gngram-lookup 0.2.1__py3-none-any.whl → 0.2.2__py3-none-any.whl

gngram_counter/lookup.py

@@ -2,6 +2,11 @@
 High-level lookup API for gngram-counter.
 
 Provides simple functions for word frequency lookups similar to bnc-lookup.
+
+Includes contraction fallback: if a contraction like "don't" is not found
+directly, the stem ("do") is looked up instead. The ngram corpus only
+contains pure alphabetic words, so contractions and their suffix parts
+(n't, 'll, etc.) are absent — but the stems are present.
 """
 
 import hashlib
@@ -11,6 +16,7 @@ from typing import TypedDict
 import polars as pl
 
 from gngram_counter.data import get_hash_file, is_data_installed
+from gngram_counter.normalize import normalize
 
 
 class FrequencyData(TypedDict):
@@ -22,6 +28,26 @@ class FrequencyData(TypedDict):
     sum_df: int  # Total document frequency across all decades
 
 
+# Contraction suffixes stored as separate tokens in the ngram corpus
+# Order matters: longer suffixes must be checked before shorter ones
+CONTRACTION_SUFFIXES = ("n't", "'ll", "'re", "'ve", "'m", "'d")
+
+# Specific stems that form 's contractions (where 's = "is" or "has").
+# NOT generalized — 's is ambiguous with possessive, so only known
+# contraction stems are listed here. Ported from bnc-lookup.
+S_CONTRACTION_STEMS = frozenset({
+    # Pronouns (unambiguously 's = "is" or "has", never possessive)
+    'it', 'he', 'she', 'that', 'what', 'who',
+    # Adverbs / demonstratives
+    'where', 'how', 'here', 'there',
+    # "let's" = "let us"
+    'let',
+    # Indefinite pronouns
+    'somebody', 'everybody', 'everyone', 'nobody',
+    'anywhere', 'nowhere',
+})
+
+
 @lru_cache(maxsize=256)
 def _load_bucket(prefix: str) -> pl.DataFrame:
     """Load and cache a parquet bucket file."""
@@ -30,13 +56,63 @@ def _load_bucket(prefix: str) -> pl.DataFrame:
 
 def _hash_word(word: str) -> tuple[str, str]:
     """Hash a word and return (prefix, suffix)."""
-    h = hashlib.md5(word.lower().encode("utf-8")).hexdigest()
+    h = hashlib.md5(normalize(word).encode("utf-8")).hexdigest()
     return h[:2], h[2:]
 
 
+def _lookup_frequency(word: str) -> FrequencyData | None:
+    """Look up frequency data for a single word form (no fallbacks)."""
+    if not word:
+        return None
+    prefix, suffix = _hash_word(word)
+    try:
+        df = _load_bucket(prefix)
+    except FileNotFoundError:
+        return None
+    row = df.filter(pl.col("hash") == suffix)
+    if len(row) == 0:
+        return None
+    return FrequencyData(
+        peak_tf=row["peak_tf"][0],
+        peak_df=row["peak_df"][0],
+        sum_tf=row["sum_tf"][0],
+        sum_df=row["sum_df"][0],
+    )
+
+
+def _split_contraction(word: str) -> tuple[str, str] | None:
+    """Split a contraction into its component parts if possible.
+
+    The ngram corpus tokenizes contractions separately (e.g., "we'll" -> "we" + "'ll").
+    This function reverses that split for fallback lookup.
+
+    Returns:
+        Tuple of (stem, suffix) if the word matches a contraction pattern,
+        or None if no contraction pattern matches.
+    """
+    for suffix in CONTRACTION_SUFFIXES:
+        if word.endswith(suffix):
+            stem = word[:-len(suffix)]
+            if stem:
+                return (stem, suffix)
+
+    # Specific 's contractions from curated allowlist (not possessives)
+    if word.endswith("'s"):
+        stem = word[:-2]
+        if stem in S_CONTRACTION_STEMS:
+            return (stem, "'s")
+
+    return None
+
+
 def exists(word: str) -> bool:
     """Check if a word exists in the ngram data.
 
