[BUG] Token Filter Order: word_delimiter_graph and synonym_graph

[aswad1]

Describe the bug

I am migrating from Solr to OpenSearch and need to maintain the same analyzer behavior that I had in Solr, where the word_delimiter_graph filter is applied before the synonym expansion.

In Solr, this order worked without issues, and I used FlattenGraphFilterFactory to handle token graph flattening after synonym processing. I need to maintain this specific order to keep consistent search behavior during the migration. Any guidance or suggestions would be greatly appreciated.

Solr schema:

     </analyzer>
     <analyzer type="query">
        <charFilter class="solr.PatternReplaceCharFilterFactory" pattern="[({.,\[\]\“\”/})]" replacement=" " />
        <tokenizer class="solr.WhitespaceTokenizerFactory" />
		<filter class="solr.StemmerOverrideFilterFactory" dictionary="lang/stemdict.txt" ignoreCase="true"/>
        <!--Move ASCII folding, lowerCase, and Hunspell before synonyms so that clean singular terms are sent to the synonyms-->
     <filter class="solr.ASCIIFoldingFilterFactory" preserveOriginal="true" />
        <filter class="solr.LowerCaseFilterFactory" />
        <filter class="solr.WordDelimiterGraphFilterFactory" generateWordParts="1" preserveOriginal="0" catenateAll="1" splitOnCaseChange="0"/>
        <filter class="solr.HunspellStemFilterFactory" dictionary="en-US.dic" affix="en-US.aff" ignoreCase="true" />
		<filter class="solr.ManagedSynonymGraphFilterFactory" managed="english" />
        <filter class="solr.PatternReplaceFilterFactory" pattern="(-)" replacement=" " replace="all" />
        <filter class="solr.FlattenGraphFilterFactory" /> <!-- required on index analyzers after graph filters -->
        <filter class="solr.StopFilterFactory" ignoreCase="true" words="stopwords.txt" />
     </analyzer>
  </fieldType>

---

OpenSearch mapping, where I am facing the error for the analyzer "text_en_index" and "text_en_query":

