First/první Previous/předcházející Home Next/další Last/poslední

MT reshaping the language industry
 05  Aktivity, agentury, chyby v překladech




Proceedings of MT Summit XVII, volume 2 Dublin, Aug. 19-23, 2019

Table 4 shows the impact of MT on the global economy, by highlighting that 7 of the largest 8 companies by market cap have a notable presence in this technology sector.

Překladatelské agentury

používají vlastní

strojové překladače
 

https://slator.com/data-research/the-slator-2018-language-service-provider-index/




8 Data for Machine Translation

Availability of data is crucial as almost all contemporary machine translation systems are based on data-driven techniques.

EU institutions have released massive volumes of freely available language resources that contain data for more than 24 EU languages and exceed 5 billion words.

Open data repositories in North America and Asia (e.g. US Government open data, Japan government open data portal) provide only few if any language resource.

Global dominance of companies like Facebook, Google, and Amazon in their primary business activities in the fields of social media, internet search and e-commerce allow them to harvest unmatchable amounts of data that they can use in other areas of their activities like MT.

 



Figure 2. Average # of errors per category.

Training started from a base level, explaining what MT is and how it integrates into the business’ workflow, to a practical module containing Post- editing guidelines.

Průměrný počet chyb podle kategorií.

 



Figure 17. Error types

Finally, a repetitive error in NMT is the change of register. In this phrase, not only the verbal form is incorrect but also changes from the second person plural to the singular even if it is the same in the source text.

6 Conclusions

The main conclusion is that although NMT seems promising in frequent language combinations, especially if English is involved, it is not obtaining the desired results in low-resource languages such as the pair Spanish-Galician.

NMT has not yet unseated RBMT and PBMT, performing, in fact, worse than these systems.

 



Lokalizační systém Microsoftu

nedosahuje u češtiny přijatelnou kvalitu, tj. 30%
 

2 The Microsoft Office Software Localisation System


First/první Previous/předcházející Články / Articles Next/další Last/poslední

Copyright Ing. Milan Čondák 02.09.2019