Update app.py

app.py

@@ -1,4 +1,4 @@
-# app.py — Universal Conlang Translator (Max Compresión Exacta)
+# app.py — Universal Conlang Translator (Max Compresión Exacta) — UI bilingüe ES/EN
 # Archivos necesarios en la raíz:
 #  - lexicon_minimax.json
 #  - lexicon_komin.json
@@ -15,8 +15,7 @@ import re
 import json
 import base64
 import zlib
-import hashlib
-from typing import Dict, Tuple, Optional
+from typing import Dict, Optional
 import gradio as gr
 
 # ------------ Archivos esperados ------------
@@ -78,7 +77,7 @@ def load_lexicons():
  EN2MINI, EN2KOMI, MINI2EN, KOMI2EN,
  ES2EN_LEMMA, EN2ES_LEMMA) = load_lexicons()
 
-# ------------ OOV reversible (modo Semi-lossless) ------------
+# ------------ OOV reversible (Semi-lossless) ------------
 ALPHA_MINI64 = "@ptkmnslraeiouy0123456789><=:/!?.+-_*#bcdfghjvqwxzACEGHIJKLMNOPRS"[:64]
 CJK_BASE = (
     "天地人日月山川雨風星火水木土金石光影花草鳥犬猫魚"
@@ -141,47 +140,7 @@ def lemma_of(tok, src_lang: str) -> str:
     else:
         return norm_en(tok.lemma_ if tok.lemma_ else tok.text)
 
-# ------------ Selección de oración predicativa ------------
-def pick_predicative_sentence(doc):
-    sents = list(doc.sents) if doc.has_annotation("SENT_START") else [doc]
-    candidates = []
-    for s in sents:
-        roots = [t for t in s if t.dep_ == "ROOT" and t.pos_ in ("VERB","AUX")]
-        if not roots:
-            continue
-        root = roots[0]
-        has_q = "?" in s.text
-        has_subj = any(t.dep_.startswith("nsubj") for t in root.children)
-        score = (1 if has_q else 0) + (1 if has_subj else 0) + (len(s) / 1000.0)
-        candidates.append((score, s))
-    if not candidates:
-        return doc
-    return sorted(candidates, key=lambda x: x[0], reverse=True)[0][1].as_doc()
-
-def is_content_token(t) -> bool:
-    return True  # No filtra para exactitud
-
-# ------------ Mapeo lema→código ------------
-def code_es(lemma: str, target: str) -> str:
-    lemma = norm_es(lemma)
-    if target == "Minimax-ASCII":
-        return ES2MINI.get(lemma) or enc_oov_minimax(lemma)
-    else:
-        return ES2KOMI.get(lemma) or enc_oov_komin(lemma)
-
-def code_en(lemma: str, target: str) -> str:
-    lemma = norm_en(lemma)
-    if target == "Minimax-ASCII":
-        if EN2MINI: return EN2MINI.get(lemma) or enc_oov_minimax(lemma)
-        return enc_oov_minimax(lemma)
-    else:
-        if EN2KOMI: return EN2KOMI.get(lemma) or enc_oov_komin(lemma)
-        return enc_oov_komin(lemma)
-
-# ------------ Fraseador compacto ------------
-TAM_MINI = {"Pres":"P", "Past":"T", "Fut":"F", "UNK":"P"}
-TAM_KOMI = {"Pres":"Ⓟ", "Past":"Ⓣ", "Fut":"Ⓕ", "UNK":"Ⓟ"}
-
+# ------------ Utilidades de análisis sintáctico ------------
 def detect_polarity(doc) -> bool:
     return "?" in doc.text
 
@@ -255,6 +214,26 @@ def _person_of_doc(doc, src_lang: str) -> Optional[str]:
     except Exception:
         return None
 