+    Performs case-insensitive lookup with automatic fallbacks:
+    1. Direct lookup of the normalized word
+    2. Contraction fallback: if word is a contraction, check if both
+       components exist (e.g., "don't" -> "do" + "n't")
+
     Args:
         word: The word to check (case-insensitive)
 
@@ -51,14 +127,27 @@ def exists(word: str) -> bool:
             "Data files not installed. Run: python -m gngram_counter.download_data"
         )
 
-    prefix, suffix = _hash_word(word)
-    df = _load_bucket(prefix)
-    return len(df.filter(pl.col("hash") == suffix)) > 0
+    word = normalize(word)
+
+    if _lookup_frequency(word) is not None:
+        return True
+
+    # Contraction fallback: check if the stem exists
+    parts = _split_contraction(word)
+    if parts:
+        stem, _ = parts
+        if _lookup_frequency(stem) is not None:
+            return True
+
+    return False
 
 
 def frequency(word: str) -> FrequencyData | None:
     """Get frequency data for a word.
 
+    Performs case-insensitive lookup with contraction fallback.
+    For contractions, returns the stem's frequency data.
+
     Args:
         word: The word to look up (case-insensitive)
 
@@ -73,19 +162,21 @@ def frequency(word: str) -> FrequencyData | None:
             "Data files not installed. Run: python -m gngram_counter.download_data"
         )
 
-    prefix, suffix = _hash_word(word)
-    df = _load_bucket(prefix)
-    row = df.filter(pl.col("hash") == suffix)
+    word = normalize(word)
 
-    if len(row) == 0:
-        return None
+    result = _lookup_frequency(word)
+    if result is not None:
+        return result
 
-    return FrequencyData(
-        peak_tf=row["peak_tf"][0],
-        peak_df=row["peak_df"][0],
-        sum_tf=row["sum_tf"][0],
-        sum_df=row["sum_df"][0],
-    )
+    # Contraction fallback: return the stem's frequency
+    parts = _split_contraction(word)
+    if parts:
+        stem, _ = parts
+        stem_freq = _lookup_frequency(stem)
+        if stem_freq is not None:
+            return stem_freq
+
+    return None
 
 
 def batch_frequency(words: list[str]) -> dict[str, FrequencyData | None]:
@@ -106,24 +197,27 @@ def batch_frequency(words: list[str]) -> dict[str, FrequencyData | None]:
         )
 
     # Group words by bucket prefix for efficient batch lookups
-    by_prefix: dict[str, list[tuple[str, str]]] = {}
+    by_prefix: dict[str, list[tuple[str, str, str]]] = {}
+    contraction_words: list[str] = []
+
     for word in words:
-        prefix, suffix = _hash_word(word)
+        normalized = normalize(word)
+        prefix, suffix = _hash_word(normalized)
         if prefix not in by_prefix:
             by_prefix[prefix] = []
-        by_prefix[prefix].append((word, suffix))
+        by_prefix[prefix].append((word, normalized, suffix))
 
     results: dict[str, FrequencyData | None] = {}
 
-    for prefix, word_suffix_pairs in by_prefix.items():
+    for prefix, entries in by_prefix.items():
         df = _load_bucket(prefix)
-        suffixes = [s for _, s in word_suffix_pairs]
+        suffixes = [s for _, _, s in entries]
 
         # Filter to all matching suffixes at once
         matches = df.filter(pl.col("hash").is_in(suffixes))
         match_dict = {row["hash"]: row for row in matches.iter_rows(named=True)}
 
-        for word, suffix in word_suffix_pairs:
+        for word, normalized, suffix in entries:
             if suffix in match_dict:
                 row = match_dict[suffix]
                 results[word] = FrequencyData(
@@ -133,6 +227,18 @@ def batch_frequency(words: list[str]) -> dict[str, FrequencyData | None]:
                     sum_df=row["sum_df"],
                 )
             else:
+                # Mark for contraction fallback
                 results[word] = None
+                contraction_words.append(word)
+
+    # Contraction fallback for words not found directly
+    for word in contraction_words:
+        normalized = normalize(word)
+        parts = _split_contraction(normalized)
+        if parts:
+            stem, _ = parts
+            stem_freq = _lookup_frequency(stem)
+            if stem_freq is not None:
+                results[word] = stem_freq
 
