kennethreitz/rhymepad.org
1
0
Fork 0
mirror of https://github.com/kennethreitz/rhymepad.org.git synced 2026-06-11 17:08:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

Overlapping rhyme layers: phrases as a second visual channel

Browse Source 
Phrases now compete in the perfect-rhyme pass anchored at their first
word's stressed vowel ("stir up" = ER AH P = "syrup"), render as
underlines while words render as fills, and can overlap in different
colors — the syrup/cup knot finally displays both schemes. Stopword-
anchored phrases attach only on exact rime identity. Segment-based
highlight renderer replaces the span-per-token one.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Kenneth Reitz
2026-06-07 01:41:49 -04:00
parent 7eaf815e5b
commit c60b9d6d43
3 changed files with 106 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app.py
+45 -5
View File
@@ -348,20 +348,33 @@ def analyze(draft: Draft):
line_toks[t["line"]].append(t)
for toks in line_toks.values():
for a, b in zip(toks, toks[1:]):
if a["word"].lower() in STOPWORDS or b["word"].lower() in STOPWORDS:
continue
pa, pb = phones_for(a["word"]), phones_for(b["word"])
if not (pa and pb):
continue
# the phrase's rime runs from the anchor's stressed vowel
# through the end of the tail: "stir up" = ER AH P, which is
# a PERFECT rhyme with syrup
pl = pa.split()
start = 0
for i in range(len(pl) - 1, -1, -1):
if pl[i][-1] in "12":
start = i
break
phrases.append({
"line": a["line"], "start": a["start"], "end": b["end"],
"word": a["word"].lower() + " " + b["word"].lower(),
"is_end": b["is_end"], "sid": a["sid"], "gid": None,
"slant": False, "vowels": _tail_vowels(pa) + _all_vowels(pb),
"rime": DIGITS.sub("", " ".join(pl[start:] + pb.split())),
# a phrase touching a stopword ("were up") may still match
# perfectly, but never competes in the vowel-only passes
"weak": (a["word"].lower() in STOPWORDS
or b["word"].lower() in STOPWORDS),
})
# pass 1: perfect rhymes (shared rime), anywhere in a line — this is
# what catches internal rhymes
# what catches internal rhymes. Phrases compete too, so "stir up"
# perfect-rhymes "syrup" even while its "up" rhymes with "cup".
by_rime = defaultdict(list)
for t in tokens:
w = t["word"].lower()
@@ -369,6 +382,10 @@ def analyze(draft: Draft):
continue
for key in rime_keys(t["word"]):
by_rime[key].append(t)
for p in phrases:
# "bought it" competes; "to me" / "but I" never found anything
if p["word"].split()[0] not in STOPWORDS:
by_rime["p:" + p["rime"]].append(p)
# biggest buckets claim their tokens first, so a word with several
# candidate pronunciations joins its best-supported rhyme group.
@@ -379,7 +396,9 @@ def analyze(draft: Draft):
toks = [t for t in toks if id(t) not in claimed]
if len(toks) < 2:
continue
distinct = {t["word"].lower() for t in toks}
# distinctness by anchor word, so "fire burns" can't pose as a
# rhyme partner for the "fire" it starts with
distinct = {t["word"].split()[0].lower() for t in toks}
end_count = sum(t["is_end"] for t in toks)
if len(distinct) < 2 and end_count < 2:
continue # the same word repeated mid-line isn't a rhyme
@@ -447,6 +466,8 @@ def analyze(draft: Draft):
if id(t) in grouped:
grouped_spans[t["line"]].append((t["start"], t["end"]))
for p in phrases:
if p["weak"]:
continue
if any(s < p["end"] and p["start"] < e
for s, e in grouped_spans[p["line"]]):
continue
@@ -487,6 +508,25 @@ def analyze(draft: Draft):
for (sid, key), toks in by_vc.items():
if len(toks) >= 2 and len({t["word"].lower() for t in toks}) >= 2:
raw_groups.append({"toks": toks, "slant": True, "key": key})
grouped.update(id(t) for t in toks)
# stopword-anchored phrases ("were up") never compete on their own,
# but an exact rime match against any grouped token lets them ride
# along — perfect phone identity carries no transitive-key risk
rime_map: dict[tuple, int] = {}
for gi, g in enumerate(raw_groups):
for t in g["toks"]:
keys = ("p:" + t["rime"],) if "rime" in t else rime_keys(t["word"])
for k in keys:
if k.startswith("p:"):
rime_map.setdefault((t["sid"], k), gi)
for p in phrases:
if id(p) in grouped or p["word"].split()[0] not in STOPWORDS:
continue
gi = rime_map.get((p["sid"], "p:" + p["rime"]))
if gi is not None:
raw_groups[gi]["toks"].append(p)
grouped.add(id(p))
# stable colors: order groups by first appearance
raw_groups.sort(key=lambda g: min((t["line"], t["start"]) for t in g["toks"]))
@@ -536,7 +576,7 @@ def analyze(draft: Draft):
toks_out = [
