes for examples (oranges and buffalos).


Another highlight of the trip was a meeting of the Khaling cultural association–presided over by our friend chairperson Dhan Bahadur Khaling (Dhana)–during which Guillaume gave a speech (in Khaling) exhorting the community members to advance the cause of their language, among other things by putting together a wikipedia page on and in Khaling.


It is interesting to note that Guillaume’s was the only speech in Khaling, despite the majority of the audience being speakers.

We were very moved to receive phengas, the traditional Khaling vest, at the assembly.  Below is a photo of some phengas.


We will return next February to work on compound verbs in Khaling–another important part of the verbal morphology.


Jacques, Guillaume, Lahaussois, Aimée, Michailovsky, Boyd, and Rai, Dhan Bahadur, 2012. “An overview of Khaling verbal morphology.” Language and Linguistics LL13.6

Endangered Languages and New Technologies

The book “Endangered Languages and New Technologies” edited by Mari C. Jones has just been published at Cambridge University Press:
New technologies can help the documentation and analysis of endangered languages through digitalisation, archiving, the creation of smart electronic resources, and by bringing new methodologies to this linguistic domain.
This book presents a range of exciting realizations and new projects from field linguists, typologists and computational linguists, to handle documentation and analysis of endangered languages in a smart and connected way.

The first part of the book focuses on the development of new technologies for endangered languages. This part opens with a paper written by Aimée Lahaussois about the application of the concept of comparable corpora to closely related endangered languages (“The Kiranti comparable corpus: A prototype corpus for the comparison of Kiranti languages and mythology”). The same stories were collected in the field (eastern Nepal) in three languages (Khaling, Thulung and Koyi) and aligned on the basis of similar semantic content. For a technical explanation of the structure behind the corpus, see the following article, written by Aimée Lahaussois and Séverine Guillaume (LACITO), “A viewing and processing tool for the analysis of a comparable corpus of Kiranti mythology”:

The material in the corpus and the corpus interface will be expanded through the HimalCo project.

Various other projects are described in the articles that make up this part, on topics as various as European dialect syntax, cross-platform keyboard layouts, machine translation, online data analysis through Collaborative Data Management System, Learning Management Systems, etc.

The second part of the book is about methodologies and how to apply these to technology: digital curation through semantic web, long-term preservation with the CINES, sound alignment software, social media, sharing platforms. The articles in this section also discuss new technologies in education and fieldwork, and revitalisation of endangered languages into communities.

Digital dictionaries are also mentioned, especially the issues of standardization and interoperability. This topic is of great concern to us with HimalCo project where we are developing dictionaries based on the LMF ISO standard. This standard provides a model for the description of online dictionaries:
This standard also defines an XML format to represent these lexical data. The goal is to publish online dictionaries in an exchangeable and parsable format.

Another area of interest for us with HimalCo is the development of a smartphone application, which would enable minority speakers to access and propose modifications for online corpora.

Fieldwork in Nepal, Oct-Nov 2014



I have just returned from another very successful field trip to Nepal.  The frequency of trips–3 a year thanks to the ANR grant–contributes to the efficiency of data collection, as it provides continuity from one visit to the next, not only for me but also for the speakers involved.

Because I work on three different languages–namely Khaling, Thulung and Koyi–one issue is to balance the time spent on each one.  Koyi has proven somewhat challenging, as the number of speakers is much lower than for the other two languages, and even then, I have only been able to work with consultants whom I would label semi-speakers.   Yet they are semi-speakers in an odd way: limited numbers of nouns, but no problem (seemingly) conjugating verbs, despite the complex verbal morphology we find in these languages.  As for Thulung, last year I met a delightful and very good consultant I have been working with since, making great progress resolving some lingering mysteries, such as reflexive derivational morphemes -si and -s (presented at Syntax of the World’s Languages in Pavia in September).  Work on Khaling has been very satisfying because of the strong Khaling community that has gathered around our main consultant, Dhana Khaling, who is chairperson of the Khaling Rai association of Kathmandu.  As a result of his position in the community, and the fact that Guillaume and I stay at his house when in Kathmandu, we are at the center of a very vibrant community: Khaling people, living in Kathmandu or visiting from the village, are forever stopping by for tea or for a cultural, social or ritual celebration, which provides a wonderful opportunity for us to interact with a great number of Khaling.