+# ------------ Mapeo lema→código y fraseadores ------------
+def code_es(lemma: str, target: str) -> str:
+    lemma = norm_es(lemma)
+    if target == "Minimax-ASCII":
+        return ES2MINI.get(lemma) or enc_oov_minimax(lemma)
+    else:
+        return ES2KOMI.get(lemma) or enc_oov_komin(lemma)
+
+def code_en(lemma: str, target: str) -> str:
+    lemma = norm_en(lemma)
+    if target == "Minimax-ASCII":
+        if EN2MINI: return EN2MINI.get(lemma) or enc_oov_minimax(lemma)
+        return enc_oov_minimax(lemma)
+    else:
+        if EN2KOMI: return EN2KOMI.get(lemma) or enc_oov_komin(lemma)
+        return enc_oov_komin(lemma)
+
+TAM_MINI = {"Pres":"P", "Past":"T", "Fut":"F", "UNK":"P"}
+TAM_KOMI = {"Pres":"Ⓟ", "Past":"Ⓣ", "Fut":"Ⓕ", "UNK":"Ⓟ"}
+
 def realize_minimax(doc, src_lang: str, drop_articles=True, zero_copula=True, semi_lossless=False, person_hint="2s"):
     root, subs, objs, obls, advs = extract_core(doc)
     tense = detect_tense(root)
@@ -275,21 +254,14 @@ def realize_minimax(doc, src_lang: str, drop_articles=True, zero_copula=True, se
     def realize_np(tokens):
         outs=[]
         for t in tokens:
-            if not USE_SPACY or is_content_token(t):
-                lem  = lemma_of(t, src_lang) if USE_SPACY else (t.text)
-                code = code_es(lem, "Minimax-ASCII") if src_lang=="Español" else code_en(lem, "Minimax-ASCII")
-                if semi_lossless and USE_SPACY and (t.tag_ in ("NNS","NNPS") or "Number=Plur" in str(t.morph)):
-                    code = f"{code}[PL]"
-                outs.append(code)
+            lem  = lemma_of(t, src_lang) if USE_SPACY else (t.text)
+            code = code_es(lem, "Minimax-ASCII") if src_lang=="Español" else code_en(lem, "Minimax-ASCII")
+            outs.append(code)
         return outs
 
     S = realize_np(subs)
     O = realize_np(objs) + realize_np(obls)
-    ADV=[]
-    for a in advs:
-        if not USE_SPACY or is_content_token(a):
-            lem  = lemma_of(a, src_lang) if USE_SPACY else a.text
-            ADV.append(code_es(lem, "Minimax-ASCII") if src_lang=="Español" else code_en(lem, "Minimax-ASCII"))
+    ADV=[code_es(lemma_of(a, src_lang), "Minimax-ASCII") if src_lang=="Español" else code_en(lemma_of(a, src_lang), "Minimax-ASCII") for a in advs] if USE_SPACY else []
 
     if zero_copula and not semi_lossless and vlem in ("ser","estar","be") and tense=="Pres" and not is_neg and not is_q:
         parts = S + O + ADV
@@ -315,21 +287,14 @@ def realize_komin(doc, src_lang: str, drop_articles=True, zero_copula=True, semi
     def realize_np(tokens, particle):
         outs=[]
         for t in tokens:
-            if not USE_SPACY or is_content_token(t):
-                lem  = lemma_of(t, src_lang) if USE_SPACY else t.text
-                code = code_es(lem, "Kōmín-CJK") if src_lang=="Español" else code_en(lem, "Kōmín-CJK")
-                if semi_lossless and USE_SPACY and (t.tag_ in ("NNS","NNPS") or "Number=Plur" in str(t.morph)):
-                    code = f"{code}[PL]"
-                outs.append(code + particle)
+            lem  = lemma_of(t, src_lang) if USE_SPACY else t.text
+            code = code_es(lem, "Kōmín-CJK") if src_lang=="Español" else code_en(lem, "Kōmín-CJK")
+            outs.append(code + particle)
         return outs
 
     S = realize_np(subs, P_SUBJ)
     O = realize_np(objs + obls, P_OBJ)
-    ADV=[]
-    for a in advs:
-        if not USE_SPACY or is_content_token(a):
-            lem  = lemma_of(a, src_lang) if USE_SPACY else a.text
-            ADV.append(code_es(lem, "Kōmín-CJK") if src_lang=="Español" else code_en(lem, "Kōmín-CJK"))
+    ADV=[code_es(lemma_of(a, src_lang), "Kōmín-CJK") if src_lang=="Español" else code_en(lemma_of(a, src_lang), "Kōmín-CJK") for a in advs] if USE_SPACY else []
 
     v_form = vcode + TAM + (NEG_M if is_neg else "")
 
