What is a Translation Memory?

A translation memory (TM) is a database of previously translated content. Whenever text is translated and confirmed as correct, the original source text and translated target text are stored in your company’s translation memory.

Translation Memory 2@3x

What is a Translation Memory?

A Translation Memory system (often referred to as a TM) is a vital piece of translation technology. It is a database built upon a company’s lexicon of previously-translated text copy. This ever-growing database contains sentences, paragraphs, and phrases that a linguist previously translated and confirmed.

Translation memories, powered by translation agency management software (like Lilt’s platform), reference the database to automatically translate text segments when a translator composes a new document. The core function of TM translation is to streamline the translator’s workflow. By creating a bank of words and phrases, you will reduce the redundancy of translating the same text repeatedly.

When you incorporate translation memory technology into your system, you will enhance the translator experience, increase efficiency, and reduce company spending.

Translation memories aren’t static, monolithic databases. Companies will create separate TMs for each language pair, content type, use case, and more. Since each of those TMs is associated with its own specific project, the TMs grow and become more useful for future projects as translation work is completed.

According to a recent CSA Research study, most translators that rely on modern tools use TMs in their day-to-day workflows, with only 12% saying that they never use them.

How does it work?

In its most basic form, translation memory technology is a dictionary of translation units stored as individual files. Translation memory software creates these ‘units’ (phrases, words, or sentences) out of the ‘source’ (the phrase in its original language) and the ‘target’ (the phrase in its translated language.)

Translation memories start off in one of two ways - from existing resources or empty. Empty TMs build a database of translated content by continually taking in each newly translated segment, word by word and sentence by sentence, over time.

TMs are accessed through computer-aided translation CAT tools, also known as machine-assisted or machine-aided human translation tools. However, translation memory is unique from completely machine-based translation systems. While machine technology uses neural networks to translate an entire source text into a target language based on predictions, TMs automatically translate specific text segments based on your company’s confirmed database. This difference is crucial: your copy won’t sound like a robot wrote it.

When a translator inputs the source language, the CAT tool assists by either:

Automatically translating to the target language using the translation memory

Using a modified translation, with the help of a translator

Having the translator create a new reference

In order to determine whether your content can be auto-filled or needs the eyes of a translator, the TM compares it to your database and ranks any matching phrases with a percentile.

If a sentence perfectly matches one that is stored in the translation memory, it will receive a 100 percent match rating. However, even if a sentence differs by one character, the maximum rating it can achieve is 99 percent. If a sentence contains similarities but falls below a 100 percent match, it is called a fuzzy match. A fuzzy match includes 75 to 99 percent of the source text. If a sentence falls below a threshold of 70 percent, it’s not a qualified match, and therefore, will not automatically translate.

Translation Memory and Translators

Translation memory is designed to easily translate any content created from repetitive blocks of text, such as technical documents. A translation memory system would be extremely useful for localizing legal and financial documents, where phrases and jargons have exact meanings.

Many languages include expressions and colloquialisms; these commonly-used phrases work well with TM translation. Additionally, many e-commerce companies regularly reference certain specs or product information. These repeated words (like colors or sizes) are also easily translated using these systems. However, some passages don’t translate well with machine technology. This technology replaces the simple words while allowing the human translator to focus on the words that matter.

Translation memory tools help linguists maximize their time by focusing on more abstract concepts rather than simple words and phrases. This way, your company has copy that reads fluidly, and your users get the best experience possible.

Creating your Translation Memory

Depending on a company’s needs, a translator memory is created either from scratch or from an existing resource. If you are building a new TM from the ground up, you start with an empty translation memory. Creating a database from an empty TM requires a long period of growth, but this process results in translations that are specific to your company’s brand voice and content.

As the database continually adds new translation pairs, it stores the units (the source text and its target text) for future use. These units now form reference points in the translation memory that correlate with a specific sentence or phrase. When you’re translating new copy, these pre-arranged units will automatically fill in so that your target replaces the source text.

If you’d prefer to get a jump start by building your database off an existing resource, you can do so easily by converting units to the new file management system.

The Technology Behind Translation Memory Systems

Translation memory functions by analyzing sentence pairs from the source document and matching them with the database’s lexicon. While many versions of translation memory systems exist, technological advancements are trending towards AI and automation utilization. These technologies allow the program to sequence sentences faster and more efficiently.

CAT tools, such as the one powered by Lilt, allows companies to achieve localization goals faster, more accurately, and more affordably. Instead of machine translation, human translators use CAT tools to aid them in their work. By enhancing translation solutions with automation, Lilt offers enterprise-level systems that conform to a business's existing workflows.

Automation Tools
One of the key features behind Lilt’s translation memory is that it automatically updates the translated segments in the platform’s CAT tool. The linguist confirms sentences and phrases. Then, through auto-propagation, the entirety of the document is scanned, and exact sentence or segment matches are automatically translated. When sentences have a “fuzzy match” (75- 99% match), translators must check these potentially-accurate translations. This aspect speeds up projects while still maintaining the highest standards.

Predictive Translation Suggestions
Maximizing a translator's time and efficacy has never been easier, and TM reduces the overhead needed to localize a document. By integrating the Lilt Platform, a translator has access to AI and automation so they can quickly process more text with accurate results. As the translator relies on the predictive translation suggestions from the TM, their speed skyrockets. The Lilt platform delivers more reliable translations using a company’s existing text content, previous translations, and asset base. As more text is entered into the database, predictions become more accurate, and documents are translated more quickly.