{"l": t["line"], "s": t["start"], "e": t["end"], "g": t["gid"],
"end": t["is_end"],
"end": t["is_end"], "ph": "vowels" in t,
"slant": t["slant"] or groups_out[t["gid"]]["slant"]}
for t in [*tokens, *phrases] if t["gid"] is not None
]
static/index.html
+36 -24
View File
@@ -111,19 +111,11 @@
outline: none;
}
#editor::placeholder { color: #5a5249; }
/* colored rhyme tokens — background tint only, so the textarea text on
top stays crisp and box metrics stay identical (no horizontal padding!) */
.rh {
border-radius: 4px;
background: color-mix(in srgb, var(--c) 15%, transparent);
}
.rh.end {
background: color-mix(in srgb, var(--c) 24%, transparent);
box-shadow: inset 0 -2px 0 0 var(--c);
}
.rh.slant.end {
box-shadow: inset 0 -2px 0 0 color-mix(in srgb, var(--c) 45%, transparent);
}
/* colored rhyme segments — words tint the background, phrases draw the
underline; both can overlap in different colors. Background/shadow
only, so the textarea text on top stays crisp and box metrics stay
identical (no horizontal padding!) */
.seg { border-radius: 4px; }
.offbeat {
text-decoration: underline wavy color-mix(in srgb, var(--r6) 45%, transparent);
text-decoration-skip-ink: none;
@@ -484,21 +476,41 @@ function render(){
analysis.groups.forEach(g=>{ groupInfo[g.id] = g; });
analysis.tokens.forEach(t=>{ (tokByLine[t.l] ||= []).push(t); });
}
const colorOf = t => `var(--r${groupInfo[t.g].color % COLORS})`;
let html = '';
lines.forEach((line, i)=>{
// only apply spans if this line still matches what the server analyzed
const fresh = analysis && analysis.lines[i] === line;
const toks = fresh ? (tokByLine[i] || []) : [];
toks.sort((a,b)=>a.s-b.s);
let pos = 0, h = '';
toks.forEach(t=>{
const g = groupInfo[t.g]; if(!g || t.s < pos) return;
h += esc(line.slice(pos, t.s));
const cls = 'rh' + (t.end ? ' end' : '') + (t.slant ? ' slant' : '');
h += `<span class="${cls}" style="--c:var(--r${g.color % COLORS})">` + esc(line.slice(t.s, t.e)) + '</span>';
pos = t.e;
});
h += esc(line.slice(pos));
const toks = (fresh ? (tokByLine[i] || []) : []).filter(t=>groupInfo[t.g]);
const words = toks.filter(t=>!t.ph);
const phrases = toks.filter(t=>t.ph);
let h = '';
if(!toks.length){
h = esc(line);
}else{
const cuts = new Set([0, line.length]);
toks.forEach(t=>{ cuts.add(t.s); cuts.add(t.e); });
const pts = [...cuts].sort((a,b)=>a-b);
for(let k = 0; k < pts.length - 1; k++){
const a = pts[k], b = pts[k+1];
const text = esc(line.slice(a, b));
if(!text) continue;
const w = words.find(t=>t.s <= a && b <= t.e);
const p = phrases.find(t=>t.s <= a && b <= t.e);
if(!w && !p){ h += text; continue; }
let style = '';
if(w) style += `background:color-mix(in srgb, ${colorOf(w)} ${w.end ? 24 : 15}%, transparent);`;
let ul = null;
if(w && w.end){
ul = w.slant ? `color-mix(in srgb, ${colorOf(w)} 45%, transparent)` : colorOf(w);
}else if(p){
ul = p.end && !p.slant ? colorOf(p)
: `color-mix(in srgb, ${colorOf(p)} 55%, transparent)`;
}
if(ul) style += `box-shadow:inset 0 -2px 0 0 ${ul};`;
h += `<span class="seg" style="${style}">${text}</span>`;
}
}
const m = fresh && analysis.meter ? analysis.meter.find(x=>x.l===i) : null;
html += (m && m.off ? `<span class="offbeat">${h}</span>` : h) + '\n';
});
tests/test_rhymes.py
+25
View File
@@ -239,3 +239,28 @@ def test_meter_coaching_needs_a_pattern_to_break():
# two lines of wildly different lengths: no dominant pattern, no flags
res = analyze(Draft(text="short line here\na very much longer line that runs on and on"))
assert not any(m["off"] for m in res["meter"])
# ------------------------------------------------------------------ layers
def test_layered_phrase_rides_second_group():
# Em's syrup knot: "up" fills with the cup chain, while "stir up"
# rides the syrup/burden group as a second layer
text = ("Maybe your cup is full of syrup and lean\n"
"fill the cup up to the brim\n"
"Maybe I need to stir up shit\n"
"the burden is mine")
syrup = group_with(text, "syrup", "burden", "stir up")
cups = group_with(text, "cup", "up")
assert "stir up" not in cups and "syrup" not in cups
def test_weak_phrase_attaches_but_never_founds():
# "were up" (stopword tail) joins an existing ER group...
text = ("Maybe your cup is full of syrup and lean\n"
"fill the cup up to the brim\n"
"shake the world up if it were up to me\n"
"the burden is mine")
group_with(text, "syrup", "burden", "were up")
# ...but two weak phrases alone can't create a group
assert "were up" not in highlighted("it were up to him\nit were up to her")