blog thumbnail (1)
Share to:

3. Big Translation Obstacles

2021-03-16
Clients who outsource their technical documents to translation service providers are very demanding. On top of asking for lightning speed delivery, they are also very particular about accuracy and fluency. Afterall, a translated passage filled with mistakes does not do much good for the professional image of the content owner.
 
Independent translators and translation service providers, therefore, are often faced with a common dilemma: they have the language skills to produce fluent texts in the target language but looking up all the bilingual equivalents of technical terms becomes a headache. To ensure that every word is correctly rendered is not impossible, but it is a very expensive process – the time and human resources needed to produce the perfectly accurate and fluent technical translation can easily cost most, if not all, of the narrow profit margin for any translator.
 
What does it take to achieve a quality translation in a cost effective manner? Chilin’s two-part answer is:
 
  1. The best possible domain specific machine translation.
  2. A post-editing tool that helps translator easily resolve difficult to translate technical terms. 
 

Custom, Domain Specific, Machine Translation

A custom, domain specific, machine translation engine can produce better results than a generic MT system. Here is an example:
 
The following is a 37-word English sentence taken from a pharmaceutical patent document.
 
Original English Text
GFP (Figure 16) was constructed by cloning the GFP fragment from pCR. Bgl-GFP-Bam as a Bglll/BamHl fragment into BamHl-digested pCMV. BEV. 2. This plasmid serves as a control in some experiments and also as an intermediate construct.
 
With so many free online translation tools available in the cyberspace, it cannot be difficult or expensive to turn this into Chinese, can it?  Let’s do just that and run the sentence through a few popular free online translation tools.
 
Original English Text
 
 
GFP (Figure 16) was constructed by cloning the GFP fragment from pCR. Bgl-GFP-Bam as a Bglll/BamHl fragment into BamHl-digested pCMV. BEV. 2. This plasmid serves as a control in some experiments and also as an intermediate construct.
Translation (1): From popular generic Machine Translation Tool A
 
通过从pCR克隆GFP片段来构建GFP(图16)。将Bgl-GFP-Bam作为Bglll / BamH1片段插入BamH1消化的pCMV中。电动汽车。 2.该质粒在某些实验中用作对照,并且也作为中间构建体。                    
 
Interestingly, the full-stops before and after the term “BEV” is read by the machine as a separate sentence and interpreted as “Battery Electrical Vehicle”, giving us “电动汽车"
 
Original English Text
 
GFP (Figure 16) was constructed by cloning the GFP fragment from pCR. Bgl-GFP-Bam as a Bglll/BamHl fragment into BamHl-digested pCMV. BEV. 2. This plasmid serves as a control in some experiments and also as an intermediate construct.
 
 
Translation (2): From popular generic Machine Translation Tool B
 
GFP(图16)是通过从pCR中克隆GFP片段构建的。Bgl-GFP-Bam作为Bglll/BamHl片段进入BamHl消化的pCMV。贝弗。2这个质粒在一些实验中作为对照,也作为中间结                                          
           
 
Again, the full-stops of the same term is causing some problem for Tool 2. This time, “BEV” is given a transliteration and becomes “贝弗”.
 
Now, let’s turn to the available sanctioned human translation.
 
Original English Text
 
GFP (Figure 16) was constructed by cloning the GFP fragment from pCR. Bgl-GFP-Bam as a Bglll/BamHl fragment into BamHl-digested pCMV. BEV. 2. This plasmid serves as a control in some experiments and also as an intermediate construct.
 
Translation (3): Human
 Translation
 
 
通过将pCR.Bgl-GFP-Bam的GFP片段作为 BglII/BamHI片段克隆至经BamHI消化的pCMV.BEV.2中而构建质粒CMV.BEV.GFP(图16)。                                                                                 
           
 
Finally, the plasmid that troubled free translation tools is rendered the right way.
 
But, have you noticed one whole sentence is missing? Human errors in translation, is indeed, not uncommon.
 
We can see that popular generic machine translation system are problematical, and sometimes even human translation are not infallible.
 
A better machine translation result can be obtained using a customized, domain specific, machine translation engine. Here is the sentence translated using a custom MT trained with Chilin bilingual sentence pairs from the pharmaceutical domain. Do you see the improvements?
 
Original English Text
 
GFP (Figure 16) was constructed by cloning the GFP fragment from pCR. Bgl-GFP-Bam as a Bglll/BamHl fragment into BamHl-digested pCMV. BEV. 2. This plasmid serves as a control in some experiments and also as an intermediate construct.                                                                                                                           
 
Custom Machine Translation
 
 
通过从pCR克隆GFP片段来构建GFP(图16)。将Bgl-GFP-Bam作为BgIII/BamHI片段插入BamHI消化的pCMV.BEV.2.中。质粒在一些实验中作为对照,也作为中间构建体。                                                                                                                                           
           
 
How do you build a custom machine translation engine? 
 
Please see the next Blog: Chilin Solution
 
A.Chan & L.Caddy