This is a post by guest writer Anna Fellet who I met in Rome last month. This further explores the theme of Controlled Language and Process Standards and continues on the themes presented Valeria Cannavina in her posting earlier this month. This slide presentation provides additional background on this case study.
=================================================================================
This article presents a case study to show the benefits of Controlled Language strategies, and highlights the key lessons learnt in the pilot project on a dedicated MT workflow created for ARREX Le Cucine, a leading Italian furniture company. This post also contains a reply to Laura Rossi’s comments on Valeria Cannavina’s previous posting on standards and the application of the CMMI model to translation.
A Business Case for Controlled Language
The goal of the ARREX project was the development of a corporate controlled language for Italian to be used in a customized authoring and (machine) translation workflow.
Why a Controlled Language?
A CL was chosen to eliminate ambiguity and complexity in product data sheets, installation and maintenance instructions (for support), catalogs, price lists, orders, reports, memos, documents for compliance. We chose to test our CL with both RbMT provided by Synthema and with SMT by Asia Online. We improved the repetitiveness of ARREX texts, and the result with RbMT was successful
As for SMT, which is a typical brute-force data-driven computing application, most of the difficulties come from the high degree of unpredictability in searching through a massive set of possible options of even in the simplest word combinations. As the number of words in a sequence increases, the precision score decreases because longer matching word sequences are more difficult to find. A controlled language increases predictability, increases statistical density and thus improves probability and boosts SMT success.
So, the single most powerful rule for authors/writers still holds its validity: one idea per sentence makes text that is easier for humans to understand also easier for MT engines to understand.
Poor source quality can lead to low quality target language content (e.g. SAP translations often result in hardly translatable/comprehensible Italian), however technical documents are ideally all written in the same “language”, even though with different idioms. Setting up terminology resources and developing writing rules enables the Italian text to be more easily handled by the MT system.
Moreover, language combinations with English are more commonly implemented, so by translating Italian into a terminologically coherent and syntactically simple English target we could use it as a starting point for other potentially successful combinations.
At the end of our preliminary investigations, we found that ARREX CL adds value to technical documentation as it allows:
- Increase in the perceived value of the product and of the whole brand: consistent, stylistically uniform, and controllable documentation (user-targeted material) created for a user/client to understand and thus helps to build customer loyalty;
- More efficient communication with clients/distribution partners/maintenance staff, thus reducing customer support calls and general costs associated with customer service;
- Reduction of translation costs (see table below);
Source text | Without CL | With CL | Difference |
Words to be translated | 70.000 | 64.000 | -8% |
Repetitions | 39.800 | 38.300 | +3% |
Words to be translated from scratch | 32.100 | 25.700 | -5% |
Human translation costs (250 words/hour) | 280 hours | 255 hours | -9% |
Human translation costs | MT post editing cost | Difference | |
Translation Costs | 280 hours | 50 hours | -80% |
We moved beyond the study on Italian CL for customized MT, and discovered that much can be done with a holistic approach to authoring workflows.
We became convinced that by adopting an ad hoc CL, by creating reusable corporate specific terminology resources, training corporate internal staff on authoring strategies for MT we could influence the company authoring workflow at a greater extent. By ad hoc CL we mean that rules are created specifically for ARREX. These rules may be valid for other domains, as well, but we had to focus on and improve inefficient writing practices unique to ARREX’s own internal corporate-speak. We were brought to focus on critical aspects that the company may not have had clear at the beginning of the project e.g. improve ARREX terminology standards by extracting most frequently used terms, and analyze synonyms and non-standard/irregular uses of terms that ARREX had already implemented.
We became convinced that by adopting an ad hoc CL, by creating reusable corporate specific terminology resources, training corporate internal staff on authoring strategies for MT we could influence the company authoring workflow at a greater extent. By ad hoc CL we mean that rules are created specifically for ARREX. These rules may be valid for other domains, as well, but we had to focus on and improve inefficient writing practices unique to ARREX’s own internal corporate-speak. We were brought to focus on critical aspects that the company may not have had clear at the beginning of the project e.g. improve ARREX terminology standards by extracting most frequently used terms, and analyze synonyms and non-standard/irregular uses of terms that ARREX had already implemented.
