KantanMT Embraces Change to Grow with the New Era of Machine Translation

Pic2

KantanMT recently launched a new interface. In this blog Laura Casanellas, Product Manager at KantanMT explores the reasons behind the change and talks about the new functionalities that have been added.

The seed for the change was planted a year ago. After joining KantanMT and familiarising myself with the product, we identified areas on our platform which could be updated and improved. The original platform was created five years ago and it was designed to fit a set of original features that, with time, had been expanded and modified.  It was time to re-focus on two of KantanMT’s core strengths: simplicity and ease of use.

At the end of last year, it was decided that the Development team was going to slow down the implementation of new features to allow time to focus on the re-design; “that does not mean that we completely stopped improving the product, as during the re-design process we have implemented a number of new features, including the availability to customize Neural Machine Translation”, mentions Marek Mazur, head of Development.

For me, it was a steep learning curve, but one I happily welcomed; as a new Product Manager, it was a great opportunity to take part on this process of growth and change.

The idea was not to deviate too much from what was already there from the beginning, as the original platform, designed by Tony O’Dowd, CEO and Chief Architect of KantanMT, already worked very well. Proof of it is that, in my time as Product Manager we have received very few comments regarding the UI; most of our client’s requests are of highly technical nature and relate to customisation of the platform for specific uses.

Prior to the implementation, we carried out several research projects; one of the most significant was a usability test performed in-house, which involved a number of new and experienced users interacting with the platform. “When you use something every day for some time, there are things that become a routine and you stop wondering if they are god or bad. That is why we decided to get test users spilt in two categories: novices and current platform users”, says Marek.

We conducted few rounds of testing, each of which included some changes in the UI, based on the feedback provided by the previous group, until we got to the final version, which is what we have today. We were aware that some parts of the platform needed to be streamlined, changed or improved and this was proven during the usability test, as most testers struggled with such areas, but we also had a couple of surprises, especially in relation to the way support information was displayed in the platform.

Usability test

Pic4

We triangulated the information gathered in the usability test with a user survey performed by our own users, and a perception questionnaire. Once all was done, we came up with the following conclusion: as the focus of the platform had widened since it was first created, some of the key terminology used needed to be replaced, especially for new users who were not familiar with the platform. Also, we streamlined the structure of the dashboard by reducing the number of tabs.

It came as a great surprise to realise that most testers did not find the Help section located at the button of the page straight way, neither they found the engine analytics area that easily. Given the fact that Machine Translation can be a complex enough process for a novice, we understood that critical resources like help and analytics needed to be more easily accessible in all areas of the platform.

This initial part was the most time consuming, but once the final design and specifications were ready, the actual coding took only a couple of months. And during the last three months, we focused on testing and adding new features to the platform. By the time of release, the KantanMT Professional Services team, our internal super users (who have been using the platform from almost its inception), together with the Development team, had been testing and using the new interface for over a month. We wanted to be absolutely certain of achieving major improvements before releasing our new UI to our clients.

Nearly every page is within two clicks

At KantanMT our research and innovation endeavours have led to some major new features – take KantanNeural or KantanFleet, for instance. In keeping with our cutting-edge technology behind the platform, we wanted the front end of the platform to look more modern and sleek. Riccardo Superbo from Professional Services reiterates, “I particularly like the layout, it is clean, simple, and modern, and it makes it easy for users to navigate through the different pages”.

The new change has brought faster performance and more economy of use. Dimitar Stherionov, head of KantanLabs, talks about a “much faster” interface, one where “the new visuals make it easier to know what is what and where to find what you need”. Pat Nagle, also member of KantanLabs, focuses on the ease of use, “the side menu makes navigation easier and faster to get around.  Nearly every page is within two clicks of the entry dashboard page”.

One of the structural improvements is the re-location of the Template function on the Training area. This change made it possible to reduce the number of tabs in the dashboard. Using templates helps our industrial users save time, especially when they have many engines; some of which might share part of the same training data.

Another important change that follows the idea of simplification, and one of my favourites, is the conversion of the previous pre- and post-processors (PEX) into one single Rule editor that can be accessed from the Training and the Translation areas, depending on the part of the content the user wants to modify: Training data (source, target or mono) or Translation data (source and / or target).

Help at eye level

Pic3.png

When I started working in KantanMT a year ago, one of the aspects of the product that most impressed me was the very comprehensive Help content that could be accessible from most of the platform areas. As KantanMT has been designed with the idea that users can create, optimise and deploy their own MT systems themselves, all features related to customisation and performance enhancement are available to the users in the platform. Somebody who uses a new feature for the first time would very possibly like to have the related Help at hand.

Through the usability test we realised that some testers could not find the Help easily, as they did not scroll down the relevant page. This came as a surprise to me, but given that information resources are important to our customers, we decided to increase the Help and information points and to include links at the top of every page. As a result, Help and additional information points are at eye level and, therefore, much more accessible than before. *

Another improvement related to information display is that all the active engine metrics are displayed within the user’s working area, below the dashboard.

Pic1

Open path to incremental NMT

KantanMT’s new interface comes with a fully functional incremental training feature for neural engines. In the words of Dimitar, “we exploit the iterative-training nature of Artificial Neural Networks and extended our platform with the ability to continue the training process of an already built KantanNeural™ engine. That is, a KantanNeural™ engine that was already built can be incrementally trained with new parallel data.”

One of the great benefits of the incremental retraining is the significant time saving that it implies. And with that, Neural MT becomes viable for industry production.

We are very happy to see this project come to fruition, as Marek says, “the new platform looks very modern, simple and it is easy to use. It was a great experience to work on the design and its implementation. In the last months, I could see how the whole team became passionate about the process”.

I agree with him, I think KantanMT is ready for the new space where the disruptive Neural technology has firmly establishing itself. We have already embraced it, adding it to our Machine Translation offer, allowing our customers the possibility to build their own Neural engines or to choose from  our more than twenty general and domain specific Neural Fleet engines.

* The Help will be significantly supported by over 20 how-to guides that will be available to our customers through our Support section soon.

About Laura Casanellas

Karl Hussey Photography 2017.Laura Casanellas is Product Manager at KantanMT. Laura brings to the team over fifteen years of experience in the Localization industry. During that time Laura worked in a variety of roles (Language Quality, Vendor Management, Content Management) and verticals (Games, Travel, IT, Automotive, Legal) and acquired extensive experience in all aspects related to Localization. Since 2011, Laura has specialised in Language Technology and Machine Translation. Her role in KantanMT involves defining new developments in collaboration with the Technical Team, Professional Services and clients alike.

 

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s