Merge pull request #2585 from lonvia/name-mutations

Introduce character mutations to token analysis

docs/customize/Tokenizers.md

@@ -99,6 +99,9 @@ token-analysis:
           - words:
               - road -> rd
               - bridge -> bdge,br,brdg,bri,brg
+      mutations:
+          - pattern: 'ä'
+            replacements: ['ä', 'ae']
 ```
 
 The configuration file contains four sections:
@@ -205,12 +208,11 @@ the `analyzer` parameter must be set. Currently there is only one implementation
 ##### Generic token analyzer
 
 The generic analyzer is able to create variants from a list of given
-abbreviation and decomposition replacements. It takes one optional parameter
+abbreviation and decomposition replacements and introduce spelling variations.
-`variants` which lists the replacements to apply. If the section is
-omitted, then the generic analyzer becomes a simple analyzer that only
-applies the transliteration.
 
-The variants section defines lists of replacements which create alternative
+###### Variants
+
+The optional 'variants' section defines lists of replacements which create alternative
 spellings of a name. To create the variants, a name is scanned from left to
 right and the longest matching replacement is applied until the end of the
 string is reached.
@@ -296,6 +298,32 @@ decomposition has an effect here on the source as well. So a rule
 means that for a word like `hauptstrasse` four variants are created:
 `hauptstrasse`, `haupt strasse`, `hauptstr` and `haupt str`.
 
+###### Mutations
+
+The 'mutation' section in the configuration describes an additional set of
+replacements to be applied after the variants have been computed.
+
+Each mutation is described by two parameters: `pattern` and `replacements`.
+The pattern must contain a single regular expression to search for in the
+variant name. The regular expressions need to follow the syntax for
+[Python regular expressions](file:///usr/share/doc/python3-doc/html/library/re.html#regular-expression-syntax).
+Capturing groups are not permitted.
+`replacements` must contain a list of strings that the pattern
+should be replaced with. Each occurrence of the pattern is replaced with
+all given replacements. Be mindful of combinatorial explosion of variants.
+
+###### Modes
+
+The generic analyser supports a special mode `variant-only`. When configured
+then it consumes the input token and emits only variants (if any exist). Enable
+the mode by adding:
+
+```
+  mode: variant-only
+```
+
+to the analyser configuration.
+
 ### Reconfiguration
 
 Changing the configuration after the import is currently not possible, although

nominatim/tokenizer/token_analysis/config_variants.py

@@ -0,0 +1,134 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Parser for configuration for variants.
+"""
+from collections import defaultdict, namedtuple
+import itertools
+import re
+
+from icu import Transliterator
+
+from nominatim.config import flatten_config_list
+from nominatim.errors import UsageError
+
+ICUVariant = namedtuple('ICUVariant', ['source', 'replacement'])
+
+def get_variant_config(rules, normalization_rules):
+    """ Convert the variant definition from the configuration into
+        replacement sets.
+
+        Returns a tuple containing the replacement set and the list of characters
+        used in the replacements.
+    """
+    immediate = defaultdict(list)
+    chars = set()
+
+    if rules:
+        vset = set()
+        rules = flatten_config_list(rules, 'variants')
+
+        vmaker = _VariantMaker(normalization_rules)
+
+        for section in rules:
+            for rule in (section.get('words') or []):
+                vset.update(vmaker.compute(rule))
+
+        # Intermediate reorder by source. Also compute required character set.
+        for variant in vset:
+            if variant.source[-1] == ' ' and variant.replacement[-1] == ' ':
+                replstr = variant.replacement[:-1]
+            else:
+                replstr = variant.replacement
+            immediate[variant.source].append(replstr)
+            chars.update(variant.source)
+
+    return list(immediate.items()), ''.join(chars)
+
+
+class _VariantMaker:
+    """ Generater for all necessary ICUVariants from a single variant rule.
+
+        All text in rules is normalized to make sure the variants match later.
+    """
+
+    def __init__(self, norm_rules):
+        self.norm = Transliterator.createFromRules("rule_loader_normalization",
+                                                   norm_rules)
+
+
+    def compute(self, rule):
+        """ Generator for all ICUVariant tuples from a single variant rule.
