Terminology management between human skills and AI
In today's global market, consistent and accurate communication is vital for maintaining your brand identity, meeting industry standards, and connecting with diverse audiences. Terminology management ensures your business speaks with one clear voice across languages and regions. At the same time, it ensures the terms you use are not only consistent but also clear to your target audience and appropriate to their cultural and professional context. Terminology management is the backbone of consistent and high-quality translations.
One crucial step in terminology management is terminology extraction—the process of identifying key terms from source texts or translations. These terms typically include industry-specific jargon, product names, abbreviations, or expressions used frequently in your company and that require consistent translation. Once identified, they are categorized and validated to create a glossary or termbase that ensures clarity and uniformity in communication.
Once integrated into Computer-Assisted Translation (CAT) tools, the glossary serves as a linguistic guide, enabling translators to maintain consistency, accuracy, and efficiency across all translation projects.
What Are the Main Approaches to Terminology Extraction?
There are several methods of terminology extraction, ranging from manual extraction to NLP or AI enabled tools. Each approach has its strengths and drawbacks.
1. Manual Extraction
Manual terminology extraction relies on skilled linguists or subject-matter experts to carefully analyse your texts and identify key terms. This process often involves reading the source material, flagging domain-specific terms, and compiling them into a glossary or termbase. The advantage of this method is that a human expert can ensure a precise identification of terms, taking into account nuances and polysemy (words with multiple meanings).
This is however very time-consuming and can be performed on a limited amount of documents. Scanning texts line by line is labor-intensive, particularly for long or complex documents. A single glossary might take hours to compile and would be in most cases based on just a few documents. Additionally, there is also the risk of human error, such as missing key terms or inconsistent selection.
2. Extraction with NLP Tools
Natural Language Processing (NLP) tools use algorithms to scan texts, detect frequently used words or phrases, and identify potential terminology. These tools often rely on statistical methods, pattern recognition, or predefined rules to extract terms and got more and more sophisticated over time. In the past, they used to extract terms on the basis of their frequency of occurrence within texts. This is however not always a reliable criterion as the words that occur more frequently in texts (e.g. articles, prepositions and conjunctions) do not really represent your corporate language or domain.
Modern tools use more advanced techniques and can, for example, compare the frequency of occurrence of all the words in a specialized text with their frequency in large non-specialized corpora. By doing so, they identify the terms that are likely candidates for a specialized glossary.
NLP tools can analyze thousands of words in a fraction of the time it takes a human. For instance, a legal contract spanning 100 pages can be processed within minutes, highlighting potential terms like "indemnity" or "force majeure". Many NLP tools also excel at identifying multi-word terms, like "artificial neural network" or "supply chain optimization", based on frequency and collocation patterns.
At the same time, the output still needs some validation and refinement by human experts to ensure accuracy and eliminate redundant or irrelevant terms.
3. Extraction with Large Language Models (LLMs)
LLMs, such as GPT-based models, use advanced deep learning algorithms trained on vast amounts of text data. These models can process input texts and suggest domain-specific terminology. Users can provide specific prompts, such as "extract technical terms related to renewable energy", to guide the process.
While powerful, LLMs are trained to always generate an output, regardless of how meaningful it is. In terminology extraction this can lead to overly broad results or to the suggestion of terms that do not appear at all in the texts submitted to the tool. As a consequence, results might be less accurate than the ones obtained with NLP tools.
4. How We Do It Right for You
At tolingo, we combine human expertise with state-of-the-art tools to deliver terminology management tailored to your needs. Here's how we make it seamless:
- Provide us with your source texts or previously translated documents.
- We extract for you the relevant terminology using a manual or semi-automatic approach and based on the use cases.
- Once the glossary has been validated by our linguists or by you in the required languages, we integrate it into our tools.
- We implement automatic QA checks to ensure your glossary is consistently applied in translations.
Terminology management may sound complicated, but we'll help you put it in clear terms!