@@ -341,7 +306,7 @@ def realize_komin(doc, src_lang: str, drop_articles=True, zero_copula=True, semi
     if is_q: out += " " + Q_FIN
     return out
 
-# ------------ Lossless (Base85 comprimido) ------------
+# ------------ Sidecars para compresión exacta ------------
 SIDECAR_B85_RE = re.compile(r"\s?§\((?P<b85>[A-Za-z0-9!#$%&()*+\-;<=>?@^_`{|}~]+)\)$")
 
 def b85_enc_raw(s: str) -> str:
@@ -367,7 +332,6 @@ def extract_sidecar_b85(text: str) -> Optional[str]:
 def strip_sidecar_b85(text: str) -> str:
     return SIDECAR_B85_RE.sub("", text).rstrip()
 
-# ------------ Custom sidecar para max compresión exacta ------------
 def custom_sidecar_enc(conlang_text: str, original_text: str) -> str:
     comp = zlib.compress(original_text.encode("utf-8"), 9)
     blob = to_custom_b64(comp, ALPHA_MINI64)
@@ -386,7 +350,7 @@ def extract_custom_sidecar(text: str) -> Optional[str]:
 def strip_custom_sidecar(text: str) -> str:
     return text.split('~')[0].rstrip() if '~' in text else text
 
-# ------------ Codificar / Decodificar léxisco puro ------------
+# ------------ Codificar/decodificar léxico puro ------------
 def encode_simple(text: str, src_lang: str, target: str) -> str:
     if not text.strip(): return ""
     def repl_es(m):
@@ -419,9 +383,6 @@ def pluralize_en(word: str) -> str:
 def pluralize(word: str, tgt_lang: str) -> str:
     return pluralize_es(word) if tgt_lang == "Español" else pluralize_en(word)
 
-PRON_ES = {"yo", "tú", "él", "ella", "nosotros", "vosotros", "ellos", "ellas", "usted", "ustedes"}
-PRON_EN = {"i", "you", "he", "she", "it", "we", "they"}
-
 mini_tail_re = re.compile(r"^(?P<stem>.+?)·(?P<tail>[PTFNQ12sp]+)$")
 
 def decode_simple(text: str, source: str, tgt_lang: str) -> str:
@@ -429,7 +390,6 @@ def decode_simple(text: str, source: str, tgt_lang: str) -> str:
         return ""
     code2es = MINI2ES if source=="Minimax-ASCII" else KOMI2ES
     code2en = MINI2EN if source=="Minimax-ASCII" else KOMI2EN
-    pron_set = PRON_ES if tgt_lang == "Español" else PRON_EN
 
     if source == "Kōmín-CJK":
         text = text.replace("？", "?").replace(" ", " ")
@@ -447,7 +407,7 @@ def decode_simple(text: str, source: str, tgt_lang: str) -> str:
     has_q = False
     is_neg = False
 
-    for i, part in enumerate(tokens):
+    for part in tokens:
         look = part.replace("[PL]", "")
         had_pl = "[PL]" in part
         pl_flags.append(had_pl)
@@ -468,7 +428,6 @@ def decode_simple(text: str, source: str, tgt_lang: str) -> str:
             lemma_tokens.append(vlem)
             pl_flags.append(False)
 
-            # Parse tail
             if tail:
                 if len(tail) > 0 and tail[0] in "PTF":
                     verb_tense = {"P": "Pres", "T": "Past", "F": "Fut"}.get(tail[0], "Pres")
@@ -487,7 +446,6 @@ def decode_simple(text: str, source: str, tgt_lang: str) -> str:
             verb_lemma = vlem
             continue
 
-        # No verbo
         w_es = code2es.get(look)
         w_en = code2en.get(look) if code2en else None
         w = w_es if tgt_lang == "Español" else (w_en or w_es or look)
@@ -500,50 +458,23 @@ def decode_simple(text: str, source: str, tgt_lang: str) -> str:
         pl_flags.append(had_pl)
 
     out_parts = []
-    greeting = None
-    wh = None
     for idx, lem in enumerate(lemma_tokens):
         if idx == verb_idx:
             conj_func = _es_conj if tgt_lang == "Español" else _en_conj
             v_conj = conj_func(verb_lemma, verb_tense, verb_person)
             if is_neg:
-                neg_prefix = "no " if tgt_lang == "Español" else "not "
-                v_conj = neg_prefix + v_conj
+                v_conj = ("no " if tgt_lang == "Español" else "not ") + v_conj
             out_parts.append(v_conj)
             continue
+        out_parts.append(pluralize(lem, tgt_lang) if pl_flags[idx] else lem)
 
