From Server-based to Web-based Translation Memory Systems:

Benchmarking, Testing and Implementation in STAR7

Enrico Giai

, Nicola Poeta

and David Turnbull

STAR7, S.p.A., Language Technologies Expert, Corso Orbassano 336, Torino, Italy

STAR7, S.p.A., Global Content, Service Line Leader, Corso Orbassano 336, Torino, Italy

STAR7, S.p.A., Language Lead, Corso Orbassano 336, Torino, Italy

Keywords: Corporate Language Management, TMS, Transit NXT, WebEdit.

Abstract: In the age of cloud computing and Software as a Service (SaaS), the need for web-based translation manage-

ment solutions is on the rise. Connecting authoring tools and other Content Management Systems (CMSs) to

Translation Management Systems (TMSs) is key to reaching a global audience quickly, effortlessly and effi-

ciently. To this end, new web-based TMSs have been developed to automate the entire information lifecycle

and allow cooperation among all stakeholders – from authoring to translation, review and publishing.

In this paper we describe the features of a web-based TMSs, identify the contexts in which such a system is

required and why, and look at their benefits. To do so, we will describe three different processes which use

three different technologies based on varying automation and collaboration needs, from a server-based instal-

lation of Transit NXT to a web-based solution such as CLM WebEdit.t

1 INTRODUCTION

Computer-Assisted Translation (CAT) tools have

been a part of the translation industry since the late

1980s. For instance, STAR Group’s STAR Transit is

the second-oldest commercial-grade CAT software

ever created and it is the oldest one still on the market.

In its essence, a CAT tool is made up of four main

components:

1. A Translation Editor, where linguists can see the

source texts and insert target translations;

2. A Translation Memory (TM) system which en-

ables pre-translation

and fuzzy matching

be-

tween the source files and previous translations,

as well as concordance searches;

3. A Termbase (TB) Management System in which

specialised glossaries are created, shared and

maintained;

4. A Quality Assurance (QA) system, which al-

lows for advanced linguistic and structural con-

formity checks.

‘Pre-translation’ is the 100% recovery of translations

from previously translated texts, independent of source-

file formats.

From the early days up until the late 2000s, a stand-

ard CAT tool would consist of a piece of software in-

stalled on the end user’s personal computer, typically

a Windows PC.

This approach was based on a software licencing

model, in which users had to acquire a physical copy

of the software or download it from the Internet; then,

the software was activated with a licence number.

This model is still in use today. However, it has

some disadvantages:

 The CAT tool’s performance is strictly depend-

ent on the local machine’s computing power;

 Translation assets – typically language pairs

(source text and target translations), TMs, TBs,

and any other reference material – must be

packed and transferred via e-mail

to the lin-

guists and back to the project managers;

 Real-time collaboration is not possible, as each

linguist is tied to the files on their local machine;

 Project automations and connectors with author-

ing systems cannot be set up, meaning that files

A ‘fuzzy match’ is a TM suggestion deriving from sim-

ilar already translated texts.

This can also cause data integrity and security issues, as

discussed in subsequent paragraphs.

216

Giai, E., Poeta, N. and Turnbull, D.

From Server-based to Web-based Translation Memory Systems: Benchmarking, Testing and Implementation in STAR7.

DOI: 10.5220/0011585400003335

In Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 3: KMIS, pages 216-220

ISBN: 978-989-758-614-9; ISSN: 2184-3228

must be manually imported and delivered to cli-

ents;

 Client portals and project monitoring dash-

boards cannot be implemented in manual work-

flows.

A step forward in this approach is the use of server-

based solutions. Many of the available CAT tools are

also offered in a ‘server’ version, which can be in-

stalled and run from a Windows Server machine. This

is beneficial to anyone in an organisation who has ac-

cess to that server, for instance via a Virtual Machine

(VM). In fact:

 CAT tools installed in a server environment can

typically leverage higher computational power

and better reliability.

 Translation assets can be transferred to a data-

base and shared across the network.

 Real-time collaboration is possible for anyone

who can access the VM.

 While running 24/7 on a dedicated server, inte-

grations can be created and run as a service to

allow for automated project creation and deliv-

ery via APIs or hotfolders

 Client portals and project monitoring dash-

boards can be implemented.

