Search with complete suggester and german analyzer - elasticsearch

Question

I created a simple index with a suggest field and a completion type. I indexed some city names. For the suggest field I use a german analyzer.
PUT city_de
{
"mappings": {
"city" : {
"properties": {
"name" : {
"type": "text",
"analyzer": "german"
},
"suggest" : {
"type": "completion",
"analyzer": "german"
}
}
}
}
}
The analyzer works fine and the search by using umlauts is good. Also the autocompletion is perfect. But I faced an issue by searching for the term wie.
Lets say I have two documents Wiesbaden and Wien with the same name as suggest completion term.
If I searching for wie I assume that the cities Wien and Wiesbaden are in the response. But unfortunately I get no response. I suppose that wie has a restriction because of the german analyzer. Because if I search for wi or wies I get valid responses.
Same is for term was, er, sie, und which looks like stemming words in german.
Do I need any additional configuration to get also a result if I search for wie or was?
Thanks!

Answer 1

The problem
Searching city names by prefix
"wie" should find "Wien" or "Wiesbaden"
Possible solution approach
For the usecase I would suggest using an edge n-gram https://www.elastic.co/guide/en/elasticsearch/reference/current/analysis-edgengram-tokenizer.html and ASCII folding the terms https://www.elastic.co/guide/en/elasticsearch/reference/current/analysis-asciifolding-tokenfilter.html.
Example
wien
token position start offset end offset
w 0 0 1
wi 1 0 2
wie 2 0 3
wien 3 0 4
wiesbaden
token position start offset end offset
w 0 0 1
wi 1 0 2
wie 2 0 3
wies 3 0 4
...
wiesbaden 8 0 9
Keep in mind that the system has to work in a asymmetric way now. The query should not be analyzed (use keyword) but the data in the index has to be analyzed.
There are two ways to achieve this:
1.) Add the query analyzer to use the query
2.) Bind the query analyzer to the field
"cities": {
"type": "text",
"fields": {
"autocomplete": {
"type": "text",
"analyzer": "autocomplete_analyzer", <-- index time analyzer
"search_analyzer": "autocomplete_search" <-- search time analyzer
}
}
}
Why does the german analyzer not work
The analyzer is designed for german text and uses an easy algorithm to remove flection and morphology.
https://www.elastic.co/guide/en/elasticsearch/reference/current/analysis-lang-analyzer.html#german-analyzer
Here is an example for the typical terms generated by this tokenizer
Hallo hier ist der Text über Wiesbaden und Wien. Es scheint angebracht über Wände und Wandern zu sprechen.
hallo 0 0 5
text 4 19 23
wiesbad 6 29 38
wien 8 43 47
scheint 10 52 59
angebracht 11 60 70
wand 13 76 81
wandern 15 86 93
sprech
If it works on city names this happens just by coincidence.