Edit LQA models

The quality assurance model is mostly about categorizing the mistakes that a translator or an editor may make.

You can set up the error categories on the QA model tab.

The main part of the tab is the list of categories. Each item of the list has a main category, a subcategory. An error category can have one or more severity levels.

The example above shows the categories of the LISA model, which is built into memoQ (among many others).

A QA model may or may not use severity levels, and it may or may not use penalty points.

When a document is graded with LQA feedback, the penalty points for each error are added up. The higher the sum, the worse the perceived quality of the translation. (Don't forget, the error categories are entered by a human reviewer!)

• If a QA model doesn't use severity levels, but it uses penalty points: Each error category has exactly one penalty value.
• If a QA model uses severity levels, but it doesn't use penalty points: A severity may or may not be used for each error category.
• If a QA model uses both severity levels and penalty points: For each error category - severity level point, you need to choose a penalty value.

  To add a category:

1. Below the list, click Add. A new row appears at the bottom of the list, with a yellow background.
2. In the Category cell, type a main category name. Press Tab.
3. In the Subcategory cell, type a subcategory name.
4. The next one or more columns are about grading the error.
• If a QA model doesn't use severity levels, but it uses penalty points: There is one column. Type a penalty value there.
• If a QA model uses severity levels, but it doesn't use penalty points: There is a column for each severity level, with a check box in each cell. Normally, memoQ uses all severity levels for a new category. Clear a check box if you don't want to use that severity level for the new category.
• If a QA model uses both severity levels and penalty points: There is a column for each severity level. Type a penalty value in each column.
• If a QA model doesn't use severity levels nor penalty points: A category has the category name and the subcategory name only. There aren't any other columns.
5. At the bottom of the table, you can type a description for the category you are editing.

• To change a category: Click a row in the table. Simply edit the cells you need to change. Press enter or click another row to save the changes.
• To delete a category: In the table, click the row you want to delete. Click Remove.

  To use penalty points: Check the The QA model uses penalty points check box.

  To use severity levels: Check the The QA model uses severity levels check box.

If you use severity levels, you need to list them at the top of the QA model tab, under Error severity. Severity levels are ordered from less severe to most severe.

• To add a severity level: Type a name in the text box below the list. Click Add.
• To change the name of a severity level: Click it in the list. Change the name in the text box below the list. Click Update.
• To remove a severity level: Click it in the list. Click Remove.
• To change the order of severity levels: Click the severity level you want to move in the list. Click Move Up or Move Down, until it gets where it belongs.

  To make sure the reviewer gives full feedback: In the translation editor, a reviewer may reject a row by pressing Shift+Enter. This is when they can enter QA category details for the row. Normally, this is enough. But you can force them to add a comment, explaining their feedback. To do this, check the Require commenting on errors check box. Then the reviewer can't save the LQA information without entering a comment, explaining of why this is an error. Use this if you suspect that the reviewer may make a lot of preferential edits, that is, edits that aren't required by the formal quality criteria.

Some automatic QA warnings may indicate a genuine translation error. You may want to include these warnings in the LQA feedback. You can include this on the QA mappings tab.

To map an automatic QA warning into a an LQA error category:

• At the bottom of the list, there is always an empty row.
• In that row, under Automatic QA type, choose a type of automatic QA warning.
• Under LQA error category, choose an error category. It must be listed on the QA model tab (see the previous section). You can add a category but not a severity here.

  To remove a mapping: Click a row in this table. At the bottom right, click Remove.

When all the LQA feedback is there for a document, memoQ can grade it - whether it's Pass or Fail. You can set this up on the Pass/Fail criteria tab.

If the LQA model uses penalty points, the Pass/Fail criteria tab looks like this:

memoQ uses the penalty points to check if a document fails or passes. But memoQ doesn't calculate the actual sum of penalty points. Instead, you can use the density of the errors - that is, the penalty points per word - to determine pass or fail.

Normally, penalty points are calculated per 1000 words. The idea is that the number of errors is usually smaller than the number of words in a document. If a document has 347 words, and the LQA returns one penalty point, the normalized value is 2.8818 (error count/ word count*1000, which in our case is 1/347*1000 = 2.8818).

You have two options to decide if a document passes or fails:

• To fail if there more errors per 1000 words than you want: Click the There are more than p penalty points per w words radio button. Then, in the p= number box, type the maximum number of penalty points you tolerate within 1000 words. You can use a different unit size. For example, you may look at the number of penalty points per 100 words, or per 10,000 words. To set the unit size, type a different number in the w= number box.
• To fail if the document is not error-free as much as you would like: Click The normalized score (1-total points/total words) is lower than s radio button. Here, memoQ calculates a number similar to a percentage. This shows how error-free the document is. Normally, memoQ uses 0.90 (90%) here. This doesn't mean that 90% of the document is error-free, but the number of penalty points is less than 10% of the word count. It's still a model of that in a way.

  When the QA model doesn't use penalty points, the Pass/Fail criteria tab looks like this:

In this list, you can add conditions. If one of the conditions is true, the quality of the document Fail.

To add a condition:

1. In the last (first empty) row of the table, type a number of errors.
2. From the Error type drop-down box, choose an error category or a severity level. So, for example, this can be Minor or Terminology - Glossary adherence. It must be listed on the QA model tab.
3. in the Number of text units box, type a whole number.
4. From the Text unit drop-down box, choose a unit (Word, Segment, Document).

   This means that memoQ counts the number of the errors of that type for every text part.

For example: Let's fail if there are 20 minor errors per 3000 words or 3 major errors per 3000 words. This is two rows in the table: One for the minor errors, and another for the major ones:

If you check the Count multiple occurrences of the same error as one check box, memoQ will count every occurrence of an error as one. This helps the translator to correct their translation. This doesn't mean that all occurrences of the same error category are counted as one: but that all occurrences of the same text that has the error are counted as one.

The LQA model and feedback is exported with an MQXLZ file: When you export a bilingual document from memoQ, the LQA model will be embedded in it. When then another user imports this bilingual document into his memoQ project, this document will contain this QA model, even when the project on the other end uses a different LQA model.