While this architecture is an improvement, adding

a web-based layer to the system would allow for

greater flexibility: the systems could be accessed di-

rectly via a web browser, without the need to physi-

cally install anything on the user system and without

the need for setting up VMs and remote accesses. In

this context, the only real requirement is a reliable In-

ternet connection – everything else is taken care of

remotely.

In this paper we will describe how STAR7’s CAT

technologies have evolved from server-based to web-

based, and what the benefits of this system are in

terms of productivity by analysing actual implemen-

tations of these technologies in the production envi-

ronment.

2 RELATED WORK

The benefits of cloud computing and web-based CAT

technologies in general were extensively described by

Muegge back in 2012. In his article, the translation

A hotfolder is an FTP folder that is constantly monitored

by a service to detect changes in its content and trigger

automated actions.

technologies expert explains the benefits of a central-

ised Translation Memory System (TMS):

 No need to install any application

 Software is always updated

 Cross-platform compatibility

 Ease of access and enhanced collaboration

 Low cost

Nonetheless, some critical drawbacks should be

considered as well:

 Constant, good quality Internet connectivity is

essential

 Privacy and confidentiality issues

 Control over and ownership of linguistic assets

Muegge concludes that the advantages far out-

weigh the drawbacks – and this is particularly true

considering the advances of the last ten years. Ac-

cording to data from Eurostat, as of 2021, 90% of EU

households now have access to a broadband Internet

connection, a 17% improvement compared with data

from 2012. As for data protection and privacy con-

cerns, the General Data Protection Regulation

(GDPR) established in 2016 has helped in ensuring

the protection of personal data and privacy in the Eu-

ropean Union and in the transfer of such data outside

the EU.

The inefficiency of the ‘traditional desktop para-

digm’ compared to SaaS solutions has also been ex-

plained by Zydroń. In his article, the focus is set on

the benefits of a centralised SaaS solution not only for

Language Service Providers (LSPs), but also for all

the stakeholders in a translation project. In addition,

in such a solution, the CAT tool would only be a part

of a much wider system of integrated components

aimed at automating tasks. As he states, “A proper

SaaS TMS/CAT solution will allow you to integrate

a customizable web-based ordering and payment sys-

tem into your own website so that customers can up-

load, get quotes, pay online and initiate the workflow

for translation jobs.”

The translation industry in general is also moving

towards a cloud-based approach. According to a sur-

vey run by Nimdzi in 2020, the majority of TMS in-

stallations are now on the cloud – be it private or pub-

lic.

In this paper we will move forward from this and

discuss how the implementation of systems with var-

ying degrees of centralisation and process automation

has been beneficial in practical terms.

From Server-based to Web-based Translation Memory Systems: Benchmarking, Testing and Implementation in STAR7

217

3 METHODOLOGY

As a member of the STAR Group Network – the

world’s thirteenth largest Language Services Pro-

vider (LSP) in terms of revenue in 2022 – STAR7 has

been using STAR Group’s technologies since its

foundation in 2000. Therefore, the technology selec-

tion benchmarking has been conducted internally and

on proprietary technologies.

The results described in this paper outline

STAR7’s evolution in terms of translation technolo-

gies and how this has been beneficial in terms of

workflow automation and scalability. In particular,

the paper reports how the implementation of the Cor-

porate Language Management (CLM) system

has

helped STAR7 to effortlessly manage millions of

words per year, and how the WebEdit Web-based

component is helping to move the bar even further.

This said, the three models presented in the results

are still all in use at STAR7: each automation level

has its own pros and cons, and the technology selec-

tion for each client is evaluated based on volumes, au-

tomation potential and turnaround requirements.

4 RESULTS

4.1 Model 1 – Server-based

Installation: Transit NXT

In this scenario, Transit NXT (STAR Group’s propri-

etary CAT tool) is installed on a Windows Server ma-

chine that can be accessed by all project managers and

any internal linguists or QA specialists. The server

also connects to the Terminology and Translation

Memory SQL Server databases, to centralise both ter-

minology and translation memories.

While this approach is beneficial as it allows for a

higher level of centralisation, the resources can only

be accessed internally within the organisation. The

‘thin’ clients

can access the Terminal Server instal-

lation and TM and TB assets only via VMs, so access

to the local area network is required. For instance, ex-

ternal linguists are excluded from this architecture.