PUT /my-index-10
{
  "settings": {
    "analysis": {
      "char_filter": {
        "custom_pattern_replace": {
          "type": "pattern_replace",
          "pattern": "[({.,\\[\\]“”/})]",
          "replacement": " "
        }
      },
      "filter": {
        "custom_ascii_folding": {
          "type": "asciifolding",
          "preserve_original": true
        },
        "custom_word_delimiter": {
          "type": "word_delimiter_graph",
          "generate_word_parts": true,
          "preserve_original": false,
          "catenate_all": true,
          "split_on_numerics": false,
          "split_on_case_change": false
        },
          "custom_synonym_filter": {
            "type": "synonym",
            "synonyms_path": "analyzers/F198001551",
            "updateable": true
          },
         "custom_hunspell_stemmer": {
          "type": "hunspell",
          "locale": "en_US"
        },
        "custom_pattern_replace_filter":{
          "type": "pattern_replace",
          "pattern": "(-)",
          "replacement": " ",
          "all":true
        },
        "custom_stemmer_override":{
          "type": "keyword_marker",
          "keywords_path":"analyzers/F225794029",
          "ignore_case": true
        },
		    "custom_synonym_graph_filter":{
			   "type": "synonym_graph",
         "synonyms_path": "analyzers/F3495229"
		  }
      },
      "analyzer": {
        "text_en_index": {
          "type":"custom",
          "char_filter": ["custom_pattern_replace"],
          "tokenizer": "whitespace",
          "filter": [
            "custom_ascii_folding",
            "lowercase",
            "custom_word_delimiter",
            "custom_hunspell_stemmer",
            "custom_synonym_graph_filter",
            "custom_pattern_replace_filter",
            "flatten_graph"
          ]
        },
        "text_en_query": {
          "type":"custom",
          "char_filter": ["custom_pattern_replace"],
          "tokenizer": "whitespace",
          "filter": [
            "custom_stemmer_override",
            "custom_ascii_folding",
            "lowercase",
            "custom_word_delimiter",
            "custom_hunspell_stemmer",
            "custom_synonym_graph_filter",
            "custom_pattern_replace_filter",
            "flatten_graph"
          ]
        },
        "text_id_tx_class_id":{
          "tokenizer": "whitespace"
        },
         "text_general_index_analyzer": {
          "tokenizer": "standard",
          "filter": [
            "lowercase"
          ]
        },
        "text_general_query_analyzer": {
          "tokenizer": "standard",
          "filter": [
            "lowercase",
            "custom_synonym_filter"
          ]
        },
        "en": {
          "tokenizer": "standard",
          "filter": [ "custom_hunspell_stemmer" ]
        },
		    "managed_synonym_analyzer":{
		      "tokenizer": "standard",
          "filter": [ "custom_synonym_graph_filter" ]
		}
      }
    }
  },
  "mappings": {
    "properties": {
      "id": {
        "type": "keyword",
        "index": true,
        "store": true
      },
      "id_tx":{
        "type": "text",
        "analyzer": "text_id_tx_class_id",
        "store": true,
        "index" :true
      },
      "description": {
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index" :true
      },
      "TM5":{
        "type":"keyword",
        "store": true,
        "index" :true
      },
      "status":{
        "type":"keyword",
        "store": true,
        "index" :true
      },
       "version":{
        "type":"keyword",
        "store": true,
        "index" :true
      },
      "class": {
        "type": "text",
        "analyzer": "text_general_index_analyzer",
        "search_analyzer": "text_general_query_analyzer",
        "store": true
      },
      "long class":{
        "type": "text",
        "analyzer": "text_general_index_analyzer",
        "search_analyzer": "text_general_query_analyzer",
        "store": true
      },
      "class_id":{
        "type":"text",
        "analyzer":"text_id_tx_class_id",
        "index":true,
        "store":true
      },
      "notes":{
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index" :true
      },
       "employee_notes":{
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index" :true
      },
      "editor_notes":{
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index" :true
      },
      "begin_effective_date":{
        "type":"date",
        "store":true,
        "index": true,
        "format":"strict_date_optional_time||epoch_millis"
      },
      "end_effective_date":{
        "type":"date",
        "store":true,
        "index": true,
        "format":"strict_date_optional_time||epoch_millis"
      },
      "goods_services":{
        "type":"keyword",
        "store":true,
        "index":true
      },
      "record_state":{
        "type":"keyword",
        "store":true,
        "index":true
      },
      "action_flag":{
        "type":"keyword",
        "store":true,
        "index":true
      },
      "creation_date":{
        "type":"date",
        "store":true,
        "index": true,
        "format":"strict_date_optional_time||epoch_millis"
      },
      "created_by":{
        "type":"keyword",
        "store":true,
        "index":true
      },
      "last_updated_date":{
        "type":"date",
        "store":true,
        "index": true,
        "format":"strict_date_optional_time||epoch_millis"
      },
      "last_updated_by":{
        "type":"keyword",
        "store":true,
        "index":true
      }
    }
  }
}

---

In OpenSearch, when I try to apply the word_delimiter_graph filter before the synonym_graph filter (as required by my use case), I receive the following error:

{
  "error": {
    "root_cause": [
      {
        "type": "illegal_argument_exception",
        "reason": "Token filter [custom_word_delimiter] cannot be used to parse synonyms"
      }
    ],
    "type": "illegal_argument_exception",
    "reason": "Token filter [custom_word_delimiter] cannot be used to parse synonyms"
  },
  "status": 400
}

---

Question:

1. Is there a known limitation in OpenSearch that prevents the word_delimiter_graph filter from being applied before the synonym_graph filter?
2. Is there any recommended workaround or configuration that would allow me to maintain the same filter order while avoiding the token graph parsing error?