-        w = pluralize(lem, tgt_lang) if pl_flags[idx] else lem
-        if w.lower() in {"hola", "hello", "hi", "hey"}:
-            greeting = w
-        elif w.lower() in {"como", "cómo", "what", "how"} and has_q:
-            wh = w
-            if tgt_lang == "Español" and w.lower() == "como":
-                wh = "cómo"
-        else:
-            out_parts.append(w)
-
-    # Reorden: Greeting + wh + S V O ADV
-    final_out = []
-    if greeting:
-        final_out.append(greeting.capitalize())
-    if wh:
-        final_out.append(wh)
-    final_out += out_parts
-
-    out_text = " ".join(final_out)
-
-    # Pregunta
+    out_text = " ".join(out_parts)
     if has_q:
         start_q = "¿" if tgt_lang == "Español" else ""
-        end_q = "?" if tgt_lang == "Español" else "?"
-        out_text = f"{start_q}{out_text.capitalize()}{end_q}"
-
+        out_text = f"{start_q}{out_text.capitalize()}?"
     return out_text
 
 # ------------ Conjugadores mínimos ------------
-_ES_SUBJ = {"1s":"yo","2s":"tú","3s":"él/ella","1p":"nosotros","2p":"vosotros","3p":"ellos"}
-_EN_SUBJ = {"1s":"I","2s":"you","3s":"he","1p":"we","2p":"you","3p":"they"}
-
 def _es_conj_regular(lemma, tense, person):
     if not lemma.endswith(("ar","er","ir")): return lemma
     stem = lemma[:-2]; vtype = lemma[-2:]
@@ -572,7 +503,7 @@ def _es_conj(lemma, tense, person):
     if lemma == "estar":
         tab = {
             "Pres":{"1s":"estoy","2s":"estás","3s":"está","1p":"estamos","2p":"estáis","3p":"están"},
-            "Past":{"1s":"estuve","2s":"estuviste","3s":"estuvo","1p":"estuvimos","2p":"estuvisteis","3p":"estuvieron"},
+            "Past":{"1s":"estuviste","2s":"estuviste","3s":"estuvo","1p":"estuvimos","2p":"estuvisteis","3p":"estuvieron"},
             "Fut":{"1s":"estaré","2s":"estarás","3s":"estará","1p":"estaremos","2p":"estaréis","3p":"estarán"},
         }; return tab[tense].get(person, tab[tense]["3s"])
     if lemma == "ir":
@@ -616,7 +547,7 @@ def _en_conj(lemma, tense, person):
     else:
         return lemma
 
