WordNet synonyms + double-click opens definitions

- mode=syn now served locally by NLTK WordNet (already a g2p-en dep),
  sense-grouped by part of speech and frequency-ranked; Datamuse gone
- Double-clicking a word in the draft opens its definition card; the
  card grew a Rhymes button for the old jump
- NLTK wordnet data baked into the Docker image and CI

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -18,5 +18,5 @@ jobs:
       # under both its old and new (nltk>=3.9) names
       - run: |
           uv run python -c "import nltk;
-          [nltk.download(p, quiet=True) for p in ('averaged_perceptron_tagger','averaged_perceptron_tagger_eng','cmudict')]"
+          [nltk.download(p, quiet=True) for p in ('averaged_perceptron_tagger','averaged_perceptron_tagger_eng','cmudict','wordnet','omw-1.4')]"
       - run: uv run pytest

Dockerfile

@@ -16,9 +16,9 @@ RUN uv sync --frozen --no-dev --no-install-project
 COPY app.py ./
 COPY static ./static
 
-# Pre-warm NLTK data so the g2p-en fallback never downloads at request time
+# Pre-warm NLTK data so g2p-en and WordNet never download at request time
 RUN uv run python -c "import nltk;\
-[nltk.download(p, quiet=True) for p in ('averaged_perceptron_tagger','averaged_perceptron_tagger_eng','cmudict')]" || true
+[nltk.download(p, quiet=True) for p in ('averaged_perceptron_tagger','averaged_perceptron_tagger_eng','cmudict','wordnet','omw-1.4')]" || true
 
 EXPOSE 8000
 

README.md

@@ -31,9 +31,11 @@ Then open <http://127.0.0.1:8000>.
 ## API
 
 - `POST /api/analyze` `{"text": "..."}` → token spans, rhyme groups,
-  per-stanza schemes
-- `GET /api/lookup?word=light&mode=rhyme|near` → frequency-ranked
-  rhymes / near rhymes, grouped by syllable count
+  per-stanza schemes, per-line meter
+- `GET /api/lookup?word=light&mode=rhyme|near|syn` → frequency-ranked
+  rhymes / near rhymes (by syllable count) and WordNet synonyms (by
+  part of speech)
 
-Synonyms come from the free [Datamuse](https://www.datamuse.com/api/)
-API, client-side.
+Dictionary definitions in the UI come from the free
+[dictionaryapi.dev](https://dictionaryapi.dev/); everything else is
+served locally.

app.py

@@ -27,6 +27,10 @@ async def lifespan(app: FastAPI):
         g2p_phones("warmup")
     except Exception:
         pass  # model unavailable; spelling fallbacks still work
+    try:
+        get_wordnet()
+    except Exception:
+        pass
     get_slant_index()
     yield
 
@@ -675,6 +679,46 @@ def get_slant_index() -> dict[str, set[str]]:
     return _slant_index
 
 
+_wordnet = None
+
+
+def get_wordnet():
+    """WordNet via NLTK (already a g2p-en dependency), data on demand."""
+    global _wordnet
+    if _wordnet is None:
+        from nltk.corpus import wordnet as wn
+        try:
+            wn.synsets("test")
+        except LookupError:
+            import nltk
+            nltk.download("wordnet", quiet=True)
+            nltk.download("omw-1.4", quiet=True)
+        _wordnet = wn
+    return _wordnet
+
+
+POS_NAMES = {"n": "noun", "v": "verb", "a": "adjective",
+             "s": "adjective", "r": "adverb"}
+
+
+def synonyms_for(w: str, limit: int) -> list[dict]:
+    """Sense-grouped synonyms from WordNet, frequency-ranked."""
+    wn = get_wordnet()
+    seen: dict[str, str] = {}
+    for ss in wn.synsets(w):
+        lemmas = list(ss.lemmas())
+        if ss.pos() in ("a", "s"):  # adjectives: pull in the satellites
+            for sim in ss.similar_tos():
+                lemmas.extend(sim.lemmas())
+        for lemma in lemmas:
+            name = lemma.name().replace("_", " ").lower()
+            if name != w and re.fullmatch(r"[a-z' -]+", name):
+                seen.setdefault(name, POS_NAMES.get(ss.pos(), ss.pos()))
+    ranked = sorted(seen.items(),
+                    key=lambda kv: (-zipf_frequency(kv[0], "en"), kv[0]))
+    return [{"word": n, "pos": p} for n, p in ranked[:limit]]
+
+
 def _ranked(words, exclude: set[str], limit: int) -> list[dict]:
     scored = []
     for w in set(words):
@@ -695,6 +739,9 @@ def _ranked(words, exclude: set[str], limit: int) -> list[dict]:
 @app.get("/api/lookup")
 def lookup(word: str, mode: str = "rhyme", limit: int = 60):
     w = word.strip().lower()
+    if mode == "syn":
+        words = synonyms_for(w, limit)
+        return {"word": w, "mode": mode, "known": bool(words), "words": words}
     phones = phones_for(w)
     if not phones:
         return {"word": w, "mode": mode, "known": False, "words": []}