---

Related component

Other

To Reproduce

N/A

Expected behavior

N/A

Additional Details

Plugins
Please list all plugins currently enabled.

Screenshots
If applicable, add screenshots to help explain your problem.

Host/Environment (please complete the following information):

• OS: [e.g. iOS]
• Version [e.g. 22]

Additional context
Add any other context about the problem here.

[nupurjaiswal]

Hi,
I tried the below approaches recommended by AWS support:

1. Tried flag “lenient: true” with synonym graph filter and word delimiter filter
   Error: "Token filter [custom_word_delimiter] cannot be used to parse synonyms"

2. Tried multiplexer filter as below:

"multiplexer_filter":{
    "type":"multiplexer"
"filters":["custom_word_delimiter","custom_hunspell_stemmer","custom_synonym_graph_filter"]

I was able to create the index, but while indexing the document I got the below error

 “raise HTTP_EXCEPTIONS.get(status_code, TransportError)(
opensearchpy.exceptions.RequestError: RequestError(400, 'illegal_argument_exception', 'Increment must be zero or greater: -1')”

which is already mentioned under: https://discuss.elastic.co/t/word-delimiter-graph-token-synonym-graph-token/278942

3. Tried multiplexer and added “lenient: true” in synonym graph filter as below:

"multiplexer_filter":{
    "type":"multiplexer"
"filters":["custom_word_delimiter","custom_hunspell_stemmer","custom_synonym_graph_filter"]
"custom_synonym_graph_filter":{
"type": "synonym_graph",
         "synonyms_path": "analyzers/F3495229",
         "lenient":true
		  }

I was able to create index and finish indexing, but when I tried querying I wasn’t getting the current result

Result:

POST /my-index-12/_analyze
{
  "analyzer": "text_en_index",
  "text": "note-book"
}

{
  "tokens": [
    {
      "token": "note book",
      "start_offset": 0,
      "end_offset": 9,
      "type": "word",
      "position": 0
    }
  ]
}

Expected result:

{
  "tokens": [
    {
      "token": "notebook",
      "start_offset": 0,
      "end_offset": 9,
      "type": "word",
      "position": 0,
      "positionLength": 2
    },
    {
      "token": "note",
      "start_offset": 0,
      "end_offset": 4,
      "type": "word",
      "position": 0
    },
    {
      "token": "book",
      "start_offset": 5,
      "end_offset": 9,
      "type": "word",
      "position": 1
    }
  ]
}

Please let me know if you have any other recommendation.

Thanks,
Nupur

[msfroh]

Hi there -- the issue here is that the synonym_graph filter will use the analyzer chain defined up to that point to tokenize the synonym file. You can't use a word_delimiter_filter to read a synonym file, because it outputs multiple tokens in the same position.

We could add a configuration setting to synonym and synonym_graph filters to allow you to specify a custom analyzer to use when reading the synonym file.

[nupurjaiswal]

Hi Michael,

Do you know how soon can this be implemented? Will I be able to specify the word_delimeter_filter before the synonym graph filter with this approach?

Thanks,
Nupur

[prudhvigodithi]

Hey, I was experimenting with the issue to understand the error better, and as suggested by @msfroh I've added a new configuration setting synonym_analyzer for synonym_graph and synonym (Draft PR with the change #16488).

Here’s the output I’ve observed.