My most recent field trip was centered around collecting Khaling ideophones–in collecting mythological texts, we had come across a number of ideophonic prefixes, and wanted to learn more about them.  They proved a very entertaining topic to work on, in light of the constant stream of visitors: they are fun to elicit and produce, were starting points for a number of wide-ranging discussions, and we all had many good laughs while collecting what ended up being close to 400 ideophones!

The data needs considerably more analysis before being written up for publication, but ideophones in Khaling (as in other languages) revolve around sensory experiences, with sound being the primary source, but also covered experiences such as texture, color.  This topic fits in nicely with previous work–Guillaume and I published a paper recently (https://benjamins.com/#catalog/journals/sl.38.2.05jac/details) about an auditory demonstrative in Khaling, which clearly taps into this grammaticalization of sensory experience, but it also connects with work we have and are carrying out on verbal morphology, as many of these ideophones are adverbial in nature.

Our next trip will also be on Khaling, in February-March 2015, and we will take the opportunity to study the many compound verbs found in Khaling.

Fieldwork in Yongning (language: Na), August 2014

Here is a short report on a field trip to Yongning that took place this month. This time, three students of Dr. Yang Liquan 杨立权 (on the right in the picture) asked to come along to learn more about fieldwork. Not that there is anything I can add to the wealth of available publications on the topic of linguistic fieldwork (a wonderfully concise and informative introduction is Dixon’s “Field linguistics: a minor manual”), but “an ounce of practice…”

Meeting Naxi, Mosuo and Bai colleagues in Lijiang, before leaving for Yongning. Left: Latami Dashi; right: Yang Liquan
Meeting Mosuo, Naxi and Bai colleagues in Lijiang, before leaving for Yongning. Left: Latami Dashi; right: Yang Liquan

The students got to practise International Phonetic Alphabet on a language that has relatively complex phonetics. The set of useful symbols for Na includes ɭ ʈ ɖ ʑ ɕ ɻ ʁ ʂ ʐ ʋ ɿ ʅ ɳ ɲ ŋ, æ ɑ ə ɤ , nasal vowels, and symbols for tone. The different language backgrounds of the students (Na/Mosuo, Bai, Mandarin dialects, English…) led to an initial division of labour among themselves: one could pick voicing differences easily, another was able to recognize different sets of apicalized vowels… This allowed for lively ‘cross-tutoring’ sessions. The aim, of course, is for each trained linguist to be able to identify and learn the entire range of phonemic oppositions found in the language under investigation.

Now the students are back to Kunming, and it will be up to them to decide to what extent they are interested to launch into fieldwork of their own — to break new ground, to make discoveries about languages of Yunnan, and to establish a sizeable record before these go out of memory.

Here is a group picture on the last day of fieldwork.

Hosts and students, at the main consultant's home. Top to bottom and left to right: Yunnan Univ. students He Wenju, Ji Xi and A Hui; the main consultant, Latami Dashilame; and her daughter-in-law Guo Geiruo
Hosts and students, at the main consultant’s home. Top to bottom and left to right: Yunnan Univ. students He Wenju, Ji Xi and A Hui; the main consultant, Latami Dashilame; and her daughter-in-law Guo Geiruo

Concerning research on Na, the main highlights were:

(i) advancing from the current state of the Yongning Na word list towards a decent dictionary, in particular verifying the identification of plants and animals;

(ii) getting answers to many small questions that had come up since the last fieldwork 9 months ago; and

(iii) recording new data, this time with a 4-channel recorder that allows for the recording of an electroglottographic signal plus audio from a table microphone, a head-mounted microphone, and an extra head-mounted microphone for a person serving as the speaker (storyteller’s) respondent.

