Arabic Machine Translation: Challenges, Progress And Best Practices
The Arabic language is rapidly gaining prominence across industries, and professional Arabic translation services are becoming indispensable for successful business expansion. But how does Machine Translation (MT) fit into this strategy?
In our blog on which MT-Engine works best for your content, we have already expanded on the best practices to select your engine and determined that language pair is the ideal measure to start from. But what does this mean for Arabic Machine Translation specifically? What are the challenges? The progress? And what should your strategy look like? Read all about it!
Arabic And Its Challenges With Machine Translation
Arabic, the most widely spoken of the Semitic languages, is a morphologically rich language (MRL) that is written from right-to-left (RTL). MRL shows its grammatical relations by changes to the words themselves rather than the relative position or addition of particles. This means that the MT-Engine, not yet highly context-aware, will not always be able to pick up the exact meaning of each word.
Research has shown that when it comes to MRL such as Arabic, morphology-aware preprocessing and pretrained embeddings will be able to contribute to far better translations.1 And despite the fact that Arabic is one of the fastest growing online languages,2 developing engines to meet these criteria is a fairly recent undertaking.
According to research by the University of Jazan,3 Arabic poses unique challenges when it comes to machine translation. The most common difficulties are related to:
- Long and complex sentence structures
- Unique word order and grammar make it hard for the machine engine to recognize the true meaning of the text.
- Words can carry multiple meanings and connotations
- Some letters of the Arabic alphabet have no equivalent in English
An additional challenge arises when it comes to an MT’s inability to make a reliable cultural and contextual interpretation of an Arabic text, or indeed any text. Cultural references and nuances of the language will be hard for an engine to pick up on since, while researchers are active in this, MT-engines are not yet context-aware.
That being said, there are ways for Arabic Machine Translation Strategies to still come up with great results, and that is if you combine MT with the Post-Editing Process. Post-Editing, the last part of the Machine Translation Post Editing (MTPE) Service, refers to the linguist expertise that is applied after the MT has produced the raw output. A native linguist specifically trained in the Post-Editing Process will proofread and edit the text and produce a qualitative piece of content.
So how impactful is the work of Post-Editors? Let’s have a look!
Why Human Expertise Remains Essential: The Value Of Post-Editing
Post-Editing is a crucial part of the MT process. Grammatical errors will be weeded out and cultural literacy and credibility will also be boosted. Additionally, Post-Editing will ensure that:
- The contextual mistakes we have already touched upon will be spotted and corrected.
- Real-World Knowledge and culture will be inserted into your content. Your content should be translated for the country and region, not just the language.
In order to reap these benefits, make sure to rely on an LSP that works with native, in-country experts with experience in your specific industry. Because when it comes to Arabic, it is worth remembering that there are a lot of different dialects that are often not mutually intelligible, and that the translator should reside in the specific region you are targeting.
Even the tone and formality can differ depending on the purpose of your content. If you are looking for a translation of media formats such as informational web content or newspapers, for example, Modern Standard Arabic (MSA) is the preferred form. However, in terms of creative content that seeks to engage – think verbal communications, marketing, and advertisement – MSA will be less appropriate. In this case, country-specific as well as regional dialects, will need to be taken into account.
Pro Tip! Want to know more about the ins and outs of MTPE? Have a look at our detailed blog “How Is Machine Translation Post-Editing (MTPE) Best Executed?â€
Which Machine Translation Engine Works Best For Arabic?
Don’t be too disappointed, but we won’t be giving you an actual engine you can work with consistently! While there is data available on which engine works best for each language pair, this is very changeable. Machine Translation is a growing market4 and is projected to reach $983.3 Million by 2022, more than double of what it was in 2016. To keep up with the growing demand for fast and accurate translations, MT-Engines are upping the game and reinventing themselves constantly. And that’s a good start.
Annual reports show changes in the best Engine for Arabic Machine Translations between 2019 and 2020. While in 2019 Modernmt was judged to be the best MT-Engine to use,5 in 2020 Systran PNMT and SDL BeGlobal were deemed more suitable.6 Even though the most compatible engines for the Arabic language can change rapidly, we are not saying you should go out and get a new one every year. We do recommend that you select more than one engine, just so you can make an informed decision and then perform consistent testing and training. But that’s something we’ll get back to shortly!
Is There An Optimal MT-Engine Specifically For My Sector?
There is some good news when it comes to MT-Engines and their compatibility with sectors. While the optimal way to assess an engine’s compatibility for your purposes is still the language pair, there is progress being made in terms of content and industry compatibility.
MT-Engines are getting smarter, and while contextual accuracy is still a while away and human translation remains paramount, it can be beneficial for you to dive into some data out there and see which engines have some resources available for English to Arabic Arabic to English translation. The above-mentioned 2020 report reveals the sectors for which there are resources available in MT. As you can see, English to Arabic already has some data available on the best MT-Engine per industry, which shows that there is already data collected for engines compatible with the Computer Hardware & Software, Energy & Water, Finances and Business.
Again, we’d like to stress that these engines might perform better in the industries, but that will only mean that the Post-Editing Process could be reduced, and time and money could be saved.
Why You Should Commit To Consistent Training And Testing
Once you have made a selection of 2 or 3 Arabic Machine Translation Engines, run a sample text through the engine. Select around 50 pages to translate to get a good overview of the performance, and compile a performance report that will inform you on the:
- Information Quality where you confirm factual and structural accuracy.
- Linguistic Quality, which includes grammatical errors, syntactical errors, and unnatural sounding language.
- Translation Quality where the general accuracy, consistency, completeness, and precision of the output is assessed.
- The Purpose of the content should also be assessed as you look into how the translation fits your content goals. As we mentioned before, some engines could work better for certain industries.
There is a chance that multiple MT-Engines will produce similar results. In that case, weigh out your options and select the engine that meets your goals best, whether it is more profitable or more productive. If you keep testing and training your engine, it will become smarter based on your specific needs and desires. So, in effect, any engine that may be reported as being superior the following year may just not be that way for you, if the investment in training and development has been targeted and consistent.
Conclusion
Arabic Machine Translation is still in need of optimization, but progress is on the way. While good MT-Engines can be identified at this stage, based on the language pair and, to some degree, industry, the use of Post-Editors and the commitment to consistent training and testing remain paramount.
At Laoret, we are fully equipped to deliver high-quality MT and MTPE Solutions through our technological authority and translators who are highly trained in the Post-Editing Process. Our Arabic linguists are native, in-country experts with subject matter expertise in your industry. Benefit from our 24/7 availability and reach out to us about your next Arabic MT Project!