Translations are sent in the form of Project Pack-

age Files (PPFs) containing language pairs, TM ex-

tracts containing only reference segments useful for

the project, TBs, and any other additional supporting

For further information, see: https://www.star-

group.net/en/downloads/star-clm.html

A ‘thin’ client is any PC that accesses a main server to

perform the actual computational operations. In this

Figure 1: Terminal Server Architecture.

material. This package is typically shared with lin-

guists via e-mail, who can then unpack the project

files to their local Transit NXT installation.

Linguists can return their Translation Package

Files (TPFs) to project managers, who can receive

them, export the completed files, and update the cen-

tralised TMs with the finished translations.

4.2 Model 2 – Server-based Installation

with FTPS File Transfer: CLM and

WebTransit

A second level of automation has been implemented

with the introduction of the CLM system. This is an

advanced Translation Management System that can

be accessed via web interface and lets clients create

new translation requests or even connect to third-

party Content Management Systems (CMSs) or hot-

folders to automate translation requests.

From there, projects are created automatically us-

ing standardised project templates, and the system en-

gages a Transit NXT command line session to auto-

mate the project creation, file import, package crea-

tion and delivery to linguists.

In this automated workflow, linguists receive a sys-

tem notification when a new translation is requested.

Instead of sharing them via e-mail, packages are sent

via a secure FTPS connection and received via

WebTransit – Transit NXT’s integrated component to

receive and distribute translation jobs created in CLM.

Translated jobs can be uploaded to the system via

WebTransit itself, and the CLM system handles both

the update of the TMs and the final file delivery to the

clients, who can retrieve their finished jobs either via

their CLM portal or in SFTP folders.

case, the PCs access a server where Transit NXT is in-

stalled, and all operations are carried out in that environ-

ment.

KMIS 2022 - 14th International Conference on Knowledge Management and Information Systems

218

The benefits of this approach are the higher level

of automation, centralisation and scalability the sys-

tem offers. Nonetheless, the system still requires lin-

guists to have Transit NXT installed on their local

machines.

4.3 Model 3 – Web-based Installation:

CLM and Webedit

In the third evolution phase, the CLM system is im-

proved upon by offering a fully fledged online Trans-

lation Editor in the form of the WebEdit module.

In this framework, the capabilities of CLM in

terms of workflow automation are combined with the

benefits of online translation, in that translators and

reviewers are no longer tied to software residing on a

local PC. This is valuable in other ways, too:

 Data security is ensured, as translation files are

no longer saved to local machines but on a cen-

tralised server.

 Data integrity is also secured, as the risk in terms

of data loss is lower and a Disaster Recovery

Plan (DRP) can be implemented in case of dis-

aster.

 Linguists can start working on one machine and

resume working on another one without the

need to transfer data.

 Linguists using an operating system other than

Windows can still be involved in translation

projects, as the web-based system is OS-

independent.

In other words, CLM WebEdit preserves all the

benefits of CLM described in Model 2, while adding

online translation and review functionalities. In this

framework, the TMS can also be useful on the client

side, as the review step can be handled internally

without the need for additional software installs.

As the system is accessible not only to lin-

guists – who are used to working in a TMS environ-

ment – but also to non-experienced users, a great deal

of effort has been made to provide a simplified, user-

friendly experience while preserving all the core

functionalities of a TMS. Features like a quick access

toolbar (1), the language pairs in column form (2), the

terminology and QA windows (2, 3) as well as the

Fuzzy matching and Concordance search windows (5,

6) are all present and easily accessible, as shown in

the figure 2.

User training and feedback have shown that the

learning curve for the use of WebEdit is particularly

favourable: both linguists and clients have reported

ease of use and clarity among the most appreciated

features.

Figure 2: WebEdit’s Translation Editor.

4.4 Model 4 – Cloud-based Installation:

Future Developments

While the approach in Model 3 is an improvement in

relation to the first two models, it can still be im-

proved upon. At the moment, a server infrastructure

is needed for the installation of the system compo-

nents, which may be an issue in terms of costs and

security concerns. Work is underway to make CLM

and WebEdit deployable as cloud-based systems,

available both in services like Azure and AWS. This

would mean outsourcing the computing power to

commercial-grade IT infrastructures for even higher

reliability and performance.

4.5 Implementation

The server-based solution has been implemented

since 2014 to automate the translation workflow of a

top-tier client in the Truck & Bus sector. In this sce-

nario, the client’s authoring system saves XML files

for translation in a shared FTPS hotfolder, which is

constantly monitored by the CLM FTP component.