Not only did this project help us understand how to clean data for MT, create resources for MT (glossaries, TM, post-editing guidelines), highlight costly and time consuming translation processes, i.e. outsourcing translation/editing and publishing, it also helped us in seamlessly adapting our work to the existing corporate strategy by addressing the internal staff’s needs directly.
WORKFLOW & VALUE
The graphic below shows how we changed the company’s workflow and the results we achieved.
Activity | ||||
WORKFLOW ANALYSIS | Goal | Before | After | |
Exhaustive map of how ARREX processes are organized and who is in charge for what. | Texts were translated externally or (sometimes) internally. | MT (internal); monolingual review with support of term base and glossaries approved by ARREX. | ||
Value | Faster, cheaper and more accurate translations and reviewer’s feedback for continuous improvement of CL and MT. | |||
Activity | ||||
RESOURCES ANALYSIS | ||||
Goal | Before | After | ||
Measuring ARREX staff performance. | No trained technical writers and no unique point of reference for the production of technical material and translated material. | Training of staff to repeat and manage the process; new professionals (pre-editing, post editing). | ||
Value | Involvement through requalification of internal staff in charge for documentation. Resources are the only feature of the whole process that cannot be cut or reduced. People will always be the key element to deliver ‘quality’. Internal staff is the best people to talk to, to understand the quality level expected. We are only providing the right tools and the right knowledge to achieve such ‘quality’ (e.g. glossaries, CL style guide, MT workflow, QA report). We offer an improvement of the quality of the process, not of the product. Products can always change. | |||
PLANNING | Goal | Before | After |
Address the workflow step by step to build long term relationships with valid collaborators. | Undocumented processes, undefined organization of roles for technical writing and translation. | Ad hoc procedures for each phase of the process, from technical writing to delivery of translated material. | |
Value | The only possible way to deal with planning is to set a common framework to communicate with ARREX to find the appropriate strategy for text editing and MT. | ||
Activity | |||
ACTION | Goal | Before | After |
Write a protocol of requirements suitable for new requests. | Fragmented process. | Independent and autonomous management. | |
Value | Flexible processes become repeatable. | ||
Repeatable processes
- Process documentation;
- Roles definition and (re)qualification;
- Building of internal writing team;
- Internal terminology approval procedure;
- Target Language Monolingual reviewer selection;
- Target Language Monolingual reviewer feedback.
Q&A to questions posed by Laura Rossi in comments of previous post.
Laura Rossi: Will translation software developers be ready to provide their customers just with what they need, instead of trying to ‘impose’ an overall comprehensive solution, which, in fact, force them to follow a specific process and workflow?
Will the definition of a standard model not be another reason for them to justify this rigidity?
ARREX was anchored to old trusted but imperfect and inefficient processes, and the change we introduced was sometimes shocking for internal technical writers. “The difficulty lies, not in the new ideas, but in escaping from the old ones” (John Maynard Keynes), but if one sees the new idea as a means to improve one’s work (and save time), participation will be natural. In this sense, ARREX drove its own change.
Laura Rossi: As long as translation and localization will be considered as an accessory activity and a cost by the customers, more than a possibly business-driving and revenue-generating task, there won't be much interest from side of the customers to rethink their internal processes and organization, as well as from side of the LSPs and translation software companies to really act as part of their customers' development and production cycle.
I fully agree with Laura’s response to Valeria’s Post, it’s time for “translation (software) companies to really act as part of their customers' development and production cycle”, but I do not agree that service providers should “teach customers to involve LSPs in an early stage of development”. I think that it’s the other way around.