static/index.html

@@ -200,6 +200,7 @@
     text-transform: uppercase; letter-spacing: .06em; margin-right: 4px;
   }
   .btn.small { padding: 5px 10px; font-size: 12px; }
+  .def-actions { display: flex; gap: 8px; }
   .chip {
     font-size: 13px; background: var(--panel-2); border: 1px solid var(--line);
     color: var(--ink); padding: 5px 10px; border-radius: 999px; cursor: pointer;
@@ -277,7 +278,7 @@ Double-click any word to look it up on the right."></textarea>
       </div>
       <div id="defBox"></div>
       <div id="lookupResults">
-        <p class="muted">Type a word and hit Go, or double-click a word in your draft. Click any result to see its definition.</p>
+        <p class="muted">Type a word and hit Go for rhymes. Double-click a word in your draft for its definition. Click any result chip for its definition too.</p>
       </div>
       <div class="legend">
         <div class="res-label">Current stanza</div>
@@ -544,7 +545,11 @@ editor.addEventListener('click', ()=>{ buildReadout(); buildLegend(); });
 editor.addEventListener('dblclick', ()=>{
   const sel = editor.value.slice(editor.selectionStart, editor.selectionEnd)
     .trim().replace(/[^A-Za-z']/g,'');
-  if(sel){ document.getElementById('lookupInput').value = sel; doLookup(); }
+  if(sel){
+    document.getElementById('lookupInput').value = sel;
+    tab('lookup');
+    showDefinition(sel);
+  }
 });
 
 /* ---------- insert at cursor ---------- */
@@ -582,13 +587,15 @@ async function doLookup(){
   defBox.innerHTML = '';
   resultsBox.innerHTML = '<p class="muted">Searching…</p>';
   try{
+    const r = await fetch(`/api/lookup?word=${encodeURIComponent(word)}&mode=${mode}`);
+    const data = await r.json();
     if(mode === 'syn'){
-      const r = await fetch(`https://api.datamuse.com/words?rel_syn=${encodeURIComponent(word)}&ml=${encodeURIComponent(word)}&max=60`);
-      const data = await r.json();
-      renderChips(`Synonyms / related for “${esc(word)}”`, data.map(d=>d.word));
+      if(!data.known){
+        resultsBox.innerHTML = `<p class="muted">No synonyms found for “${esc(word)}”.</p>`;
+        return;
+      }
+      renderByPos(word, data.words);
     }else{
-      const r = await fetch(`/api/lookup?word=${encodeURIComponent(word)}&mode=${mode}`);
-      const data = await r.json();
       if(!data.known){
         resultsBox.innerHTML = `<p class="muted">“${esc(word)}” isn't in the pronunciation dictionary.</p>`;
         return;
@@ -619,9 +626,16 @@ function defCard(word, inner){
       <div class="defhead"><b>${esc(word)}</b><span class="phon" id="defPhon"></span>
         <span class="defx" title="Dismiss">×</span></div>
       ${inner}
-      <div class="def-actions"><button class="btn small" id="defInsert">Insert at cursor</button></div>
+      <div class="def-actions">
+        <button class="btn small" id="defInsert">Insert at cursor</button>
+        <button class="btn small" id="defRhymes">Rhymes</button>
+      </div>
     </div>`;
   defBox.querySelector('#defInsert').addEventListener('click', ()=> insertAtCursor(word));
+  defBox.querySelector('#defRhymes').addEventListener('click', ()=>{
+    document.getElementById('lookupInput').value = word;
+    doLookup();
+  });
   defBox.querySelector('.defx').addEventListener('click', ()=>{ defBox.innerHTML=''; });
 }
 async function showDefinition(word){
@@ -647,6 +661,21 @@ function renderChips(label, words){
   resultsBox.innerHTML = `<div class="res-label">${label}</div>` + chipHtml(words.slice(0,50));
   wireChips();
 }
+function renderByPos(word, items){
+  if(!items.length){ resultsBox.innerHTML = `<p class="muted">No synonyms for “${esc(word)}”.</p>`; return; }
+  const byPos = {};
+  items.forEach(d=>{ (byPos[d.pos || 'related'] ||= []).push(d.word); });
+  let h = `<div class="res-label">Synonyms for “${esc(word)}”</div>`;
+  ['noun','verb','adjective','adverb','related'].forEach(pos=>{
+    if(!byPos[pos]) return;
+    h += `<div class="res-label">${pos}s</div>` + chipHtml(byPos[pos]);
+    delete byPos[pos];
+  });
+  Object.keys(byPos).forEach(pos=>{ h += `<div class="res-label">${esc(pos)}</div>` + chipHtml(byPos[pos]); });
+  resultsBox.innerHTML = h;
+  wireChips();
+}
+
 function renderBySyllable(word, items, label){
   if(!items.length){ resultsBox.innerHTML = `<p class="muted">No results for “${esc(word)}”.</p>`; return; }
   const bySyl = {};

tests/test_rhymes.py

@@ -297,3 +297,14 @@ def test_perfect_subgroup_fuses_with_slant_family():
             "Unplugged toaster")
     group_with(text, "soldier", "shoulder", "older", "colder", "holster",
                "coaster", "roaster", "poster", "toaster")
+
+
+def test_lookup_synonyms_wordnet():
+    words = {w["word"] for w in lookup("happy", mode="syn")["words"]}
+    assert "glad" in words
+    big = {w["word"]: w["pos"] for w in lookup("big", mode="syn")["words"]}
+    assert "large" in big and big["large"] == "adjective"
+
+
+def test_lookup_synonyms_unknown_word():
+    assert lookup("xqzzqx", mode="syn")["known"] is False