+        """
+        parts = re.split(r'(\|)?([=-])>', rule)
+        if len(parts) != 4:
+            raise UsageError("Syntax error in variant rule: " + rule)
+
+        decompose = parts[1] is None
+        src_terms = [self._parse_variant_word(t) for t in parts[0].split(',')]
+        repl_terms = (self.norm.transliterate(t).strip() for t in parts[3].split(','))
+
+        # If the source should be kept, add a 1:1 replacement
+        if parts[2] == '-':
+            for src in src_terms:
+                if src:
+                    for froms, tos in _create_variants(*src, src[0], decompose):
+                        yield ICUVariant(froms, tos)
+
+        for src, repl in itertools.product(src_terms, repl_terms):
+            if src and repl:
+                for froms, tos in _create_variants(*src, repl, decompose):
+                    yield ICUVariant(froms, tos)
+
+
+    def _parse_variant_word(self, name):
+        name = name.strip()
+        match = re.fullmatch(r'([~^]?)([^~$^]*)([~$]?)', name)
+        if match is None or (match.group(1) == '~' and match.group(3) == '~'):
+            raise UsageError("Invalid variant word descriptor '{}'".format(name))
+        norm_name = self.norm.transliterate(match.group(2)).strip()
+        if not norm_name:
+            return None
+
+        return norm_name, match.group(1), match.group(3)
+
+
+_FLAG_MATCH = {'^': '^ ',
+               '$': ' ^',
+               '': ' '}
+
+
+def _create_variants(src, preflag, postflag, repl, decompose):
+    if preflag == '~':
+        postfix = _FLAG_MATCH[postflag]
+        # suffix decomposition
+        src = src + postfix
+        repl = repl + postfix
+
+        yield src, repl
+        yield ' ' + src, ' ' + repl
+
+        if decompose:
+            yield src, ' ' + repl
+            yield ' ' + src, repl
+    elif postflag == '~':
+        # prefix decomposition
+        prefix = _FLAG_MATCH[preflag]
+        src = prefix + src
+        repl = prefix + repl
+
+        yield src, repl
+        yield src + ' ', repl + ' '
+
+        if decompose:
+            yield src, repl + ' '
+            yield src + ' ', repl
+    else:
+        prefix = _FLAG_MATCH[preflag]
+        postfix = _FLAG_MATCH[postflag]
+
+        yield prefix + src + postfix, prefix + repl + postfix

nominatim/tokenizer/token_analysis/generic.py

@@ -7,145 +7,44 @@
 """
 Generic processor for names that creates abbreviation variants.
 """
-from collections import defaultdict, namedtuple
 import itertools
-import re
 
-from icu import Transliterator
 import datrie
 
-from nominatim.config import flatten_config_list
 from nominatim.errors import UsageError
+from nominatim.tokenizer.token_analysis.config_variants import get_variant_config
+from nominatim.tokenizer.token_analysis.generic_mutation import MutationVariantGenerator
 
 ### Configuration section
 
-ICUVariant = namedtuple('ICUVariant', ['source', 'replacement'])
-
 def configure(rules, normalization_rules):
     """ Extract and preprocess the configuration for this module.
     """
     config = {}
 
-    config['replacements'], config['chars'] = _get_variant_config(rules.get('variants'),
+    config['replacements'], config['chars'] = get_variant_config(rules.get('variants'),
-                                                                  normalization_rules)
+                                                                 normalization_rules)
     config['variant_only'] = rules.get('mode', '') == 'variant-only'
 
+    # parse mutation rules
+    config['mutations'] = []
+    for rule in rules.get('mutations', []):
+        if 'pattern' not in rule:
+            raise UsageError("Missing field 'pattern' in mutation configuration.")
+        if not isinstance(rule['pattern'], str):
+            raise UsageError("Field 'pattern' in mutation configuration "
+                             "must be a simple text field.")
+        if 'replacements' not in rule:
+            raise UsageError("Missing field 'replacements' in mutation configuration.")
+        if not isinstance(rule['replacements'], list):
+            raise UsageError("Field 'replacements' in mutation configuration "
+                             "must be a list of texts.")