Query

curl -X PUT "localhost:9200/my-index-15" -H "Content-Type: application/json" -d '{
  "settings": {
    "analysis": {
      "char_filter": {
        "custom_pattern_replace": {
          "type": "pattern_replace",
          "pattern": "[({.,\\[\\]“”/})]",
          "replacement": " "
        }
      },
      "filter": {
        "custom_ascii_folding": {
          "type": "asciifolding",
          "preserve_original": true
        },
        "custom_word_delimiter": {
          "type": "word_delimiter_graph",
          "generate_word_parts": true,
          "catenate_all": true,
          "split_on_numerics": false,
          "split_on_case_change": false
        },
        "custom_synonym_filter": {
          "type": "synonym",
          "synonyms": [
            "laptop, notebook",
            "smartphone, mobile phone, cell phone",
            "tv, television"
          ],
          "updateable": true,
          "synonym_analyzer": "standard"
        },
        "custom_pattern_replace_filter": {
          "type": "pattern_replace",
          "pattern": "(-)",
          "replacement": " ",
          "all": true
        },
        "custom_stemmer_override": {
          "type": "keyword_marker",
          "keywords": [
            "run",
            "running",
            "jumps",
            "quick"
          ],
          "ignore_case": true
        },
        "custom_synonym_graph_filter": {
          "type": "synonym_graph",
          "synonyms": [
            "laptop => notebook",
            "smartphone, mobile phone, cell phone => smartphone",
            "tv, television => television",
            "word processor => text editor"
          ],
          "synonym_analyzer": "standard"
        }
      },
      "analyzer": {
        "text_en_index": {
          "type": "custom",
          "char_filter": ["custom_pattern_replace"],
          "tokenizer": "whitespace",
          "filter": [
            "custom_ascii_folding",
            "lowercase",
            "custom_word_delimiter",
            "custom_synonym_graph_filter",
            "custom_pattern_replace_filter",
            "flatten_graph"
          ]
        },
        "text_en_query": {
          "type": "custom",
          "char_filter": ["custom_pattern_replace"],
          "tokenizer": "whitespace",
          "filter": [
            "custom_stemmer_override",
            "custom_ascii_folding",
            "lowercase",
            "custom_word_delimiter",
            "custom_synonym_graph_filter",
            "custom_pattern_replace_filter",
            "flatten_graph"
          ]
        },
        "text_id_tx_class_id": {
          "tokenizer": "whitespace"
        },
        "text_general_index_analyzer": {
          "tokenizer": "standard",
          "filter": [
            "lowercase"
          ]
        },
        "text_general_query_analyzer": {
          "tokenizer": "standard",
          "filter": [
            "lowercase",
            "custom_synonym_filter"
          ]
        },
        "en": {
          "tokenizer": "standard"
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "id": {
        "type": "keyword",
        "index": true,
        "store": true
      },
      "id_tx": {
        "type": "text",
        "analyzer": "text_id_tx_class_id",
        "store": true,
        "index": true
      },
      "description": {
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index": true
      },
      "TM5": {
        "type": "keyword",
        "store": true,
        "index": true
      },
      "status": {
        "type": "keyword",
        "store": true,
        "index": true
      },
      "version": {
        "type": "keyword",
        "store": true,
        "index": true
      },
      "class": {
        "type": "text",
        "analyzer": "text_general_index_analyzer",
        "search_analyzer": "text_general_query_analyzer",
        "store": true
      },
      "long class": {
        "type": "text",
        "analyzer": "text_general_index_analyzer",
        "search_analyzer": "text_general_query_analyzer",
        "store": true
      },
      "class_id": {
        "type": "text",
        "analyzer": "text_id_tx_class_id",
        "index": true,
        "store": true
      },
      "notes": {
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index": true
      },
      "employee_notes": {
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index": true
      },
      "editor_notes": {
        "type": "text",
        "analyzer": "text_en_index",
        "search_analyzer": "text_en_query",
        "store": true,
        "index": true
      }
    }
  }
}'

Output

{"acknowledged":true,"shards_acknowledged":true,"index":"my-index-15"}

Query

curl -X POST "localhost:9200/my-index-15/_analyze" -H "Content-Type: application/json" -d '{
  "analyzer": "text_en_index",
  "text": "note-book"
}'  | jq '.'

