calliope-ts 0.0.2 → 0.0.4

package/dist/clio/tagfix.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Correct a sentence's tags in place-safe fashion (returns a new array).
+ * `tokens` and `tags` are the en-pos outputs, index-aligned.
+ */
+export declare function correctTags(tokens: string[], tags: string[]): string[];
+//# sourceMappingURL=tagfix.d.ts.map

package/dist/clio/tagfix.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tagfix.d.ts","sourceRoot":"","sources":["../../src/clio/tagfix.ts"],"names":[],"mappings":"AAwCA;;;GAGG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,EAAE,CA2DtE"}

package/dist/clio/tagfix.js

@@ -0,0 +1,101 @@
+// tagfix.ts — Pre-parse POS-tag correction layer.
+//
+// FinNLP's en-pos tagger is structurally sound but carries a small tail of
+// SYSTEMATIC tag errors that matter enormously for verse analysis, because a
+// wrong tag flips a word's content/function status (→ its stress tier) and
+// derails the en-parse dependency tree built from the tags.  This pass runs
+// BETWEEN en-pos and en-parse (see parseDocument in parser.ts), so corrected
+// tags repair both the tagging AND the resulting dependency structure — a
+// post-hoc fix of the parse could never do that.
+//
+// Every rule below targets an error class actually observed in this repo's
+// trials; rules are deliberately narrow (anti-gaming: each must be justified
+// by the error it fixes, not by benchmark deltas).
+/** Zero-derived irregular past participles that en-pos tags NN/VBP after a
+ *  have-auxiliary ("had quit", "has put", "have read").  Only forms whose
+ *  participle is identical to the base/noun spelling — the -en/-ed forms tag
+ *  fine on their own. */
+const ZERO_PARTICIPLES = new Set([
+    'quit', 'put', 'set', 'cut', 'hit', 'let', 'shut', 'cast', 'cost', 'hurt',
+    'burst', 'split', 'spread', 'bet', 'wed', 'read', 'rid', 'shed', 'thrust',
+    'slit', 'bid', 'broadcast', 'upset', 'sunburst',
+]);
+const HAVE_FORMS = new Set(['have', 'has', 'had', 'having', "'ve", "'d"]);
+/** Archaic / Early-Modern-English forms en-pos has no lexicon entries for —
+ *  ubiquitous in the verse this toolkit exists to scan. */
+const ARCHAIC_TAGS = {
+    thou: 'PRP', thee: 'PRP', ye: 'PRP',
+    thy: 'PRP$', thine: 'PRP$',
+    art: 'VBP', wert: 'VBD', wast: 'VBD',
+    doth: 'VBZ', hath: 'VBZ', dost: 'VBZ', hast: 'VBZ', saith: 'VBZ',
+    didst: 'VBD', hadst: 'VBD', wouldst: 'MD', couldst: 'MD', shouldst: 'MD',
+    shalt: 'MD', wilt: 'MD', canst: 'MD', mayst: 'MD', 'mightst': 'MD',
+    wherefore: 'WRB', whither: 'WRB', whence: 'WRB',
+    hither: 'RB', thither: 'RB', yon: 'JJ', yonder: 'RB',
+    ere: 'IN', oft: 'RB', anon: 'RB',
+};
+/**
+ * Correct a sentence's tags in place-safe fashion (returns a new array).
+ * `tokens` and `tags` are the en-pos outputs, index-aligned.
+ */
+export function correctTags(tokens, tags) {
+    const out = tags.slice();
+    for (let i = 0; i < tokens.length; i++) {
+        const w = tokens[i].toLowerCase();
+        // 1. The pronoun "I".  en-norm lowercases sentence-initial "I" → "i",
+        //    which en-pos then reads as a foreign word / letter name (FW).
+        if (w === 'i' && out[i] === 'FW')
+            out[i] = 'PRP';
+        // 2. Archaic forms (thou/thy/doth/shalt/wherefore…): en-pos guesses
+        //    NN/JJ/FW for these, wrecking both stress class and the parse.
+        //    Guard "art": only when a pronoun precedes ("thou art"), since the
+        //    noun reading ("the art of…") is the modern default.
+        const archaic = ARCHAIC_TAGS[w];
+        if (archaic && !/^(NNP|NNPS)$/.test(out[i])) {
+            if (w === 'art') {
+                const prev = i > 0 ? tokens[i - 1].toLowerCase() : '';
+                if (prev === 'thou' || prev === 'ye' || prev === 'you')
+                    out[i] = 'VBP';
+            }
+            else {
+                out[i] = archaic;
+            }
+        }
+        // 3. Perfect-tense zero participles: have-form + ("quit"/"put"/"read"…)
+        //    tagged as NN/VBP/VBD → VBN, so en-parse builds the verb chain
+        //    instead of treating the participle as a direct-object noun
+        //    ("I had quit the programming paradigm").  An intervening adverb
+        //    ("had just quit") is allowed.
+        if (ZERO_PARTICIPLES.has(w) && /^(NN|NNS|VBP|VBD|VB)$/.test(out[i])) {
+            const prev1 = i > 0 ? tokens[i - 1].toLowerCase() : '';
+            const prev2 = i > 1 ? tokens[i - 2].toLowerCase() : '';
+            const prev1IsAdv = i > 0 && /^RB/.test(out[i - 1]);
+            if (HAVE_FORMS.has(prev1) || (prev1IsAdv && HAVE_FORMS.has(prev2))) {
+                out[i] = 'VBN';
+            }
+        }
+        // 4. Impossible gerunds: a VBG tag on a token that does not end in
+        //    -ing/-in' cannot be a gerund/present participle — it is an en-pos
+        //    lexicon glitch.  The right tag depends on context: before a noun it
+        //    is a noun modifier ("wisdom"/VBG teeth → NN); after a subject
+        //    pronoun it is a finite verb ("as they bicycle/VBG through" → VBP,
+        //    which keeps "through" a phrasal particle in the parse).  With no
+        //    deciding context, leave the tag alone (en-parse treats VBG
+        //    verb-ishly, the safer default).
+        if (out[i] === 'VBG' && !/in[g'’]?$/.test(w)) {
+            const prevTag = i > 0 ? out[i - 1] : '';
+            const nextTag = i + 1 < tokens.length ? out[i + 1] : '';
+            if (/^NNS?$/.test(nextTag))
+                out[i] = 'NN';
+            else if (prevTag === 'PRP')
+                out[i] = 'VBP';
+        }
+        // 5. Vocative "O" ("O wild West Wind"): en-pos gives NNP/JJ; it is an
+        //    interjection (and must not become a content word with a beat by
+        //    default).  Only the bare capital O — "o'er" etc. are handled by the
+        //    aphaeresis lexicon in stress.ts.
+        if (tokens[i] === 'O' && i + 1 < tokens.length && out[i] !== 'UH')
+            out[i] = 'UH';
+    }
+    return out;
+}