Lilt’s Adaptive Neural Machine Translation Engine
Lilt’s translation memory suggestions are provided by our Adaptive Neural Machine Translation (NMT) engine. One of the core concepts behind AI-powered technology is its ability to continually learn and advance its functionality, and this gets put to the test with the NMT. The NMT gets trained and refined with your organization’s language to improve its predictive models. Therefore, this platform continually adapts to new linguistic environments with every translation unit added to the database. Built-in collaboration features bring this beyond machine translation: a community of translators can query the group and compose more accurate transitions based on syntax and context.

Additional Features
Along with translation memory, the Lilt platform also utilizes a termbase (a database containing terminology and concept-oriented entries.) This feature improves accuracy and consistency when translating context-based phrases in a business’s document.

Why Use Translation Memory Software?

The basic premise behind adopting translation memory is self-evident. It creates channels that reduce time and effort for the human translator. Many text documents are composed of repetitive sentences and phrases; it would be a waste of resources to rely on the translator to continually revisit the same sentence. You want to put the talents of the linguist to good use.

By integrating TM software, an enterprise can:

Develop a consistent branding voice: As the same string of words frequently shows up in the text, translation memory ensures that a consistent translation is used across the document or copy.

Use a single platform for multiple users: It’s likely that a company, large or small, will require the talents of multiple translators. Translation Memory software offers a centralized database that multiple users can reference, constructing a unified language throughout the company.

Reduce overhead : When it comes to localization, time is money. Translation memory exponentially speeds up the process of translation documentation so that you can maximize on its ever-growing database. As the lexicon grows, your need for continual translation decreases. Now, your translators are freed up to tackle the more complex phrases in your copy.

Become more adaptable: As translation units are continually revisited and refined by editors and translators, your brand can modify its voice and copy with ease.

Translation Memory Management

Translation memory software is customized for each company; no two databases are the same. Languages have nuance: you can find different intents and meanings behind words depending on the context in which they are used. A single sentence can hold many meanings. By incorporating translation memory into your company’s workflow, you can reach more clients while ensuring your words ring true to your brand.

This is why translation memory management is critical to a company’s operations. With all translation units stored in a company’s personalized translation memory, routine management must ensure a consistent tone across all localization procedures.

Managing Your TM
Each translation unit can be tailor-made. In the Language Assets interface, you can find a compendium of all translation memories. Here, you can import and export translation unit files. When a new sentence is added to the information storehouse, the database queries it against existing entries, displaying the results to the translator. Translation units that come back as context matches (a 100 percent match) are added to the system. Any fuzzy match (less than 100 percent) will be manually amended by your translator. This automation turns your database-building into a simple endeavor.

You can also import or export existing language assets to build your TM. This system saves time and effort when creating new localization content, so you don’t have to start with an empty database.

Translation Memory Alignment
Translation memory alignment is a useful tool that turns your previously-translated documents into translation units. It provides a quick solution for translators working with multiple languages. By importing two separate languages into the system, translation memory alignment automatically pairs them into translation units.

TM alignments eliminate the need to build a database from scratch. Simply use your pre-existing documents, and the alignment function will compare source and target language files. Then, it will then either 1) import them into a new or existing TM, 2) combine sentence segments, or 3) remove sentence segments.

With translation memory alignment, even an empty database can be quickly filled and immediately usable.

Translation Memory vs. Machine Translation

It’s easy to assume that translation memory and machine translation are interchangeable. They both incorporate computer technology to decipher one language into another. However, each offers a unique aspect of the localization process.

Modern-day machine translation is simply the process of converting one language to another. It relies entirely on AI and automation to acquire new data in order to parse sentence syntax. Unlike TM, machine translation does not provide percentages for matches, so the reviewer must judge the translation quality. While machine translation has its uses, you can often end up with copy that doesn’t have the flow of normal, human language.

Translation memory works on a different structure. Translation memories use translation units to automatically convert predefined segments, which have been approved by specialized linguists.

One of the main advantages that Lilt’s translation memory has is that it becomes more advanced as users add to the translation memories and termbases. When content is created through the Lilt platform, the database is automatically updated. You can translate with confidence, knowing that your translation unit reference has already been verified for accuracy.

What Can Translation Memory Be Used For?

Translation memory can be used for all of a company’s translation needs to enhance its users’ experience.

Ensuring Brand Consistency
Regardless of content or application, translation memory can ensure that a company’s branding and voice are consistent across its channels or internal office exchanges. With the CAT tools built into the Lilt Platform, translators can maintain cohesion between multiple translation memories, termbases, and style guides.

Improving Company Efficiency
The features found in Lilt’s CAT program showcase its effectiveness for translating documents through its Translation Memory Leverage. This function analyzes and measures the reliance that a translator attributes to the TM. When a company first incorporates translation memory into its localization workflow, the leverage will report low reliance. But, as more and more translation units are added to the database, the leverage of the TM will increase dramatically.

Lilt’s translation memory is designed to work in tandem with our adaptive machine translation operation. When both systems work concurrently, they learn and advance at exponential rates.

As modular content is on the rise, the need to parse sentences down into smaller segments is becoming more necessary. Businesses and organizations rely on Content Management Systems (CMS) to manage their information. As portions of text are extracted from documents (rather than the whole document), translation memory makes the process faster and more economical.

Streamlining Localization to Improve the User Experience
Translation memories are the premier solution for every company's localization needs. As global communication continues to expand, effective communication across multiple languages is vital to removing barriers. By instituting a translation memory into translators’ toolsets, they can instantly increase their productivity while providing a consistent standard of quality.

To learn how a TM can help scale your localization programs, contact us today.