Noted - Slack, Messenger and AI Storytelling

Every two weeks, News Labbers come together for a “Family Day” to share project updates and other notes from their work. We don’t bring our relatives, but we do invite the wider family of people who are interested in driving innovation in News, rather tnan keeping it exclusively for News Labbers. During the 90 minutes we share what we have been doing for the past fortnight — whether it’s progress on a prototype, reflections from a recent conference or hackathon, or a day shadowing reporters in New Broadcasting House. We also collect news around journalism, technology and innovation to share with the team.

Here are three topics we had on the docket.

Slack: Smarter, Faster and More Open

The Times Digital Development team opened their internal Slack channel to the general public by creating a GitHub repository full of useful resources for journalists. We also took note of a new partnership between IBM Watson and Slack, which aims to make bots on the platform smarter using the company’s conversation and speech APIs. According to Liat Clark, reporting for Wired, IBM plans to share a Botkit Watson plugin that lets Watson Conversation communicate directly with Slack. An application Starter Kit for developers will also be made available.

One of the more intriguing examples of Slack automation was from an article in Digiday describing how Buzzfeed uses the service to push trending local news stories out to a global audience. It said that regional and national editors receive a notification on the desktop application when a local story passes a certain threshold for success. BuzzFeed head of European growth Luke Lewis is quoted as saying that editors can then “translate, adapt or iterate creatively on it” for a global audience.

The BBC covers a wide range of news across the country and the world. While editors can choose to look up analytics on individual stories, there is no internal notification system built into newsroom workflow that alerts editors when a story was doing particularly well on social media. Perhaps a future project for News Labs … ?

Messenger Bots: In or Out?

We don’t think the platform itself is out, but some of the bots we were most excited about in our #General slack channel over the last month have proven to be a bit less fun after a few weeks of use. We had a couple of die-hard New York Times Election bot followers who were pleased to see the organisation making use of its rich data on polls and surveys in a bot service. However, we quickly felt this feature wasn’t nearly as clever as the scripted updates coming straight from NYT political correspondent, Nicholas Confessore. By this Sunday, it seemed like Confessore himself was trying to get users to click on decidedly “non-bot” content — a clue to the fate of the project, from an analytics standpoint?

Some of us have also been following the Washington Post’s “Feels” bot, which asked users to share how the election was affecting their emotional well-being. This seemed like a good idea, but was the election making anyone feel anything other than exhaustion in the final stretch? From recent updates, the “here’s how you compare” graphics also seem to suggest that the users who are still submitting their feels may not representative of the population of the United States.

A graphic from the Washington Post Feels bot
How bot users said they felt the day after the U.S. presidential election

One alternative we noted was the election bot created by Canada’s Globe and Mail, “GloBot”. GloBot differed from its U.S. counterparts by not only delivering news around the election, but letting users see the candidates’ positions on policy points like the environment and the U.S. / Canada relationship. This content wasn’t updated, so it perhaps has limited utility for long-term use beyond a vehicle for pushing articles.

Augmented Reality, Virtual Reality and Automation

The BBC doesn’t have a print legacy, but it was still interesting to see announcements for Condé Naste China’s upcoming experiment with Vivepaper — a new augmented reality magazine experience that lets users interact with content using an HTC Vive headset or more standard cardboard device.

We also revisited the automated image tagging technology that Google released in September. Show and Tell is a “neural image caption generator” that uses TensorFlow to identify objects in images. The algorithm reportedly achieves over 93% accuracy. The examples on its README file don’t give us a strong sense as to whether this technology could be easily adapted for news or breaking news, but Google published a research paper on the topic that might give us a better idea. BBC R&D is developing similar technology from within the BBC on facial recognition and tagging.

Additional Linking

From the News Labs Slack Channel

  1. We were excited to see two audio transcription tools at Mozfest - AutoEdit and Trint. One of our developers found another work-in-progress, Adobe’s Project VoCo. We’ve been developing our own product called Transcriptor, which offers automated text-to-speech technology and caption creation. It’s interesting to see what other tools emerge to fill this space, and what technologies they make use of, as we continue to think of new ways to help BBC journalists find and edit audio and video footage more easily.
  2. We stumbled upon an open-source argument analysis platform called Argumen, which we found interesting as we think about ways to expand upon our work with chat bots. The platform lets users add branches to statements, explaining them with “because” or choosing to debate them further with clauses that start with “however” and “but”. Perhaps crowdsourced data could be a source for conversational logic for a project in the future.