A recording session
Checking signals and adjusting input levels at the beginning of a  recording session

In this case, the respondent was the student A Hui, who is a native speaker of (another dialect of) Na, and has special interest in a language and oral literature that have not been fully passed down to people of her generation. Her discussions with the hostess extended well after the recording sessions were over.

Talk about real communication: discussions in the Na (Mosuo) language, outside the (home-made) recording booth
Discussions in the Na (Mosuo) language, outside the recording room

The weather was cool and the nights cold (rainy season), but there were occasional sunny intervals. Electricity is now available most of the time, and with the help of a voltage stabilizer, plus a UPS for short power outages, we were able to use electronic equipment without any glitches. Extra batteries carried us through the occasional power cut.

The ritual of ‘walking around the mountain’ took place during our stay, allowing us a half-day off from linguistic work, and a view of the magnificent scenery of Yongning, under a spectacular cloudy sky.

Hill top above the Yongning monastery, on the day of a ritual: walking around the mountain
Hill top above the Yongning monastery, on the day of a ritual: walking around the mountain

We are grateful to those who made this intensive fieldwork possible, and in particular to our hosts and consultants.

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Project currently developing on GitHub — out of public view

The project blog has gone quiet in the past 5 months, as texts and dictionaries are in the making in a (nonpublic) GitHub repository. Tools for data migration from Toolbox / MDF format to XML and LaTEX are now operational. Painstakingly collected linguistic data and innovative technology make a heady cocktail: we were excited to see the very earliest professional-looking draft versions of our dictionaries.

Aimée Lahaussois made a field trip to Nepal earlier this year; Guillaume Jacques is beginning one in Southwest China; and this post’s author will visit again Yongning (also in Southwest China) in August. Lots of tasks are under way. In GitHub jargon: hundreds of ‘Commits’ (updates of documents and scripts), 9 Wiki pages, (currently) 33 ‘Issues’ (tasks to be done)… and high hopes of having great things to showcase here in due course.

Alexis Michaud

I'm a linguist specialized in phonetics, focusing on un(der-)documented languages.

More Posts - Website

Follow Me:

Article on Associated Motion in Japhug Rgyalrong

The article Jacques (2013) has been recently published. While the HimalCo project has already produced several presentations at various conferences, this is the first published article of the project.

Like all the scholarly productions of our project, it is based on corpus data (on the Japhug language) that will eventually all be made available on the Pangloss website. The article has two aims: first, to describe the associated motion construction in Japhug Rgyalrong and second, to use these data to contribute to the discussion on cross-category harmony (the idea that word order and affix order are correlated in some way). Since the second part might only interest theoretical linguists, I will now offer some reflections on the first part.

Languages with associated motion have a set of verbal affixes that describe a motion event occurring before, during or after the action referred to by the verb. In addition, they also have a competing construction with a motion verb and its complement. Since none of the world’s major languages has associated motion, this category has been relatively neglected, though it does not appear to be particularly rare. In the article, I focus on the semantic differences between the two constructions and show that a syntactic test involving negation works only with the motion-verb construction, not with the associated-motion construction.

For further information on Associated Motion as a cross-linguistic category, I recommend the work of Antoine Guillaume (2006, 2009) who has extensively worked on Tacanan languages. He and I are planning to work together on the typology of Associated Motion.



Guillaume, Antoine (2006). La catégorie du “mouvement associé” en cavineña: apport à une typologie de l’encodage du mouvement et de la trajectoire. Bulletin de la société linguistique de Paris 101 (2): 415–436.

Guillaume, Antoine (2009). Les suffixes verbaux de mouvement associé en cavineña. Faits de Langues : Les Cahiers 1: 181–204.

Jacques, Guillaume 2013. Harmonization and disharmonization of affix ordering and basic word order. Linguistic Typology, 17 (2):187–215

Himalco – Parallel corpora in languages of the Greater Himalayas