     return results

gngram_counter/normalize.py

@@ -0,0 +1,48 @@
+"""Text normalization utilities for gngram-counter.
+
+Handles normalization of Unicode apostrophe variants and other text
+transformations to ensure consistent matching against the ngram corpus.
+
+Ported from bnc-lookup normalize.py.
+"""
+
+# Unicode characters that should normalize to ASCII apostrophe (U+0027)
+# Ordered by likelihood of occurrence in English text
+APOSTROPHE_VARIANTS = (
+    '\u2019'  # RIGHT SINGLE QUOTATION MARK (most common smart quote)
+    '\u2018'  # LEFT SINGLE QUOTATION MARK
+    '\u0060'  # GRAVE ACCENT
+    '\u00B4'  # ACUTE ACCENT
+    '\u201B'  # SINGLE HIGH-REVERSED-9 QUOTATION MARK
+    '\u2032'  # PRIME
+    '\u2035'  # REVERSED PRIME
+    '\u02B9'  # MODIFIER LETTER PRIME
+    '\u02BC'  # MODIFIER LETTER APOSTROPHE
+    '\u02C8'  # MODIFIER LETTER VERTICAL LINE
+    '\u0313'  # COMBINING COMMA ABOVE
+    '\u0315'  # COMBINING COMMA ABOVE RIGHT
+    '\u055A'  # ARMENIAN APOSTROPHE
+    '\u05F3'  # HEBREW PUNCTUATION GERESH
+    '\u07F4'  # NKO HIGH TONE APOSTROPHE
+    '\u07F5'  # NKO LOW TONE APOSTROPHE
+    '\uFF07'  # FULLWIDTH APOSTROPHE
+    '\u1FBF'  # GREEK PSILI
+    '\u1FBD'  # GREEK KORONIS
+    '\uA78C'  # LATIN SMALL LETTER SALTILLO
+)
+
+# Pre-compiled translation table for fast apostrophe normalization
+_APOSTROPHE_TABLE = str.maketrans({char: "'" for char in APOSTROPHE_VARIANTS})
+
+
+def normalize_apostrophes(text: str) -> str:
+    """Normalize Unicode apostrophe variants to ASCII apostrophe."""
+    return text.translate(_APOSTROPHE_TABLE)
+
+
+def normalize(text: str) -> str:
+    """Normalize text for ngram lookup.
+
+    Applies: apostrophe variant conversion, lowercase, strip whitespace.
+    """
+    return normalize_apostrophes(text).lower().strip()

{gngram_lookup-0.2.1.dist-info → gngram_lookup-0.2.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: gngram-lookup
-Version: 0.2.1
+Version: 0.2.2
 Summary: Static Hash-Based Lookup for Google Ngram Frequencies
 Home-page: https://github.com/craigtrim/gngram-lookup
 License: Proprietary
@@ -32,7 +32,7 @@ Description-Content-Type: text/markdown
 [![PyPI version](https://badge.fury.io/py/gngram-lookup.svg)](https://badge.fury.io/py/gngram-lookup)
 [![Downloads](https://pepy.tech/badge/gngram-lookup)](https://pepy.tech/project/gngram-lookup)
 [![Downloads/Month](https://pepy.tech/badge/gngram-lookup/month)](https://pepy.tech/project/gngram-lookup)
-[![Tests](https://img.shields.io/badge/tests-58-brightgreen)](tests/)
+[![Tests](https://img.shields.io/badge/tests-131-brightgreen)](https://github.com/craigtrim/gngram-lookup/tree/main/tests)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
 
 Word frequency from 500 years of books. O(1) lookup. 5 million words.
@@ -74,11 +74,11 @@ gngram-freq computer
 
