When 400 People Share One Conversation: Omnio at the Global Pastoralist Gathering, Mongolia

Snapshot

Table 1. Deployment snapshot

Table 2. Results at a glance

Executive Summary

The Global Pastoralist Gathering brought more than 400 herders, government representatives, researchers, civil society organisations, youth delegates and Indigenous representatives to Mongolia. The programme ran in English and Mongolian.

For a meeting convened specifically to hear from pastoralist communities, that presents a structural problem. The knowledge the gathering existed to collect is held in local languages by people who, in many cases, speak nothing else. Conventional simultaneous interpretation can bridge four or five languages at realistic cost. Participants at this event actively used 38.

acAIberry deployed Omnio across both the rural field visits and the Ulaanbaatar plenary. The technical footprint in the main venue was a single audio connection to the house mixing console. Participants listened on their own phones, choosing a language at the point of use rather than being offered a fixed shortlist decided months earlier.

172 people used the platform across more than 40 sessions, receiving 78 hours of live interpretation in 38 languages. Organisers received a complete documentary record of every session, in the original language and in English, as a by-product of the interpretation itself.

The deployment was not without difficulty. Connectivity at the four field sites was unstable and live audio transmission was unreliable at the outset. Live text translation, which requires a fraction of the bandwidth, held under the same conditions and kept participants connected to the discussion while the team stabilised audio on site.

1.1 The organisers

The gathering was convened by the Ministry of Food, Agriculture and Light Industry (MOFALI) of Mongolia together with Green Gold MRRC. Mongolia is an apt host. Pastoralism is not a heritage activity there but a working part of the national economy, and the country’s rangelands face the same pressures being discussed on stage.

1.2 The event

The programme was deliberately split across two very different environments.

Field visits, 11 and 12 August

Delegations travelled to Bayankhongor, Arkhangai, Selenge and Bulgan to see rangeland management in practice and speak with local herders. These are working rural locations, not conference facilities.

Plenary sessions, 14 and 15 August

The gathering reconvened at the Blue Sky Hotel and Tower in Ulaanbaatar for formal sessions, with more than 40 sessions delivered across the programme.

That split matters to anyone evaluating this case study. A platform that performs in a hotel ballroom has proved comparatively little. The field days were the real test.

2.1 A room wider than its languages

The gathering drew participants from Asia, Africa, Europe and the Americas. Proceedings ran in English and Mongolian.

The arithmetic of conventional interpretation is unforgiving. Interpreters are costly and must be booked well in advance. Booths consume floor space. Channels and headsets have to be assigned to named languages weeks before anyone arrives, which requires organisers to predict which languages will walk through the door. Most events settle at four or five. Every language beyond that is a budget line that has to be defended.

Participants at this gathering used 38 languages.

2.2 Who gets excluded, and why it matters here

The languages that survive the shortlisting process are usually those in which international institutions already operate. The consequence is quiet but consistent. The participants most likely to spend a conference listening to a language they do not speak are frequently the ones the meeting was convened to hear.

At a pastoralist gathering this inverts the entire purpose. Expertise on drought response, grazing rotation, animal health and land tenure is built locally and transmitted in local languages. It does not arrive pre-translated. Flying in the people who hold that knowledge does not, on its own, bring the knowledge into the discussion. Without a language route into the room, the result is attendance rather than participation.

2.3 Constraints on any solution

Three practical constraints shaped what would work.

• Language demand could not be predicted. Asking delegates to declare a language months ahead would have produced an incomplete list.

• Any solution had to work in two environments, a fully equipped hotel and rural field sites with limited infrastructure.

• Delegates could not be asked to collect, carry, charge and return specialist hardware across four days and several hundred kilometres of travel.

The deployment was measured against four outcomes.

1. Participants who speak neither English nor Mongolian should be able to follow sessions as they happen, not afterwards.

2. Language coverage should reflect who actually attended rather than a shortlist fixed in advance.

3. Integration should not require venue modification, hardware distribution or dedicated interpretation infrastructure.

4. Organisers should finish the event holding a documented account of what was said.

4.1 Architecture

Omnio operates as an access layer sitting between a single audio source and an unlimited number of individual listeners.

Audio capture

An Omnio operator laptop was connected directly to the venue audio mixing console, taking the same feed that drives the room speakers. This is the whole venue-side installation. No booths were constructed, no microphones added, no headsets distributed and nothing introduced that a house sound engineer would need to accommodate.

Processing and distribution

The live feed is processed and distributed as parallel language streams. Each additional language is a stream rather than a physical channel, so the marginal cost of the 38th language is not materially different from that of the second.

Participant access

Participants used their own phones. They open Omnio, select the session, select a language and listen. Alongside the audio, each user received a live transcript in both the original language and their selected language.

Output to organisers

Every session generated an original audio recording, an original-language transcript and an English translated transcript, delivered to MOFALI and Green Gold MRRC.

Table 3. Deployment summary

4.2 Why dual output mattered

The decision to deliver audio and text together was made on accessibility grounds. It turned out to be the most consequential design choice of the deployment, for reasons covered in section 6.