package/dist/display.d.ts

@@ -28,11 +28,5 @@ export interface ReadingStanza {
  * the rest of the line.  No original character is ever dropped.
  */
 export declare function projectStressOntoLine(rawLine: string, words: ClsWord[]): string;
-/**
- * Reading view: the poem itself in its original formatting, each syllable
- * coloured by 4-tier relative stress, followed by a same-structure block of
- * per-line stress maps + meter (with top-3 fit scores).  This is the whole
- * output for this mode — not the full per-line analytic dump.
- */
 export declare function renderReadingView(stanzas: ReadingStanza[]): string;
 //# sourceMappingURL=display.d.ts.map

package/dist/display.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"display.d.ts","sourceRoot":"","sources":["../src/display.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,OAAO,EAMP,UAAU,EAGX,MAAM,YAAY,CAAC;AAqJpB,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAkPjF;AAMD,wBAAgB,YAAY,IAAI,MAAM,CAUrC;AAuHD;;;;GAIG;AACH,wBAAgB,gBAAgB,IAAI,MAAM,CAsBzC;AAqCD,4DAA4D;AAC5D,MAAM,WAAW,WAAW;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,EAAE,UAAU,EAAE,CAAC;CACvB;AAED,4DAA4D;AAC5D,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,WAAW,EAAE,CAAC;CACtB;AAmCD;;;;;;;;;;GAUG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,MAAM,CAuC/E;AAmMD;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,MAAM,CAsDlE"}
+{"version":3,"file":"display.d.ts","sourceRoot":"","sources":["../src/display.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,OAAO,EAMP,UAAU,EAGX,MAAM,YAAY,CAAC;AAgNpB,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,UAAU,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAySjF;AAMD,wBAAgB,YAAY,IAAI,MAAM,CAUrC;AAuHD;;;;GAIG;AACH,wBAAgB,gBAAgB,IAAI,MAAM,CAsBzC;AAqCD,4DAA4D;AAC5D,MAAM,WAAW,WAAW;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,EAAE,UAAU,EAAE,CAAC;CACvB;AAED,4DAA4D;AAC5D,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,WAAW,EAAE,CAAC;CACtB;AAmCD;;;;;;;;;;GAUG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,MAAM,CAuC/E;AA2OD,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,MAAM,CA0FlE"}