 ## Docs
 
-- [API Reference](docs/api.md)
-- [CLI Reference](docs/cli.md)
-- [Data Format](docs/data-format.md)
-- [Use Cases](docs/use-cases.md)
-- [Development](docs/development.md)
+- [API Reference](https://github.com/craigtrim/gngram-lookup/blob/main/docs/api.md)
+- [CLI Reference](https://github.com/craigtrim/gngram-lookup/blob/main/docs/cli.md)
+- [Data Format](https://github.com/craigtrim/gngram-lookup/blob/main/docs/data-format.md)
+- [Use Cases](https://github.com/craigtrim/gngram-lookup/blob/main/docs/use-cases.md)
+- [Development](https://github.com/craigtrim/gngram-lookup/blob/main/docs/development.md)
 
 ## See Also
 
@@ -91,5 +91,5 @@ Data derived from the [Google Books Ngram](https://books.google.com/ngrams) data
 
 ## License
 
-Proprietary. See [LICENSE](LICENSE).
+Proprietary. See [LICENSE](https://github.com/craigtrim/gngram-lookup/blob/main/LICENSE).
 

gngram_lookup-0.2.2.dist-info/RECORD

@@ -0,0 +1,11 @@
+gngram_counter/__init__.py,sha256=JsgQYIF5LcYqdhWuDuVhrlt5eVebk36CsXQK9Q3H5ZA,374
+gngram_counter/cli.py,sha256=7PScHhnjNoi0so0IGGZ7ipi0bgILtfQmZ0PPCxJCO_k,861
+gngram_counter/data.py,sha256=HvESF1lc9v7HPbemksnvzvV460ts9gBjvACMZZao9qs,1089
+gngram_counter/download_data.py,sha256=vlggDDszmI29UJA9H17AK-_BTNOcjq9oWoKju4DDCTU,2663
+gngram_counter/lookup.py,sha256=r67ulgLPM0zkIWyulQsmsRVbIZt9J1APQ1974DWgX1k,7564
+gngram_counter/normalize.py,sha256=UDUPk4Mb-fcdIy-4WAivFnk33H6gwxxD3oKFHq2tNNg,1693
+gngram_lookup-0.2.2.dist-info/LICENSE,sha256=9r2EF9XQjpHEtltPlomXEmegbVVhZsVHzygSPfiid_E,1497
+gngram_lookup-0.2.2.dist-info/METADATA,sha256=DSF-z85Un8wSJQs55r0Q6_O5K9c9fyDuUi9rW1NMFIU,3323
+gngram_lookup-0.2.2.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+gngram_lookup-0.2.2.dist-info/entry_points.txt,sha256=bzFME4Um0_lWLTo2JcvFseBUSD7Gk7r-156Cr_wssnM,109
+gngram_lookup-0.2.2.dist-info/RECORD,,

gngram_lookup-0.2.1.dist-info/RECORD

@@ -1,10 +0,0 @@
-gngram_counter/__init__.py,sha256=JsgQYIF5LcYqdhWuDuVhrlt5eVebk36CsXQK9Q3H5ZA,374
-gngram_counter/cli.py,sha256=7PScHhnjNoi0so0IGGZ7ipi0bgILtfQmZ0PPCxJCO_k,861
-gngram_counter/data.py,sha256=HvESF1lc9v7HPbemksnvzvV460ts9gBjvACMZZao9qs,1089
-gngram_counter/download_data.py,sha256=vlggDDszmI29UJA9H17AK-_BTNOcjq9oWoKju4DDCTU,2663
-gngram_counter/lookup.py,sha256=8WThcRWmIYPBgHTwfOYNSN1wTgddnBXCx6moNwulKXU,3992
-gngram_lookup-0.2.1.dist-info/LICENSE,sha256=9r2EF9XQjpHEtltPlomXEmegbVVhZsVHzygSPfiid_E,1497
-gngram_lookup-0.2.1.dist-info/METADATA,sha256=okH1jbNz8k9EsgS8aFnOdeslJwY7wUhDSFwpWptFeD4,2952
-gngram_lookup-0.2.1.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-gngram_lookup-0.2.1.dist-info/entry_points.txt,sha256=bzFME4Um0_lWLTo2JcvFseBUSD7Gk7r-156Cr_wssnM,109
-gngram_lookup-0.2.1.dist-info/RECORD,,