Using TAUS QE - Off The Shelf

At TAUS we have created an effective and user-friendly solution for anyone searching for reliable QE services: EPIC API. All you have to do is obtain an API key by signing up for a free trial, or purchasing a subscription. Now you are all set to start improving your translation workflows.

Sounds easy enough, but how does it actually work?

In this guide, you will find out exactly how to use our QE service for your specific wishes, no technical knowledge or software installations required! Whether you just want to try out a few samples without fuss, or you aim to integrate EPIC API into your current pipeline.

We offer both no-code and high-code options for any use case you might have. Read on and start estimating the quality of your translations today.

Why EPIC Delivers Accurate and Reliable Translation Quality Scores

Before we get into the nitty-gritty details, you might be wondering why you should be using EPIC API in the first place? Here’s what makes TAUS QE perform so well.

Firstly, we are lucky to have obtained lots and lots of data over the years. Our database contains 7.5 billion words in over 500 language pairs, to be exact. This means that you don’t have to worry about scrambling enough data together for accurate results. We have already done that work for you.

Additionally, we have created our own benchmarking techniques that make sure our models perform as closely to human reviewers as possible, without the bias and subjectivity that human evaluations often contain.

And last but not least, TAUS has a truly neutral position in the market as we don’t offer any translation services. We don’t benefit from high or low scores, so you can trust our QE scores to be truthful.

Key API features That Enhance Translation Workflows

These are some of the features that make EPIC API stand out when it comes to user-experience and efficiency.

• Full control over metrics
  For each request, you choose which of the available metrics you want to use. You can retrieve a more traditional QE score which tells you the overall quality of your translation, or opt for a linguistic metric that judges your data on aspects such as grammaticality, fluency and naturalness.
  
  
• Automatic (custom) model switching
  If the metric is not specified, the default TAUS QE score is used. This saves you the trouble of having to specify a metric each time. As a bonus, if you have a custom model, this model will be used automatically whenever the corresponding language pair is scored. No further changes to the request needed.
  
  
• Option to upload a file or plain text
  Sometimes you only have a few segments that you input as plain text, sometimes you have an entire file of samples to score. EPIC API supports both datatypes seamlessly.
  
  
• Automated Post-editing (APE)
  We don’t just offer QE scores here at TAUS. If any of your segments are not up to your standard, you can have them instantly edited by our fine-tuned LLM’s. This reduces the need for any manual corrections. Simply add a threshold to your request. Any segments that are scored below this threshold will be automatically refined. 

Things You Need For Your Request

Before you can start using the API there’s a few things you need to collect. After obtaining these five things, you are ready to make a request!

API key
You can find this on your account page after you’ve applied for a free trial or subscription.

Post url
This is a link that you need to add to your request so it gets sent to the right location. You will find the urls in the TAUS API documentation at developer.taus.net. You can make two types of requests depending on your data type. You can either attach a file, or send your samples as plain text. Each request type has its own url. On the documentation you will find which url is needed for your case.

Metric
In the TAUS API documentation you will find all available metrics you can use to estimate the quality of your translations. Choose a metric to add to your request. If no metric is specified, our default model TAUS QE v2.0 will be used. If you have made a custom model with us in a certain language pair, this model will automatically be used.

Data
Your data can be a .tsv file with two columns containing source and target translations, or it can be plain text in the format below. If you want to specify a metric, please refer to the documentation at https://developer.taus.net for more examples on how to add this to the data format.

{
    "source": {
        "value": "This is a test.",
        "language": "en"
    },
    "target": {
        "value": "Das ist ein Test.",
        "language": "de"
    }
}

 

APE Threshold
If you want to use the Automatic Post-editing service, please contact us so we can activate it for your API key. Then all you have to do is add a threshold to your request. Optionally you can also add a low threshold.

In the example below, all segments that have a score lower than 0.9, but higher than 0.3 will be edited by our specially trained LLM’s. In the response, you will see the QE score of the segments, and a revised version of the underperforming translation. Please refer to the documentation for examples on what a full request including APE would look like.

"apeConfig": {
     "threshold": 0.9,
     "lowThreshold": 0.3
}

How to Get Started with TAUS EPIC (No-Code Setup Guide)

Perhaps you want to try out EPIC API yourself before sending it over to your technical team, or maybe you don’t feel like starting up your programming software just for a quick test. Either way, we got you covered!

The easiest way to test out the API without any coding is to try out our demo. Just upload a file and get scores back instantly. However, the demo doesn’t offer the full capabilities that you can get by using the API directly. 

Want to have full control over the metrics and thresholds, but without all the technical hassle? Postman offers a handy User Interface that allows users without any technical background to send API requests.

Here’s how to use it:

Step 1 - Create free Postman account

Go to postman.com and create a free account. This allows you to send and receive unlimited requests. Click on the + symbol to create a new request.

Step 2 - Choose request type 
Refer to the TAUS API documentation and choose the right url for your data, either plain text or file upload. You can find the urls on the right-hand panel of the website. Scroll down until you see a dropdown box with the word POST on it. This one is for plain text uploads. If you scroll even further down you will find the one for file uploads. Click on the correct dropdown for your data and you will see both the production and sandbox urls.

For example, the sandbox url needed for a plain text POST request is: https://api.sandbox.taus.net/2.0/estimate/.

Step 3 - Add url and API Key
Go to your Postman request page, add your url to the url box, and make sure the request is set to POST (instead of GET). Now you can add your api key by clicking on “Authorization” and choosing API key as the Auth Type. Type “Api-Key” in the Key field, and paste your key in the Value field.

Step 4 - Add data as plain text 
(If you want to upload a file containing translations, skip this step and go to step 5.)

Click on the “Body” tab and choose “raw” as datatype. Make sure the format is set to JSON in the dropdown menu right next to it. Now in the text field, paste any example data you find on our documentation or input your own data in the same format.

Step 5 - Upload data as file
Click on the “Body” tab and choose “form-data” as datatype. In the Key field type ‘file’, and in the dropdown right next to it, choose “File”. In the corresponding Value field, you can upload a file from your computer.

In the following Key fields type “source_language”, “target_language”, “metric_uid” and “metric_version”. In the corresponding Value fields type the right value. In the example image below you can see what this should look like.

Step 6 - Send your request
Click send and receive your QE scores within seconds! You will find the response containing your output underneath your request. In the dropdown menu you can choose which format you want to have your response in: JSON, HTML, XML or JavaScript. Click the three dots in the right corner to export your results to a file. 

Interpreting Results

Now that you’ve made your request and retrieved the results, you might be wondering what the scores mean? The TAUS QE models return a score from 0 to 1. The closer to 1, the better your translation is. The exact borders between good and bad translations are subjective, and depend on your use case and domain. It is up to you and your team to decide on the threshold between good and bad translations. However, we do offer a standard scheme to help you get started.

Here’s a quick guideline on how to interpret the TAUS QE scores:

• Above 0.9: good translations
• 0.88-0.9: gray area (can be good, might have issues)
• Below 0.88: usually indicates at least minor errors
• Below 0.8: serious errors
• Below 0.7: poor quality

Want to get more insights on QE scoring and interpreting results? Feel free to check out our free QE scoring guide with real-life examples and best practices here.