+
+        config['mutations'].append((rule['pattern'], rule['replacements']))
+
     return config
 
 
-def _get_variant_config(rules, normalization_rules):
-    """ Convert the variant definition from the configuration into
-        replacement sets.
-    """
-    immediate = defaultdict(list)
-    chars = set()
-
-    if rules:
-        vset = set()
-        rules = flatten_config_list(rules, 'variants')
-
-        vmaker = _VariantMaker(normalization_rules)
-
-        for section in rules:
-            for rule in (section.get('words') or []):
-                vset.update(vmaker.compute(rule))
-
-        # Intermediate reorder by source. Also compute required character set.
-        for variant in vset:
-            if variant.source[-1] == ' ' and variant.replacement[-1] == ' ':
-                replstr = variant.replacement[:-1]
-            else:
-                replstr = variant.replacement
-            immediate[variant.source].append(replstr)
-            chars.update(variant.source)
-
-    return list(immediate.items()), ''.join(chars)
-
-
-class _VariantMaker:
-    """ Generater for all necessary ICUVariants from a single variant rule.
-
-        All text in rules is normalized to make sure the variants match later.
-    """
-
-    def __init__(self, norm_rules):
-        self.norm = Transliterator.createFromRules("rule_loader_normalization",
-                                                   norm_rules)
-
-
-    def compute(self, rule):
-        """ Generator for all ICUVariant tuples from a single variant rule.
-        """
-        parts = re.split(r'(\|)?([=-])>', rule)
-        if len(parts) != 4:
-            raise UsageError("Syntax error in variant rule: " + rule)
-
-        decompose = parts[1] is None
-        src_terms = [self._parse_variant_word(t) for t in parts[0].split(',')]
-        repl_terms = (self.norm.transliterate(t).strip() for t in parts[3].split(','))
-
-        # If the source should be kept, add a 1:1 replacement
-        if parts[2] == '-':
-            for src in src_terms:
-                if src:
-                    for froms, tos in _create_variants(*src, src[0], decompose):
-                        yield ICUVariant(froms, tos)
-
-        for src, repl in itertools.product(src_terms, repl_terms):
-            if src and repl:
-                for froms, tos in _create_variants(*src, repl, decompose):
-                    yield ICUVariant(froms, tos)
-
-
-    def _parse_variant_word(self, name):
-        name = name.strip()
-        match = re.fullmatch(r'([~^]?)([^~$^]*)([~$]?)', name)
-        if match is None or (match.group(1) == '~' and match.group(3) == '~'):
-            raise UsageError("Invalid variant word descriptor '{}'".format(name))
-        norm_name = self.norm.transliterate(match.group(2)).strip()
-        if not norm_name:
-            return None
-
-        return norm_name, match.group(1), match.group(3)
-
-
-_FLAG_MATCH = {'^': '^ ',
-               '$': ' ^',
-               '': ' '}
-
-
-def _create_variants(src, preflag, postflag, repl, decompose):
-    if preflag == '~':
-        postfix = _FLAG_MATCH[postflag]
-        # suffix decomposition
-        src = src + postfix
-        repl = repl + postfix
-
-        yield src, repl
-        yield ' ' + src, ' ' + repl
-
-        if decompose:
-            yield src, ' ' + repl
-            yield ' ' + src, repl
-    elif postflag == '~':
-        # prefix decomposition
-        prefix = _FLAG_MATCH[preflag]
-        src = prefix + src
-        repl = prefix + repl
-
-        yield src, repl
-        yield src + ' ', repl + ' '
-
-        if decompose:
-            yield src, repl + ' '
-            yield src + ' ', repl
-    else:
-        prefix = _FLAG_MATCH[preflag]
-        postfix = _FLAG_MATCH[postflag]
-
-        yield prefix + src + postfix, prefix + repl + postfix
-
-
 ### Analysis section
 
 def create(transliterator, config):
@@ -171,19 +70,43 @@ class GenericTokenAnalysis:
         else:
             self.replacements = None
 
+        # set up mutation rules
+        self.mutations = [MutationVariantGenerator(*cfg) for cfg in config['mutations']]
+
 
     def get_variants_ascii(self, norm_name):
         """ Compute the spelling variants for the given normalized name
             and transliterate the result.
