Translation reuse via translation memories and terminology management has played a critical role in reducing the business cost of localization and translation, but there are some restrictions inherent in the reuse paradigm. Translation reuse employs computer-assisted techniques to accumulate and store the results of what is fundamentally a human translation and terminology research effort. Both the terminology databases and the translation memories used by translators as part of computer-assisted translation workstations are necessarily populated by the actions of the translators themselves. As human translators solve terminological or translation problems by dint of research involving parallel texts from the Internet or other document corpora, they create records of those solutions and store them. Over time, as other problems are solved, terminology databases and translation memories are populated with many potential translations for the technical terms and unusual sentences that are often encountered in specialized translation and software localization. These solutions can then be reused in the context of translation workstation software. Although there is an accumulation of translation and terminological data over time, there is a time lag between the advent of any given translation project and the point when translation databases for the project reach an optimal size and scope.

Time lag is not the only problem with the translation reuse paradigm. There are usually also significant restrictions on the scope of records in translation data-bases. The scope of a record, defined as the number of texts and contexts consulted before making and recording a translation decision, is significantly constrained by the time available for translation research. Due to the constant pressure of deadlines, translators and localizers typically pursue the identification and documentation of terminological or translation equivalents by consulting parallel texts (documents of the same text type) and background texts (documents in the same domain) only to the point where they are satisfied that they have found an acceptable equivalent. Deadlines do not motivate searching additional parallel or background texts for other possible equivalents that may be better or more accurate matches for a source-language term or phrase. The translation quality benefits of extended research may be outweighed by the economic liabilities of extended project time. Of course, this assumes that resources for doing extended terminological or translation research, in the form of collections of parallel and background texts, are readily available and in a useful form. Fundamentally, the basic dependence on human effort to populate terminology glossaries and translation memories has several inherent constraints: the time required to populate databases, economic or other restrictions on availability, research time, and the scope or range of documents consulted to solve terminology and translation problems.

Current business policy in the language industry dictates that localization and translation vendors retain and aggregate the terminology databases and translation memories accumulated by their translators and localizers, creating a shared pool of language resources. As a translation company continues to populate and aggregate its databases in the domains in which it translates, the time lag and human effort for any given subject domain declines, while the range of coverage increases. However, as new domains are added to the translation portfolio (a phenomenon called domain shift), the lag / scope / effort problem will reoccur. Even assuming a retention and accumulation policy, the number of parallel texts and research documents consulted before translators make and record translation decisions will still be dependent on access to corpora with relevant documents (the Internet, document repositories, digital libraries) and, of course, project deadlines. A database or translation memory accumulated by pooling the resources of several translators is larger than one compiled by a single individual, but individual records or alignments are still constrained by the same limitations as those of smaller databases.