add information about ICU normalisation

3.8/aql/functions-string.md

@@ -133,6 +133,20 @@ value used is 0xFFFFFFFF, and the final xor value is also 0xFFFFFFFF.
 CRC32("foobar") // "D5F5C7F"
 ```
 
+NORMALIZE_UTF8
+--------------
+
+`NORMALIZE_UTF8(text) → normalized`
+
+In Unicode there may be more than one representation of a glyph. 
+Use [ICU Normalization](http://www.unicode.org/reports/tr15/) in order to convert
+to a similar character. It may also be useful to run this normalization before 
+writing the documents into the database. If you have strings with different representations, 
+functions like [`FIND_FIRST`](#FIND_FIRST) may not find all matches you desire.
+
+- **text** (string): a UTF8-string
+- returns **nomalized UTF8-string** (string): the normalized string.
+
 ENCODE_URI_COMPONENT()
 -----------
 

3.8/indexing-index-basics.md

@@ -402,6 +402,19 @@ not be enabled for other types of queries or conditions.
 For advanced full-text search capabilities consider [ArangoSearch](arangosearch.html).
 {% endhint %}
 
+
+Indexes and non-ASCII texts
+---------------------------
+Before strings are put into an index, they are
+[normalized by using ICU](http://www.unicode.org/reports/tr15/). There are several characters
+in the Unicode space, that have a similar meaning. In order to have all variants of them
+in a result set when querying, the strings are normalized for the index.
+This slightly changes the behaviour of `FILTER` statements with `==` -
+comparisons when ran on non-indexed document attributes. While the index may still be useful
+by fetching a little more results then you want to actually work with, you may want to have an 
+additional `FILTER MD5(doc.attr) == MD5(@comparisonstring)` to make sure in the end result
+only contains the actual values you desire. 
+
 Indexing attributes and sub-attributes
 --------------------------------------
 
@@ -689,4 +702,3 @@ become unsustainable if this list grows to tens of millions of entries.
 
 Building an index is always a write heavy operation (internally), it is always a good idea to build indexes
 during times with less load.
-