4.3 On-site operation

acAIberry staffed the deployment on site across both the field and plenary components, operating the audio connection, supporting participants at the point of use and responding to conditions as they changed. This was an attended deployment rather than software handed over at the door, which is what made the field-site intervention described below possible within the event itself.

5.1 Reach

172 people used Omnio across more than 40 sessions, receiving 78 hours of live interpretation in 38 languages. Those languages included English, Mongolian, French, Spanish, Hungarian, Turkish, German, Mandarin, Japanese, Korean, Hindi and Arabic, alongside 26 others.

172 users against 400 or more attendees is worth reading correctly. A substantial share of participants worked comfortably in English or Mongolian and had no need for interpretation. The meaningful figure is not the percentage of the room that used the platform. It is that the portion of the room previously without any language route into the discussion had one.

5.2 Participant outcomes

Adnen Helali, Tunisia

Speaks Arabic and no working second language. Followed the event in Arabic throughout. He found Omnio very professional, very clear and quite accurate in translation. said that this is by far the best interpretation service he has ever used at any event.

Saijrah, China

Speaks only Mandarin, with no English. He is the case that matters most in this document, because he did not stop at listening.

In the standard version of his four days, Saijrah sits through the sessions, waits for a colleague with spare time and a shared language, and receives a summary of what he missed. Summaries arrive after the moment has passed. By the time he knows what was said, the discussion has moved on and the chance to respond is gone. Whatever he knows about herding in his own region stays in the room only as long as he is standing in it, and leaves when he does.

Instead he followed the sessions as they happened and took part in the discussion.

That is the line where the platform stops being an accessibility feature and becomes something else. Listening is one-directional. A delegate who can only receive is an audience member at a meeting convened to hear him. Saijrah crossed from receiving the conversation to being in it, in a room where nobody on stage was speaking his language, and he did it holding a phone rather than waiting for an interpreter to be assigned to Mandarin.

He came to find the team afterwards, at length, to thank them.

Pastoralist delegates, Turkey and Hungary

One speaks no language other than Turkish. The other does not speak English. Both followed proceedings in full. Both participated in the discussions, which was the greatest achievement of Omnio.

These are representative rather than exceptions. Participants approached Acaiberry team repeatedly across the four days, and the consistent theme was not the quality of the translation but the experience of taking part in the event rather than sitting through it.

5.3 Organiser outcomes

Organisers of an international gathering carry the language problem more directly than anyone else in the building. They are the ones who selected the working languages, who know exactly which delegates those languages leave out, and who field the quiet apologies afterwards.

Ms. Burmaa Dashbal, Executive Director, Green Gold MRRC

Both MOFALI and Green Gold MRRC reported that the service was effective, and that participants had come to them directly to say so.

Organisers also finished the gathering holding a complete session record: original audio, original transcript and English translation for more than 40 sessions. Under a conventional arrangement this is a separate workstream with its own budget and a turnaround measured in weeks.

A case study that reports only successes is of no use to anyone planning a similar deployment.

6.1 Connectivity at field sites

The problem

Internet connectivity at Bayankhongor, Arkhangai, Selenge and Bulgan was unstable. Continuous live audio is the most bandwidth-intensive function the platform performs, and audio transmission was not reliable at the start of the field programme.

What held

Live text translation continued to perform effectively under the same conditions, requiring far less bandwidth than streamed audio. Participants who would otherwise have lost the discussion entirely followed it as text in their own language.

Resolution

The team addressed audio transmission on site and stabilised it during the field visits rather than logging it as a lesson for afterwards. The plenary sessions in Ulaanbaatar, with venue connectivity and a direct console feed, did not run into the same difficulty. Omnio was very effective at the plenary sessions in Ulaanbaatar, and participants were able to use both audio and text services.

6.2 What we would do differently

• Position text as a primary output from the outset. In low-connectivity environments the transcript is not a supplement to the audio. It is the resilient channel, and participants should be briefed accordingly before they leave the hotel.

• Set expectations before signal loss rather than during it. A participant who has been told that audio may thin out and that the transcript will carry them experiences a mode change. A participant who has not been told experiences a failure. The technical reality is identical.

• Survey connectivity at field sites in advance where possible. Rural conditions are knowable ahead of time and should shape the operating plan rather than be discovered during it.

Language coverage stops being a forecasting exercise

The organisers did not select 38 languages. Participants did, individually, at the moment of use. Any coverage decision made months in advance will be wrong, and it will be wrong about the delegates who travelled furthest.

Integration is a single audio connection

If a venue has a mixing console, the infrastructure question is largely settled. This also means interpretation can be added late in planning, which conventional arrangements do not allow.

Design for the difficult venue

The ballroom will be fine. The plan should be built around the field day, the breakout room and the site visit, because those are the conditions that determine whether participants trust the service.

Documentation and interpretation are the same task

Processing speech into multiple languages produces a transcript as a by-product. Organisers who currently commission transcription separately are paying twice for one piece of work.

Access is a precondition, not an outcome

Interpretation cannot make a contribution more persuasive or guarantee it will be taken seriously. What it does is remove the barrier standing in front of every other question. A delegate who cannot follow a session cannot agree with it, object to it, correct an error in it or build on it.