Translation Proxy: More Websites, More Work

With new websites being created every day and with the ever-increasing demand for multilingual content online, translation proxy has become quite popular. Offered by tech-enabled language service providers (LSPs) as well as technology companies, proxy helps dynamically translate website content on the fly. As a result, end-users gain access to a website in the desired language in real time.

In the past, the technique used by proxies to find translatable content on the original websites often failed, resulting in sections remaining unlocalized or a mix of languages being displayed. It mostly worked fine for simple static websites. However, over the course of 2020, the market saw a number of developments, so it’s a good time to take a quick tour of a few  of the latest and greatest translation proxies and what they’re capable of.

What is translation proxy?

Proxy is an intermediary server separating website visitors from the destinations they browse. Traffic flows through the proxy server on its way to the requested address. It then comes back — usually through the same proxy server — with the proxy forwarding end-users the data received from the website.

So, a translation proxy is a system that's layered on top of a website — like a mirror where the reflections are the website's translated versions. It functions as an ad-hoc localized website placed in between the visitor’s requesting site and a website in the original language.

Source: Nimdzi

It can operate as follows:

1. When a visitor wants to access a website in a local language, the proxy detects their location and browser language settings. 

2. The proxy’s server, which sits between a source website and a visitor, receives the request for the translated site and routes the request to another web server. 

3. Content is translated in real time either through machine translation (MT) or human translation (e.g., with the help of a translation memory, or TM). TMs store previous translations and are leveraged by the web server.

4. The second web server recreates the structure of the site in the needed language.

When content changes or is added on the original website, a web crawler monitors it and the adjustment occurs on the proxy site either by querying the database or triggering a translation task. In the latter case, translations will need to be produced and then are typically pushed online within a business day or so instead of in real time.

The proxy and JavaScript website localization market

Based on Nimdzi’s market data, the web proxy and JavaScript (JS)-based market subsegment brings together 25 different companies.

Source: The Nimdzi Language Technology Atlas 2020

The proxy subsegment encompasses the following groups of companies:

• Technology specialists such as Text United and Easyling who mainly sell their technology via LSPs. Large LSPs whitelabel Easyling’s solution — e.g., Lionbridge (Lionbridge Proxy) or Amplexor with SiteSync (previously with Transplicity).
• Industry-leading LSPs with revenues upwards of USD 500 million who have diversified their client offering by developing proprietary technology such as Transperfect (GlobalLink/Onelink).
• Technology-driven LSPs that sell their professional services in combination with a proxy solution, such as WeGlot (which also automatically handles multilingual SEO).

Most of the proxy solutions do not allow the translated content to sit on the customer's content management system (CMS) which gives rise to security concerns.

A closer look at two technology examples

Easyling

Easyling can be used to publish translations in a variety of ways. The solution features three major publishing modes: proxy, JavaScript, and static HTML. Each publishing mode focuses on a different use case. In its proxy mode, Easyling sits between the original CMS and the visiting browser, operating on the source language responses. As it must be injected between the two,  this might mean additional firewall configuration or clearance by INFOSEC depending on the organization.

That’s why, since 2020, Easyling has been developing a JavaScript-powered translation solution that bypasses the proxy aimed at highly regulated or privacy-conscious content owners. In this publishing mode, the translation is handled by a JavaScript snippet provided by the service, which downloads the segment pairs (source + target) with associated metadata and applies the translations in the client’s browser. 

Any content which cannot be translated is reported back to Easyling for handling as part of the regular maintenance workflow. In this phase, the client resumes generation of content (new articles, blog posts, product pages, etc.) that requires translation. Easyling detects the new content, ingests it, and enters it into the LSP's translation process, which ends with the content being imported back into Easyling and the site appearing fully translated again. 

This makes the JavaScript offering particularly suitable for translating sites on corporate networks that are otherwise inaccessible to the proxy option: it runs on the visitor’s machine and the Easyling backend doesn’t need to handle sensitive data. But as the content needs to be ready before it can be translated in the browser, a flash of untranslated text (a phenomenon where you see the source language content for a brief instant  before the browser signals that everything has loaded and the content is replaced with the localized versions) is inevitable. Notably, server-side translation sidesteps this, since the browser will only ever see localized content in the first place.

MotionPoint

MotionPoint positions itself as a specialist in the web proxy translation system subsegment of the TMS cohort. Their technology does not store customers’ content on its server, and provides interaction between the website and its users on a browser level. This makes the solution compliant with various IT security standards. Furthermore, the security and compliance framework in place meets the needs of customers in highly regulated industries such as healthcare or the financial industry.

MotionPoint’s website localization solution is said to require no integration effort and to take fewer than 5 hours on average to deploy. Their EasyLink technology “welcomes first-time website visitors in their preferred languages.” This reduces bounce rates and increases time on-site.

MotionPoint’s proxy can work with JavaScript to identify and translate content that’s generated on-the-fly by product inventory applications or single-page applications.

Autopilot instant translations?

There are known downsides of translation proxy. While in the ideal deployment scenario, the localization effort remains unseen to the end-user, in reality bits and pieces of unlocalized content are still noticeable (remember that one random time you wanted to shop on an international website?). 

Raw MT, which is often used to deliver translated content on the fly, may fail and confuse the customer even further. Also, the approach itself implies that content relevant in the source language remains relevant in all other languages, which is not always the case.

Not to mention that the website in question may be hosted on a network isolated from the outside, preventing access to the crawler — or there may not even be a route in.

Autopilot instant translations of high quality are NOT a widespread reality.

However, the ability of translation proxies to distinguish a website’s translatable content from its code are improving. Currently, some web crawlers are able to identify translatable content within HTML, CSS, JavaScript, JSON, XML, and other formats at the start of a website translation project. Worthy proxy solutions are also capable of detecting translatable content in graphics, videos, PDFs, apps, etc.

The trend and demand for “effortless localization” helped proxies develop in a way that enables more accurate language “swapping” in real time. Furthermore, development of solutions that bypass proxies help present a localized user experience for owners of highly regulated content.