Providers should be able to integrate seamlessly in a company strategy for content, and detect processes that can be improved. This is what we could define as a holistic approach to content creation, where translation is only one piece of a broader internal and external corporate communication puzzle.
Laura Rossi: I think the landscape is actually changing, but the change is still quite slow, especially from the side of the customers, and I wonder what will be able to cause the shift on a massive scale from the traditional way of seeing translation as a 'service industry' to consider it an essential part of a business.
I might be wrong, but I suspect this shift is driven by economic imperatives, and MT offers terrific time and cost savings. Now MT is the right technology, handling repetitive tasks to let humans do what they are best at, but technology can be applied to processes, not to outcomes. One useful approach to a realistic, sustainable translation market is to explicitly differentiate between processes and outcomes.
This is why we focus on the quality of the process, instead of the quality of the product/output, and think of a Customer Centered Business Model, with single services satisfying multiple needs.
“Quality in a product or service is not what the supplier puts in. It is what the customer gets out and is willing to pay for. A product is not quality because it is hard to make and costs a lot of money, as manufacturers typically believe. Customers pay only for what is of use to them and gives them value. Nothing else constitutes quality” (Peter Drucker).
“Quality in a product or service is not what the supplier puts in. It is what the customer gets out and is willing to pay for. A product is not quality because it is hard to make and costs a lot of money, as manufacturers typically believe. Customers pay only for what is of use to them and gives them value. Nothing else constitutes quality” (Peter Drucker).
We see that:
- translation students are not trained on MT (in Italy), and mostly they don’t have a sense for the realities of the professional translation workplace after graduation;
- professional translators are very suspicious of MT, and generally do not welcome new ways of approaching the job, perhaps because they don’t have direct access to the client company, due to agencies (LSPs) intermediation;
- Agencies (LSPs) see MT only as a means to pay translators much less than what they pay them presently.
We believe there are possibilities for new approaches to content creation, and translation management, and that companies wishing to change the way they write and translate their content, like ARREX did, will drive this change, not LSPs.
Laura Rossi: Can we avoid the trap ‘we-are-following-a-standard-or-model-therefore-we-are-good’, which, as you say, can ‘hook’ the customers, but, in my view, does not necessarily ensure their satisfaction?
How can we make sure that a possible specific translation industry standard process model will be flexible and modular enough so as to avoid the risk of LSPs and customers ‘anchoring’ to that as a ‘given’ and a ‘must’?
During the ARREX project we saw that it was hard for clients to ask for specific services since they see translation as marginal and take it for granted, and as Renato Beninatto often says, “translation is really like toilet paper, it’s only important when it’s not there when you need it.”
We ended seeing translation as a product, and not as a service provided with methods akin to those of industrial production. With the commoditizatioin of translation, i.e. with almost no difference between suppliers, there is an undue and ineffective emphasis on prescriptive standards and the ‘we-are-following-a-standard-or-model-therefore-we-are-good’ scenario. Prescriptive standards, though, are useless for those LSPs wishing to differentiate, and to adapt their service to the client’s needs, because they may have to change their service and approach for the unique requirements of each client. Process Standards are not, cannot be, and should not be laws, not even strict regulations because every company is different. The general state of information asymmetry between the LSP and client, make a process standard useful only if it implies transparency and flexibility. Reiterative and rigid procedures, instead, lead to static monolithic workflows.
We ended seeing translation as a product, and not as a service provided with methods akin to those of industrial production. With the commoditizatioin of translation, i.e. with almost no difference between suppliers, there is an undue and ineffective emphasis on prescriptive standards and the ‘we-are-following-a-standard-or-model-therefore-we-are-good’ scenario. Prescriptive standards, though, are useless for those LSPs wishing to differentiate, and to adapt their service to the client’s needs, because they may have to change their service and approach for the unique requirements of each client. Process Standards are not, cannot be, and should not be laws, not even strict regulations because every company is different. The general state of information asymmetry between the LSP and client, make a process standard useful only if it implies transparency and flexibility. Reiterative and rigid procedures, instead, lead to static monolithic workflows.