         """
+        variants = self._generate_word_variants(norm_name)
+
+        for mutation in self.mutations:
+            variants = mutation.generate(variants)
+
+        return [name for name in self._transliterate_unique_list(norm_name, variants) if name]
+
+
+    def _transliterate_unique_list(self, norm_name, iterable):
+        seen = set()
+        if self.variant_only:
+            seen.add(norm_name)
+
+        for variant in map(str.strip, iterable):
+            if variant not in seen:
+                seen.add(variant)
+                yield self.to_ascii.transliterate(variant).strip()
+
+
+    def _generate_word_variants(self, norm_name):
         baseform = '^ ' + norm_name + ' ^'
+        baselen = len(baseform)
         partials = ['']
 
         startpos = 0
         if self.replacements is not None:
             pos = 0
             force_space = False
-            while pos < len(baseform):
+            while pos < baselen:
                 full, repl = self.replacements.longest_prefix_item(baseform[pos:],
                                                                    (None, None))
                 if full is not None:
@@ -207,24 +130,9 @@ class GenericTokenAnalysis:
 
         # No variants detected? Fast return.
         if startpos == 0:
-            if self.variant_only:
+            return (norm_name, )
-                return []
 
-            trans_name = self.to_ascii.transliterate(norm_name).strip()
+        if startpos < baselen:
-            return [trans_name] if trans_name else []
+            return (part[1:] + baseform[startpos:-1] for part in partials)
 
-        return self._compute_result_set(partials, baseform[startpos:],
+        return (part[1:-1] for part in partials)
-                                        norm_name if self.variant_only else '')
-
-
-    def _compute_result_set(self, partials, prefix, exclude):
-        results = set()
-
-        for variant in partials:
-            vname = (variant + prefix)[1:-1].strip()
-            if vname != exclude:
-                trans_name = self.to_ascii.transliterate(vname).strip()
-                if trans_name:
-                    results.add(trans_name)
-
-        return list(results)

nominatim/tokenizer/token_analysis/generic_mutation.py

@@ -0,0 +1,56 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Creator for mutation variants for the generic token analysis.
+"""
+import itertools
+import logging
+import re
+
+from nominatim.errors import UsageError
+
+LOG = logging.getLogger()
+
+def _zigzag(outer, inner):
+    return itertools.chain.from_iterable(itertools.zip_longest(outer, inner, fillvalue=''))
+
+
+class MutationVariantGenerator:
+    """ Generates name variants by applying a regular expression to the name
+        and replacing it with one or more variants. When the regular expression
+        matches more than once, each occurence is replaced with all replacement
+        patterns.
+    """
+
+    def __init__(self, pattern, replacements):
+        self.pattern = re.compile(pattern)
+        self.replacements = replacements
+
+        if self.pattern.groups > 0:
+            LOG.fatal("The mutation pattern %s contains a capturing group. "
+                      "This is not allowed.", pattern)
+            raise UsageError("Bad mutation pattern in configuration.")
+
+
+    def generate(self, names):
+        """ Generator function for the name variants. 'names' is an iterable
+            over a set of names for which the variants are to be generated.
+        """
+        for name in names:
+            parts = self.pattern.split(name)
+            if len(parts) == 1:
+                yield name
+            else:
+                for seps in self._fillers(len(parts)):
+                    yield ''.join(_zigzag(parts, seps))
+
+
+    def _fillers(self, num_parts):
+        """ Returns a generator for strings to join the given number of string
+            parts in all possible combinations.