Output

{
  "tokens": [
    {
      "token": "notebook",
      "start_offset": 0,
      "end_offset": 9,
      "type": "word",
      "position": 0,
      "positionLength": 2
    },
    {
      "token": "note",
      "start_offset": 0,
      "end_offset": 4,
      "type": "word",
      "position": 0
    },
    {
      "token": "book",
      "start_offset": 5,
      "end_offset": 9,
      "type": "word",
      "position": 1
    }
  ]
}

@nupurjaiswal @aswad1 can you please confirm if the query and the output is as expected ? If you have any additional queries to test this behaviour please share.

Thank you
@getsaurabh02

[prudhvigodithi]

Some more tests which works as expected.

curl -X PUT "localhost:9200/no-synonym-analyzer-test" -H "Content-Type: application/json" -d '{
  "settings": {
    "analysis": {
      "filter": {
        "custom_synonym_graph_filter": {
          "type": "synonym_graph",
          "synonyms": [
            "note book, note-book => notebook",
            "mobile phone, mobile-phone => smartphone"
          ]
        }
      },
      "analyzer": {
        "test_analyzer": {
          "type": "custom",
          "tokenizer": "whitespace",
          "filter": [
            "lowercase",
            "custom_synonym_graph_filter"
          ]
        }
      }
    }
  }
}'

 curl -X POST "localhost:9200/no-synonym-analyzer-test/_analyze" -H "Content-Type: application/json" -d '{
  "analyzer": "test_analyzer",
  "text": "note-book"
}' | jq '.'

When used custom synonym_analyzer.

curl -X PUT "localhost:9200/synonym-test-19" -H "Content-Type: application/json" -d '{
  "settings": {
    "analysis": {
      "analyzer": {
        "keyword_synonym_analyzer": {
          "type": "custom",
          "tokenizer": "keyword",
          "filter": ["lowercase"]
        },
        "standard_synonym_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": ["lowercase"]
        },
        "whitespace_synonym_analyzer": {
          "type": "custom",
          "tokenizer": "whitespace",
          "filter": ["lowercase"]
        },
        "keyword_test": {
          "tokenizer": "keyword",
          "filter": ["lowercase", "keyword_syn_filter"]
        },
        "standard_test": {
          "tokenizer": "standard",
          "filter": ["lowercase", "standard_syn_filter"]
        },
        "whitespace_test": {
          "tokenizer": "whitespace",
          "filter": ["lowercase", "whitespace_syn_filter"]
        }
      },
      "filter": {
        "keyword_syn_filter": {
          "type": "synonym_graph",
          "synonyms": ["note book, notebook => notebook"],
          "synonym_analyzer": "keyword_synonym_analyzer"
        },
        "standard_syn_filter": {
          "type": "synonym_graph",
          "synonyms": ["note book, notebook => notebook"],
          "synonym_analyzer": "standard_synonym_analyzer"
        },
        "whitespace_syn_filter": {
          "type": "synonym_graph",
          "synonyms": ["note book, notebook => notebook"],
          "synonym_analyzer": "whitespace_synonym_analyzer"
        }
      }
    }
  }
}'

curl -X POST "localhost:9200/synonym-test-19/_analyze" -H "Content-Type: application/json" -d '{
  "analyzer": "whitespace_test",
  "text": "note book"
}' | jq '.'

Scenario when used custom synonym_analyzer with custom which has word_delimiter_graph and synonym_graph

When used "synonym_analyzer": "custom_synonym_analyzer" meaning to pass a custom analyzer to synonym_analyzer, this again goes back to user configured analyzer chain and expectation is to ensure the synonym_graph is passed before word_delimiter_graph (or handled as required), else it will throw the same error. Example notice the custom_synonym_analyzer.