New files in the hotfolder trigger the creation of new

projects by using project templates, which are se-

lected based on codes in each file name. Translation

and review jobs are sent to selected translators based

on a ranking logic, and the completed files are saved

back to the FTPS folder, ready to be handled by the

client’s authoring system.

Figure 3: Server-based workflow automation with FTPS

hotfolder.

From Server-based to Web-based Translation Memory Systems: Benchmarking, Testing and Implementation in STAR7

219

The number of words processed for the client has

been rising ever since implementation in 2014 and

has now reached approximately 520 million words

per year. This volume would not be feasible in a tra-

ditional environment, both in terms of workforce and

in terms of system capabilities: for instance, the main

TM for the customer now contains 80 million unique

segments in more than 35 languages. The result is an

exceptionally high level of leverage in terms of recov-

ery from existing translations.

Figure 4: Total words managed for the client in STAR

CLM, by target language (2021).

The system has also been used to manage the trans-

lation requests of clients ranging from the Agriculture

sector to Automotive, as well as clients in the Home

Appliances and Fashion sectors, proving that the sys-

tem is well suited to all kinds of workflows. In situa-

tions where no connection between CMS and TMS

needs to be established, clients can also use CLM’s

Client Portal to upload translation requests and trigger

the same automated workflows.

With the release of CLM WebEdit in 2020,

STAR7 has been hard at work migrating existing cli-

ents from the previous server-based system to the new

web-based solution. In addition, new clients have

been migrated to WebEdit, as the online translation

and review module is also attractive in terms of in-

house client review. In that respect, a client in the

field of Sports & Fitness has successfully imple-

mented the CLM WebEdit solution to request InDe-

sign catalogue translations from STAR7 while man-

aging internal market reviews by using WebEdit. Pre-

viously, the final step was performed using comments

in PDFs, in which the client would report corrections

that STAR7 had to make in both the target files and

in the TM. Using WebEdit has drastically improved

productivity, as corrections are now directly imple-

mented in the working files and in the TM.

5 CONCLUSION

In this paper we set out to describe different automa-

tion models and workflows using STAR7’s transla-

tion technologies. As translation processes require

higher automation levels and translation volumes

grow higher, the need for reliable, structured and scal-

able solutions grows consequently. This is why

STAR7 decided to adopt the server-based model first,

and the web-based model later, to ‘future-proof’

translation workflows. As new technologies and IT

architectures are developed, research activities are

constantly pushed forward to optimise translation

workflows and attract existing or prospective clients

with additional features and processes aimed at sim-

plifying tasks that could otherwise be automated. An

area that is currently under development is that of Ma-

chine Translation (MT) and Post-Editing Machine

Translation (PEMT) workflows, which have been

successfully implemented in CLM using STAR MT

technology as well as commercial MT engines.

Potential future developments can be made – as al-

ready mentioned – in cloud computing and in imple-

menting Artificial Intelligence models to improve

upon existing processes that are still human-driven, to

assist the many actors in the translation industry.

REFERENCES

Eurostat Data on Households with Broadband Access,

https://ec.europa.eu/eurostat/data-

browser/view/tin00073/default/bar?lang=en, last ac-

cessed 2022/08/04.

Muegge, U. (2012). The silent revolution: Cloud-based

translation management systems. In tcworld, July 2012,

17–21.

Nimdzi 100: The 2022 ranking of the largest language ser-

vices providers in the world, https://www.nim-

dzi.com/nimdzi-100-top-lsp/, last accessed 2022/08/04

Nimdzi History of TMS Technology, https://www.nim-

dzi.com/the-history-of-tms-technology-from-the-80s-

to-2010, last accessed 2022/08/04.

Nimdzi Language Technology Atlas: https://www.nim-

dzi.com/nimdzi-language-technology-atlas-2020, last

accessed 2022/08/04.

Official Journal of the European Union, Regulation (EU)

2016/679, https://eur-lex.europa.eu/legal-con-

tent/EN/TXT/HTML/?uri=CELEX:32016R0679&fro

m=EN, last accessed 2022/08/04.

Zydroń, A. (2012): Perspectives: Cloud computing, SaaS

and translation tools. In MultiLingual, January 2012,

pp. 20–21.

KMIS 2022 - 14th International Conference on Knowledge Management and Information Systems

220