+        """
+        return itertools.product(self.replacements, repeat=num_parts - 1)

settings/icu_tokenizer.yaml

@@ -59,6 +59,13 @@ token-analysis:
       mode: variant-only
       variants:
           - !include icu-rules/variants-de.yaml
+      mutations:
+          - pattern: ä
+            replacements: ["ä", "ae"]
+          - pattern: ö
+            replacements: ["ö", "oe"]
+          - pattern: ü
+            replacements: ["ü", "ue"]
     - id: el
       analyzer: generic
       mode: variant-only

test/bdd/db/import/naming.feature

@@ -58,3 +58,48 @@ Feature: Import and search of names
         | រាជធានីភ្នំពេញ |
         | 東京都 |
         | ပုဗ္ဗသီရိ |
+
+
+    Scenario: German umlauts can be found when expanded
+        Given the places
+            | osm | class | type | name+name:de |
+            | N1  | place | city | Münster      |
+            | N2  | place | city | Köln         |
+            | N3  | place | city | Gräfenroda   |
+        When importing
+        When sending search query "münster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "muenster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "munster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "Köln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "Koeln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "Koln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "gräfenroda"
+        Then results contain
+            | osm |
+            | N3  |
+        When sending search query "graefenroda"
+        Then results contain
+            | osm |
+            | N3  |
+        When sending search query "grafenroda"
+        Then results contain
+            | osm |
+            | N3  |

test/python/tokenizer/token_analysis/test_generic_mutation.py

@@ -0,0 +1,89 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Tests for generic token analysis, mutation part.
+"""
+import pytest
+
+from icu import Transliterator
+
+import nominatim.tokenizer.token_analysis.generic as module
+from nominatim.errors import UsageError
+
+DEFAULT_NORMALIZATION = """ '🜳' > ' ';
+                            [[:Nonspacing Mark:] [:Cf:]] >;
+                            :: lower ();
+                            [[:Punctuation:][:Space:]]+ > ' '
+                        """
+
+DEFAULT_TRANSLITERATION = """ ::  Latin ();
+                              '🜵' > ' ';
+                          """
+
+class TestMutationNoVariants:
+
+    def make_analyser(self, *mutations):
+        rules = { 'analyzer': 'generic',
+                  'mutations': [ {'pattern': m[0], 'replacements': m[1]}
+                                 for m in mutations]
+                }
+        config = module.configure(rules, DEFAULT_NORMALIZATION)
+        trans = Transliterator.createFromRules("test_trans", DEFAULT_TRANSLITERATION)
+
+        self.analysis = module.create(trans, config)
+
+
+    def variants(self, name):
+        norm = Transliterator.createFromRules("test_norm", DEFAULT_NORMALIZATION)
+        return set(self.analysis.get_variants_ascii(norm.transliterate(name).strip()))
+
+
+    @pytest.mark.parametrize('pattern', ('(capture)', ['a list']))
+    def test_bad_pattern(self, pattern):
+        with pytest.raises(UsageError):
+            self.make_analyser((pattern, ['b']))
+
+
+    @pytest.mark.parametrize('replacements', (None, 'a string'))
+    def test_bad_replacement(self, replacements):
+        with pytest.raises(UsageError):
+            self.make_analyser(('a', replacements))
+
+
+    def test_simple_replacement(self):
+        self.make_analyser(('a', ['b']))
+
+        assert self.variants('none') == {'none'}
+        assert self.variants('abba') == {'bbbb'}
+        assert self.variants('2 aar') == {'2 bbr'}
+
+
+    def test_multichar_replacement(self):
+        self.make_analyser(('1 1', ['1 1 1']))
+
+        assert self.variants('1 1456') == {'1 1 1456'}
+        assert self.variants('1 1 1') == {'1 1 1 1'}
+
+
+    def test_removement_replacement(self):
+        self.make_analyser((' ', [' ', '']))
+
+        assert self.variants('A 345') == {'a 345', 'a345'}
+        assert self.variants('a g b') == {'a g b', 'ag b', 'a gb', 'agb'}
+
+
+    def test_regex_pattern(self):
+        self.make_analyser(('[^a-z]+', ['XXX', ' ']))
+
+        assert self.variants('a-34n12') == {'aXXXnXXX', 'aXXXn', 'a nXXX', 'a n'}
+
+
+    def test_multiple_mutations(self):
+        self.make_analyser(('ä', ['ä', 'ae']), ('ö', ['ö', 'oe']))
+
+        assert self.variants('Längenöhr') == {'längenöhr', 'laengenöhr',
+                                              'längenoehr', 'laengenoehr'}