This is why a scalable and transparent path like CMMI is useful. Only if client and customer are transparent in processes, can they find the most adequate actions to interact. The client will explain (and understand) its own level of maturity (requirements) to leverage the service provided, and the provider will be able to address the unique needs of the customer.
When it comes to adopting standards, a company does not know exactly where the ensuing process changes will lead. It can also lead to other changes that were not originally envisioned. This happened to ARREX as well: when they saw that along with improving their authoring and writing strategy, they could also act on other issues, i.e. translation, they did not hesitate in considering that, as well.
Laura Rossi: Is it really possible to capture in a standard something as ‘subjective’ as quality?
One could wonder what standards really mean to customers. Are they all concerned about “quality”? Quality is subjective in this sense that it is subjective and dependent on each customer.
A list of ‘ad hoc’ requirements to measure the level of adequacy of the service for the particular customer is useful both to assess customer satisfaction, service improvements, and to define client’s profile and demands in different domains. It is also true that if the quality of the product is subject to the assessment of the client, you can not say the same for the evaluation of the process. In fact, process standards should aim at increasing and improving the quality of the process. This can only be done with transparency in client-vendor relationships.
In our experience, we saw that processes such as post-editing can be measured either by the customer, in terms of satisfaction with the final result (via a series of requirements that must be met by the output of the MT output), and by the monolingual reviewer in a questionnaire on ‘linguistic’ aspects of translation and measurements of price/time/productivity. In this sense, requirements can be defined differently as: MT output for publication, pre-translation or internal use.
What I think would be extremely useful, and hope to see promoted in the industry, is a framework for pricing, especially for post-editing, to help customer-vendor relationships be more transparent. Crowdsourcing, as well, could be better and more widely accepted and used with a clear, simple, and common standards framework. Repeatable processes are worth sharing.
A list of ‘ad hoc’ requirements to measure the level of adequacy of the service for the particular customer is useful both to assess customer satisfaction, service improvements, and to define client’s profile and demands in different domains. It is also true that if the quality of the product is subject to the assessment of the client, you can not say the same for the evaluation of the process. In fact, process standards should aim at increasing and improving the quality of the process. This can only be done with transparency in client-vendor relationships.
In our experience, we saw that processes such as post-editing can be measured either by the customer, in terms of satisfaction with the final result (via a series of requirements that must be met by the output of the MT output), and by the monolingual reviewer in a questionnaire on ‘linguistic’ aspects of translation and measurements of price/time/productivity. In this sense, requirements can be defined differently as: MT output for publication, pre-translation or internal use.
What I think would be extremely useful, and hope to see promoted in the industry, is a framework for pricing, especially for post-editing, to help customer-vendor relationships be more transparent. Crowdsourcing, as well, could be better and more widely accepted and used with a clear, simple, and common standards framework. Repeatable processes are worth sharing.
Mark Zuckerberg said “By giving people the power to share, we're making the world more transparent.” It has proved to be a very profitable strategy, as well.
Anna graduated in 2007 in modern languages and cultures at the University of Padua, and in 2009 in technical and scientific translation at LUSPIO of Rome with a final dissertation on '‘Machine Translation: productivity, quality, customer satisfaction.’ At present she works as a freelance translator and subtitle translator and on a pilot project on Italian Controlled Language and Machine Translation with LUSPIO University, Asia Online, Synthema and ARREX Le Cucine. She can be reached at anna@s-quid.it http://www.s-quid.it
![clip_image002[4]](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjHO5MzH2VvduO3RCDHv3F0Vq7CAPtUIkTMVBK-Rbd1RG5E4w_z_dY3ZpDacZFdGcQqBFP0jOgox4g-VtT0cqTurcv2-Ed0QI6fku46JVYeJOmamcu-yaVslLGkDoenejvrR6gkjVP_pVwS/s1600-h/clip_image00243.jpg "clip_image002[4]")