All concepts cannot be translated simultaneously and it is necessary to determine which parts of the terminology should be translated in the project and in which sequence. The preparation therefore includes the selection (creation, allocation) of specific translation subsets. Also, in the preparation, re-use of translations from other NRC’s and code systems should be considered.

3.1 Translation subset selection 

The aim of subset selection is to provide translators and reviewers with the best possible prerequisites for their work. The project will benefit from using the same translator working with concepts within the same subject field, e.g. heart conditions including body structures, disorders and procedures.

The choice of order of areas to be translated can be determined by local use cases or by choosing to translate concepts that create maximal efficiency. Different approaches for identifying the initial subsets can be chosen depending on local needs. For example, one could prioritise concepts:

• which recur, link, qualify and confer meaning to other concepts in different contexts, e.g. body structures and qualifier values. Translating such concepts in the beginning of the project maximises re-use of translations, which increases both efficiency and consistency. It is advisable to translate concepts representing attribute targets for clinical finding and disorder concepts before translating the clinical finding hierarchy, using ECL queries. This supports both consistency and efficiency.
• which are relatively easy to translate, because they are similar and/or well-defined – for the same reasons
• which are to be part of pilot projects in a given health or social care context, or part of a specific research project – especially if your resources are limited.

One can also choose to avoid certain branches or types of concepts. Some branches are more difficult to translate than others, because they incorporate more culture- or country-specific concepts. For instance, if there is no private health care in your country, you will have little use for the 49 concepts for private referral to. Other branches may contain much more detail than you require. For instance, the branch 418019003 |Accidental event (event)| contains an astonishing wealth of accidents, e.g. 214871001 |Motor vehicle nontraffic accident involving collision, not on public highway, between motor vehicle, except off-road motor vehicle, and pedestrian, passenger on motorcycle injured (event)|. Translating these may require much time for relatively little gain. The same could be said about parts of the food hierarchy. Translating 226674006 |Welsh cake (substance)| into Norwegian makes little sense. 

Selection of subsets is preferably managed by a team including specialists in health and social care, terminologists and experts on SNOMED CT.

3.2 Re-use of translations from external sources (NRC’s or from different code systems)

A pragmatic approach for adding translations may be to “borrow” existing translations from external sources, e.g. other medical terminologies or translations produced by NRC’s in different member countries who share the same target language(s) or other code systems, such as Orphanet if already translated into the target language.

When considering the reuse of “legacy” translations, it is important to first analyse any differences in scope, regional variation and formal aspects. In order for the descriptions to become part of the national SNOMED CT extension, the NRC should verify that the translated concepts have the same meaning and granularity, and that the translation has been produced in a qualitative manner including clinical validation, similar to the SNOMED CT translation approach. If this is the case, it is still possible that the translation needs to be adapted in order to comply with regional appropriateness, formal requirements and linguistic guidelines.

Depending on the nature of the external source, there are a number of considerations when selecting and recycling existing translations:

• Translations belonging to the same code system, i.e. SNOMED CT but produced by different NRC’s. In other words, the concept in the source language has the same ID, has the same meaning and position in the hierarchy. The target audience however is a different country or region, and the language flavour may imply semantic, lexical as well as formal differences. For instance, the Belgian NRC uses Dutch descriptions produced by the NRC of The Netherlands as a starting point for untranslated concepts. Even though grammar and spelling is standardised for Dutch in both regions, there may be discrepancies in meaning between Belgium and The Netherlands that need to be verified. In addition, different guidelines related to formatting and formal choices may apply in the two member countries. E.g. in The Netherlands the FSN is translated whereas in Belgium this is not the case.

Example:

Translations belonging to different code systems but for which there is a semantic mapping between (source language) concepts. For instance, the SNOMED CT to Orphanet Map Release is the product of a joint project carried out under a collaboration agreement between the two parties. As this map links rare diseases managed by Orphanet to concepts in SNOMED CT, it could be an option for NRC’s to also recycle the translated terms available in Orphanet. As translation guidelines applied by Orphanet may deviate, the NRC needs to make sure that also the formal aspects of the recycled descriptions correspond to their own formal requirements. E.g. CMT2B is an acceptable synonym in Orphanet, however in SNOMED CT every acronym needs to be followed by the fully written out form.

Example:

Translations belonging to different code systems and for which there is only a lexical similarity between the concepts in the source language. It may be risky to recycle translations from external sources as the meaning of terms that share the same lexical morphology in the source language may be different (i.e. homonyms but also different level of granularity). Examples of external resources:

•  WHO: 
  • ICD-10/ICD-11
  • the Anatomical Therapeutic Chemical (ATC) classification system
•  International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH): MedDRA
•  SAM V2 (Belgium)
•  Terminologia Anatomica

3.3 Creating a text corpus

Easy access to medical texts is highly valuable when translating SNOMED CT into the target language. There might already exist text corpora consisting of medical texts in the target language for research purposes at universities or other higher education institutions. If so, this could be re-used for translation of SNOMED CT.

If there is no corpus available, creating one should be considered. It should be noted, however, that this might require quite some time and resources. Criterias such as balance, sampling and representativeness should be taken into account when considering which content to include in a corpus. Also right tooling for searching and analysing the corpus should be made easily available for the translators. One option is to collaborate with researchers of corpus linguistics when creating a text corpus. Another is to seek assistance from a National Language Bank if there is one in the country.