Looking forward to EMNLP 2028: NLP conferences and climate change

EMNLP 2018 is underway, and I am in Brussels with many of my collaborators. There are over 2500 participants, making it the largest NLP conference ever, with scientists from all over the world. These are exciting times for the field of natural language processing.

But let’s look slightly further ahead: What will EMNLP look like in a decade?

The context for this question is the recent special report from the Intergovernmental Panel on Climate Change, which states the facts in plain terms: catastrophic climate change is inevitable; it will destabilize ecosystems and societies; and to stand even a chance of averting far worse outcomes, we must cut worldwide carbon emissions in half over the next decade—and then keep cutting them. To accomplish this, we will need to remake modern society at all levels, from global to individual.

Remaking society at a global level will require sustained, coordinated action across all sectors of human activity on a scale never before seen, and enormous changes to policy. Individual action is not enough, but individuals will need knowledge, experience, and the moral authority that comes from setting examples of positive alternatives. So, if you are an individual scientist who wants to contribute, what can you do to cut your own carbon emissions in half? For many of us, it would require a single behavioral change: stop flying, because air travel is our single biggest source of carbon emissions, often dwarfing all other sources. This is certainly true for me, and it is likely true for many scientists in my field. In short: flying to conferences and other professional events has many direct and indirect costs.

Conferences cost the scientific community in other ways. Independent scientists and those in the Global South often cannot attend due to cost. Some are prevented from attending by lengthy visa applications or denied visas. Scientists from underrepresented groups are harassed and assaulted. Women are more likely to face gendered expectations for family care that make it difficult for them to travel. In short, conferences are exclusive, and exclusivity keeps out those who might otherwise advance our field.1

So the cost of bigger and bigger NLP conferences is high. Let’s weigh that cost against the benefits. Why are conferences important for our field, or any scientific field?

  • In NLP and related fields—mainly in computing—they are the primary publication venues for scientific work.
  • They are a major venue for more informal dissemination of scientific work in the form of talks, posters, and hallway conversations.
  • They are shared spaces where we meet other scientists, generate new ideas, and start new collaborations.

Are these benefits only possible with conferences? Consider:

  • The conflation of publication and conference attendance is simply due to the historical accident that computer science came of age in the jet age. There are many other publication models, and academics love debating them so I won’t dwell on them here, except to say this: TACL ought to solve this problem, but it doesn’t. It won’t replace conference publication a long as it so stingy with acceptances—most people resort to conference submissions because TACL is unattainable. This deserves rethinking.
  • Talks are easily recorded, and indeed many recent NLP conferences have all of their talks online. Talks can and have been delivered remotely, but our conferences explicitly discourage this. They shouldn’t.
  • Paradoxically, the growth in the number of conferences and the number of researchers makes it less likely that you’ll actually meet the people in the field that you hope to talk to. Simply going to one conference every year or so means that you’ll miss meeting many of your fellow researchers who went to one of the other 3-5 big conferences this year. But you can confer with them year-round through any number of technologies, public (e.g. social media) or private (e.g. videoconference, chat, email, etc.).

Personally, I’m skeptical of the idea that conferences are vital for intellectual exchange: of the many meetings I’ve attended over the course of my career, the number that substantially influenced my research has been very small. All of them were small events attended by a few dozen people who collaborated intensively in small groups for a week or more. That kind of focus is exactly the opposite of what we find in today’s enormous conferences. And, having collaborated successfully with remote partners several times over the past few years, I’m skeptical that this requires colocation.

I’m sure it’s healthy for scientific communities to collectively pause, take stock, learn about recent discoveries, and interact spontaneously with each other. The technology already exists to enable all of these things without flying across the world, even if it isn’t cohesively packaged yet. Remote collaboration tools are not fully carbon neutral, but they are far less damaging than flying. Many climate scientists have already quit flying, and some academics are already experimenting with virtual conferences. Technologists have a role to play in mitigating climate change, and it is easy to imagine that we could iterate and scale virtual alternatives to conferences in just a few years.

Obviously, there are many other ways you can reduce your carbon footprint and contribute to the enormous task ahead. But I was trained as a computer scientist, and I know that if you want to lower costs, you start with the biggest cost first because that’s where the biggest gains are. My goal isn’t to minimize other things you can do, and it isn’t to tell you to just stop conferencing. It’s an invitation: if you believe that the scientific exchange that happens at conferences is important, how can you reimagine that exchange to make it much less damaging to the environment and far more inclusive than it is right now?

  1. Exclusivity was built into scientific conferences from the beginning and this needs to end. ↩︎