What's Context-Sensitive Translation?

Our innovative solution, Context-Sensitive Translation, considers the context of a translation project. It goes beyond typical automation by utilizing AI to make decisions based on past translation content. It's different from all translation technology out there because it not only examines sentence structure (syntactic) but also comprehends its meaning (semantic). The article provides examples illustrating the impact of this approach in translation.

Context-Sensitive translation as BWX uses advanced technology, like Large Language Models, to handle data and give more accurate results. With this tech, when a translator is working, it checks if there's already a similar sentence in the translation memory. It also suggests things based on the glossary the translator has created. This helps make translations better and more consistent.

Do you know the difference between semantics and syntax?‍

A language can be organized in various ways to analyze its elements. Among these ways are syntax and semantics.

Syntax deals with how elements function in a sentence and how they relate to each other. It studies how the subject arranges information in a sentence to communicate in a specific language.

The rules of syntax depend on the language you are speaking or writing in. It can be more restrictive, or quite flexible. English word order follows the subject-verb-object sequence. It's usually the same for Portuguese, French, Spanish and some others. 

One of the main rules of Syntax in the English Language is that a complete sentence requires a subject and a verb to express a complete thought. As we see in the example below:

On the other hand, semantics refers to the meaning of words, texts and phrases. Language professionals use the acknowledgment about syntax to communicate with purpose.

To be a master in semantics requires a repertory of variations in meaning of a word in different contexts and the ability to persuade and create aesthetic effects, such in poetry. In translation, it is fundamental to producing the best meaning match. And this search is maybe the most fun part of translation. 

For instance, in Portuguese there is the word "saudade" to express when you miss something. There's no such word in english. But the best substitute in daily life is the verb "miss". "Miss something or someone..." You can also use "homesickness" to refer to when you miss home. In literature, you also find the term "long for", as stated below:

Our BWX performs a semantic analysis precisely because it evaluates the meaning of words within a context, allowing the translator to choose the best match, considering the translator's style, thematic references, and textual genre. 

The 3 pillars of integrated AI for translation‍

A CAT Tool, or computer-assisted translation software, assists translators by providing recommendations from three traditional sources: Translation Memory, Glossary, and Machine Translation.

Every source can be viewed as a good friend that brings valuable context about the previous translations. It's like reviewing your text in three collections, resulting in the best combination.

For Context-Sensitive Translation, each source is an important pillar of translation. Let's understand each of them.

1. Machine Translation

This pre-translation process involves automatic translations from platforms such as Google, DeepL, Microsoft, and similar sources. Let's see how our platform, also a CAT tool, handles Machine Translation?

As shown below, our platform displays the pre-translation in a preview along the screen, dividing the text into sentences. This setup allows the translator to focus on details, enhancing efficiency for subsequent processes. Take a look at the example below, featuring an article from our blog translated from English to Spanish.

Machine translation is valuable because it significantly speeds up translation work. However, it's a standardized process. What's the issue with using MT alone? It translates the same sentence in the same way every time, and you can't optimize or personalize the output. 

That's where our Context-Sensitive solution comes in. It's precious because every previous translation is taken into account. This means that once an error or a translation preference is identified, the suggestions are fed back as a pre-translation.

1. Glossary

Every glossary is a collection of terms. In this context, it represents a translator's vocabulary. With the glossary, when you change a word in the text, all subsequent words with the same meaning will also change. This wouldn't be possible just with MT.

The advantage? The translator no longer has to repeatedly fix the term and gains productivity time.

In the side panel of our CAT Tool, terms in the glossary are highlighted in yellow. See how the glossary works as a dictionary in the example: 

1. Memory Translation

Translation memory considers entire sentences from the text. It's not as rigid as the glossary. It learns and improves as translations are made. Therefore, it informs the translator about the proximity of translated sentences compared to what has been done before.

Translation memory works on a scale of 50% to 100% proximity. In the Bureau Works' BWX tool, you can choose your minimum threshold of proximity for translation. For example, choosing that when the proximity is equal to or higher than 80%, the translator will be informed that a match has been found!

With our Context-Sensitive solution, the translator gains in style, feeling they've put their touch in the translation. And you know how authorship is valuable for translators.

Testing with a document to translate from American English to Mexican Spanish, the CAT Tool indicated one match of more than 100%. Check the test below:

Why is AI a decision-making tool?‍

There are still prejudices surrounding post-AI editing. The main criticisms revolve around it being a way to save translation work, leaving everything up to the AI's responsibility.

But the truth is that these tools are allies. They organize the translation work. Simply put, AI consults information sources (machine translation, glossary, and memory translation) and delivers a translation based on these contexts.

The ideal scenario is to use AI to make a decision, not to translate. What does this mean? It means that the final word about all suggestions given by the tools will always belong to the translator. If the final refinement is human, the content delivered will also be.

‍

‍

‍

‍

‍