Google Sheets — Excelling at Understanding
Natural Language Processing (NLP) has become part of many consumers’ home lives, and its use is set to increase even more. Assistants like Amazon Echo and Google Assistant are used for everything from setting timers or checking the weather to ordering food. However, until now, the use of NLP in software has largely remained the domain of developers familiar with Machine Learning and data scientists, using complex structured API calls for services like Rasa or DialogFlow, or developing bespoke models with lots of data and a large amount of trial and error.
Google is pushing the boundaries of NLP internally, as my last article highlighted, and they are also keen to see the technology adopted more widely by external developers, as their latest release, ‘Semantic Reactor’, demonstrates. Their accompanying blog post indicates that they understand that there is a large intimidation factor associated with using NLP in projects, and this release is designed to lessen that factor’s effect.
Semantic Reactor is an add-on for Google Sheets that allows a developer to use a spreadsheet to analyse pieces of text, and sort lines either according to their similarity to the input text or according to the appropriateness of the line as a response to the input. This allows developers to gain the benefits of a trained NLP model without categorising their training data into ‘intents’, which can often be limiting and unnatural for the training data, as well as a rather blunt tool for analysis. In addition, it allows developers to analyse the ‘similarity’ of the text in comparison to all of the training data, rather than just seeing the most similar match. This allows for nuance within the processing of natural language (something that Byte required for its Zegna chatbot for the ‘What Makes a Man’ campaign), and the ability to have software respond to suggestions further down the list of results based on complex business logic, rather than just the NLP model.
Google Sheets is undeniably ubiquitous — millions of companies use it every day for their spreadsheet needs, and as a result, there are already several tools and add-ons (both official and unofficial) that can be combined with Semantic Reactor to produce really exciting results. Moreover, its widespread use means that many software developers are already familiar with the Sheets API, likely leading to increased adoption rates relative to a brand new API release, which could require weeks of development to integrate into a new system. However, the somewhat rigid and simplistic structure of the Google Sheets interface means that this is unlikely to become widespread as a tool for production bots, and it is likely that developers will stick to APIs such as DialogFlow for these purposes. Nevertheless, Google should be commended for producing a tool for such a powerful technology with such a low barrier to entry.
As part of this release, and to overcome what it has identified as other barriers to NLP’s adoption, Google has also made available a ‘lite’ version of their Universal Sentence Encoder (the technology that converts natural language sentences to vectors, which underpins software like Semantic Reactor). This ‘lite’ version will allow developers to include it in client-side applications (whether on mobile devices or front-end websites), rather than requiring it to be run on a server, again reducing obstacles that might preclude a developer from exploring the possibilities that NLP might bring to their application.
Google admits that these smaller models and tools are far from perfect, and that will admittedly be off-putting to some developers. Nevertheless, given the myriad of advantages that NLP brings to user interfaces, the removal of the adoption barriers that Google has identified is to be welcomed. More robust alternatives remain available for more specific use cases, but being able to explore this technology within the familiar surroundings of Google Sheets should allow thousands of new developers to incorporate NLP into their existing setups.