Adding @msfroh for inputs here, should this be handled by the user when configuring the analyzer chain for custom analyzer which will be used in synonym_analyzer?

curl -X PUT "localhost:9200/custom-test-11" -H "Content-Type: application/json" -d '{
  "settings": {
    "analysis": {
      "analyzer": {
        "custom_synonym_analyzer": {       
          "type": "custom",
          "tokenizer": "whitespace",
          "filter": [
            "lowercase"
          ]
        },
        "test_analyzer": {
          "type": "custom",
          "tokenizer": "whitespace",
          "filter": [
            "lowercase",
            "custom_word_delimiter",
            "custom_advanced_synonym"
          ]
        }
      },
      "filter": {
        "custom_word_delimiter": {
          "type": "word_delimiter_graph",
          "generate_word_parts": true,
          "catenate_all": true,
          "preserve_original": false,
          "split_on_numerics": false,
          "split_on_case_change": false
        },
        "custom_advanced_synonym": {
          "type": "synonym_graph",
          "synonyms": [
            "note-book => notebook",
            "note book => notebook",
            "mobile-phone => smartphone",
            "mobile phone => smartphone"
          ],
          "synonym_analyzer": "custom_synonym_analyzer"  
        }
      }
    }
  }
}'

[dblock]

[Catch All Triage - 1, 2, 3]

[nupurjaiswal]

@prudhvigodithi The mapping you posted earlier with "synonym_analyzer": "standard" is not working. I am getting the error "Token filter [custom_word_delimiter] cannot be used to parse synonyms"

[prudhvigodithi]

Hey @nupurjaiswal, did you try to build OpenSearch from my fork and test it? I have created a draft PR #16488 with the proposal to use synonym_analyzer, this is not yet released.
Thanks

[nupurjaiswal]

@prudhvigodithi I was using Opensearch on AWS. I am not running opensearch locally. I cloned your project from here: https://github.com/prudhvigodithi/OpenSearch/tree/bug-fix-synonym and tried to use "gradlew assemble", but I am facing errors while building it. Can you please share the instruction on how to build and run Opensearch using your fix? That would be helpful

[prudhvigodithi]

Hey @nupurjaiswal sure,

• Clone my fork https://github.com/prudhvigodithi/OpenSearch/tree/bug-fix-synonym.
• Make sure you have java 21 on your local.
• Assuming your local is linux

./gradlew :distribution:archives:linux-tar:assemble -x :distribution:docker:buildArm64DockerImage -x :distribution:docker:buildDockerImage -x :distribution:docker:buildPpc64leDockerImage -x :distribution:docker:buildS390xDockerImage -x distribution:docker:docker-arm64-export:exportArm64DockerImage -x :distribution:docker:docker-export:exportDockerImage -x :distribution:docker:docker-ppc64le-export:exportDockerImage -x :distribution:docker:docker-s390x-export:exportDockerImage 

• ./gradlew run -PnumNodes=3 this should start OpenSearch on port 9200.

Once started you can run your PUT query. #16263 (comment)

Also FYI I got some feedback on the PR, will take a look address the comments.

[nupurjaiswal]

Hi @prudhvigodithi I followed for your instruction: Cloned the repo "https://github.com/prudhvigodithi/OpenSearch.git", Installed java 21, followed your command build and run opensearch. The build was successful. I used postman to do the PUT query, I am still getting the below issue:

{
    "error": {
        "root_cause": [
            {
                "type": "illegal_argument_exception",
                "reason": "Token filter [custom_word_delimiter] cannot be used to parse synonyms"
            }
        ],
        "type": "illegal_argument_exception",
        "reason": "Token filter [custom_word_delimiter] cannot be used to parse synonyms"
    },
    "status": 400
}

[prudhvigodithi]

Hey @nupurjaiswal can you please confirm you tried with bug-fix-synonym (git clone https://github.com/prudhvigodithi/OpenSearch.git -b bug-fix-synonym) branch from my fork ?

[nupurjaiswal]

Thanks a lot, @prudhvigodithi. I realized I was on the wrong branch, but I was able to create the index and test queries once I switched to your branch. Your fix solved the issue. I will continue with the remaining testing and keep you updated.