We’re super excited at TfWL with all the start-ups we have taking part in cohort 2.

Due to COVID restrictions, cohort 2 has not been able to take place from our innovation Lab in Newport.

Consequently, cohort 2 has been taking place remotely.

We’re collaborating with out start-ups online, every week. COVID may have locked us down, but it has not knocked us out!

Cohort 2 is producing some fantastic ideas and solutions. We want to shine the spotlight on each start-up that is helping to innovate Transport for Wales Rail.

Today we’re introducing you to, Wordnerds.

Who are Wordnerds?

Wordnerds are a deep tech start-up doing exciting things in the world of Big Data and Artificial Intelligence (AI). Using cutting-edge AI, Deep Learning and Natural Language Processing alongside old-school corpus linguistics techniques, Wordnerds are teaching computers to read – and genuinely understand – language. 

Their software-as-a-Service linguistics engine gives brands the ability to automate the understanding of what their staff, customers and stakeholders are saying – at scale and irrespective of the vocabulary they use, or how liberally they sprinkle their text with emojis, sarcasm and irony. 

Their Background

Regulated industries have proved hugely successful for Wordnerds. Regulators mean industries strive for improvements, and Wordnerds have certainly found in the Utilities and Rail sector, organisations are looking to their customers to inform better CX strategies.  

Within the Rail sector, they’re working with a number of UK TOCs, System Operators and an industry watchdog to:  

  • Identify real-time passenger sentiment across Routes and Stations, supporting rail staff to make better data-led, customer-orientated decisions. 
  • Gain greater insights into industry wide standards, giving CX teams a holistic view of performance, benchmarking against industry-average standards and competitors. 
  • Identify the changes in passenger expectations towards travel as the UK moves out of lockdown, and inform strategies for increasing passenger confidence. 
  • Motivate rail companies to review customer feedback sources, from social channels, emails, surveys, live chat, complaints, that would otherwise remain ignored or underused.  

USP

The main difference between Wordnerds and the “social listening” platforms on the market is that they approach the problem from a linguistics-first – not a data science first – perspective.

Wordnerds have developed a first-of-a-kind contextual word embedding model that allows the user to train and fine tune their own artificial intelligence to find themes in the data that matter to them. This means users are able to surface common themes within their data, irrespective of vocabulary, sarcasm, misspellings or the liberal use of emojis.  

As a result, there are 3 ways in which Wordnerds stands out from its competitors;

  1. Smart filtering
  2. Topic clustering
  3. More accurate sentiment analysis

Foundations

In December 2015, CEO of Gateshead-based digital agency Daykin & Storey, Pete Daykin answered a 48-hour tech challenge from Nissan, facilitated by the Digital Catapult Northeast/Tees Valley and attended by over thirty agencies and software companies from around the world. 

Nissan wanted to explore whether it was possible to use social media to identify the tiny number of quality issues they experience in new cars. Of the 3,000 tweets an hour that mention Nissan or one of their brands, could they identify the one or two a day that might potentially indicate a problem on the production line? 

Pete recalled meeting a freakishly tall Scotsman, Steve, in a pub some months earlier. 

Steve was a linguist and had just finished a project at Newcastle University identifying the author of an unknown Victorian ghost story manuscript from the linguistic fingerprint of the writing. 

At the end of the first day, Pete rang Steve, explained the situation and asked a question. If Pete could use some new Artificial Intelligence techniques to identify and tag linguistic elements in tweets (verbs, nouns, prepositions etc.), could Steve work out a way of spotting problems from the structure of the language, not the vocabulary? 

After all, language is surprising. People say things like: “My gearbox is toast” or, “My suspension is goosed”, things that you wouldn’t find with a keyword search. 

Steve sat up all night, wrote 21 syntactic rules and provided three examples of things other software would miss but they could spot.  

The pair’s pitch to Nissan the following day went something like: “You asked us to look at your data: we didn’t. You wanted to see our software: we don’t have any. What we do have is a half-arsed idea of how we can combine really new AI with really old linguistics to solve your problem.” 

They won the challenge, received a small budget from Nissan to work on a proof of concept and set about the task. Wordnerds was born. 

Meet the Team

Pete Daykin – CEO and Co-Founder: https://www.linkedin.com/in/pete-daykin-2a11952/ 

Twitter – @peterdaykin 

Steve Erdal – Chief Scientific Officer and Co-Founder: https://www.linkedin.com/in/steveerdal/ 

Twitter – @StevieTheGiant 

Helen Precious – Big Data Consultant: https://www.linkedin.com/in/helen-precious-019a65b6/  

Twitter – @HelenPrecious 

Steph Clish – Operations Manager: https://www.linkedin.com/in/stephanie-clish-b1177a4a/ 

Twitter – @stellaclish 

Why was TfWL a good fit for Wordnerds?

“Innovation happens at the intersection of two disciplines. For us, that was Corpus Linguistics, and Natural Language Processing. These two disciplines have been on a certain trajectory – and to a certain extent you could argue everything has been achieved in those two areas already by someone. But by combining these two things, we’ve been able to solve problems that couldn’t have been solved on their own.  

Innovation programmes such as these keep us moving – in order to maintain our standing in a hugely competitive market place, it’s imperative we keep pushing to stay ahead of the curve, and at the top of the tech stack! 

We love working with like-minded people. We recognise, through our sponsors and product owners, we share an ambition to achieve exceptional insights from unstructured text. In turn, we are motivated to help the team at TfW develop and deliver an exceptional CX strategy – gathering insights across communication channels – enabling customers to communicate however they feel comfortable, irrespective of vocabulary, grammar, emojis or sarcasm.”  

What does TfWL Hope to Achieve Working With TfWL?

“In an ideal world, we hope to develop the relationships with our Programme Owners and Sponsors that outlast the TfWL programme. We recognise that we are not the experts here – and so for us, we hope to learn from those who are, to help strengthen our platform for this industry.  

This programme affords us the time and connections to really dive into the business challenges and processes that Mike and the team are facing, in order to understand how we can best support them in gathering customer insights.  

In the longer term, I hope we’ll be set up to generate a long-term working relationship, with a view to engaging in other innovative opportunities we can achieve with such vast, industry-specific data resources to pull from.”  

Follow Wordnerds

www.wordnerds.ai 

https://www.linkedin.com/company/wordnerds/

https://twitter.com/word_nerdy 

https://www.facebook.com/WordnerdsAI

https://www.instagram.com/wordnerds.ai/