-# ------------ Semi-lossless (rutas) ------------
+# ------------ Rutas principales ------------
 def _build_with_spacy(text: str, src_lang: str, target: str,
                       drop_articles: bool, zero_copula: bool, semi_lossless: bool) -> str:
     nlp = nlp_es if src_lang=="Español" else nlp_en
@@ -629,7 +560,7 @@ def _build_with_spacy(text: str, src_lang: str, target: str,
 def build_sentence(text: str, src_lang: str, target: str,
                    drop_articles: bool, zero_copula: bool, mode: str, max_comp_exact: bool = False) -> str:
     if not text.strip(): return ""
-    semi = True  # Siempre semi-lossless
+    semi = True  # siempre semi-lossless
     core = _build_with_spacy(text, src_lang, target, drop_articles, zero_copula and not semi, semi_lossless=semi) if USE_SPACY else encode_simple(text, src_lang, target)
     if max_comp_exact:
         return custom_sidecar_enc(core, text)
@@ -640,31 +571,19 @@ def universal_translate(text: str, src: str, tgt: str,
                         mode: str, max_comp_exact: bool = False) -> str:
     if not text.strip(): return ""
     if src == tgt: return text
-
-    # Natural → Conlang
     if src in ("Español","English") and tgt in ("Minimax-ASCII","Kōmín-CJK"):
         return build_sentence(text, src, tgt, drop_articles, zero_copula, mode, max_comp_exact)
-
-    # Conlang → Natural (considera sidecars)
     if src in ("Minimax-ASCII","Kōmín-CJK") and tgt in ("Español","English"):
-        # Custom sidecar para exact
         orig = extract_custom_sidecar(text)
         if orig is not None: return orig
-        # Fallback b85 si hay
         orig = extract_sidecar_b85(text)
         if orig is not None: return orig
-        # Semi-lossless
         return decode_simple(strip_custom_sidecar(strip_sidecar_b85(text)), src, tgt)
-
-    # Natural ↔ Natural (lemas)
     if src in ("Español","English") and tgt in ("Español","English"):
         return translate_natural(text, src, tgt)
-
-    # Conlang ↔ Conlang (simplificado)
     if src in ("Minimax-ASCII","Kōmín-CJK") and tgt in ("Minimax-ASCII","Kōmín-CJK"):
         orig = extract_custom_sidecar(text)
         if orig is not None:
-            # Preserva sidecar
             core = strip_custom_sidecar(text)
             es_lemmas = decode_simple(core, src, "Español")
             words = re.findall(r"\w+|[^\w\s]+", es_lemmas)
@@ -672,28 +591,20 @@ def universal_translate(text: str, src: str, tgt: str,
             for w in words:
                 if re.fullmatch(r"\w+", w):
                     code = ES2MINI.get(norm_es(w)) if tgt=="Minimax-ASCII" else ES2KOMI.get(norm_es(w))
-                    if not code:
-                        code = enc_oov_minimax(w) if tgt=="Minimax-ASCII" else enc_oov_komin(w)
-                    out.append(code)
+                    out.append(code or (enc_oov_minimax(w) if tgt=="Minimax-ASCII" else enc_oov_komin(w)))
                 else:
                     out.append(w)
-            out_text = " ".join(out)
-            return custom_sidecar_enc(out_text, orig)
-        # Sin sidecar, normal
-        core = text
-        es_lemmas = decode_simple(core, src, "Español")
+            return custom_sidecar_enc(" ".join(out), orig)
+        es_lemmas = decode_simple(text, src, "Español")
         words = re.findall(r"\w+|[^\w\s]+", es_lemmas)
         out=[]
         for w in words:
             if re.fullmatch(r"\w+", w):
                 code = ES2MINI.get(norm_es(w)) if tgt=="Minimax-ASCII" else ES2KOMI.get(norm_es(w))
-                if not code:
-                    code = enc_oov_minimax(w) if tgt=="Minimax-ASCII" else enc_oov_komin(w)
-                out.append(code)
+                out.append(code or (enc_oov_minimax(w) if tgt=="Minimax-ASCII" else enc_oov_komin(w)))
             else:
                 out.append(w)
         return " ".join(out)
-
     return "[No soportado]"
 
 def translate_natural(text: str, src_lang: str, tgt_lang: str) -> str:
@@ -719,91 +630,306 @@ def round_trip(text, src, tgt, mode, max_comp_exact):
     back    = universal_translate(conlang, tgt, src, True, False, mode, max_comp_exact)
     return conlang, back
 
-# ------------ UI y explicaciones ------------
+# =====================================================================================
+# ========================== UI bilingüe con selector global ==========================
+# =====================================================================================
+
+ALL_LANGS = ["Español","English","Minimax-ASCII","Kōmín-CJK"]
+
+# Texto ES
 EXPLAIN_ES = """
-**Modo único:Sin ninguna casilla marcada: 0% de compactación (el conlang es similar al original, sin omisiones ni compresión extra).</br>
-Omitir artículos marcada: Aproximadamente 10-15% de compactación (elimina artículos como "el", "la", que representan alrededor del 10% de las palabras en textos españoles típicos).
-</br>Cópula cero marcada: Aproximadamente 5-10% de compactación (omita verbos copulativos como "ser" o "estar" en oraciones afirmativas presentes, común en ~5-10% de las oraciones).
-Ambas (Omitir artículos + Cópula cero): Aproximadamente 15-20% de compactación (combinación aditiva de omisiones).
-</br>Max Compresión Exacta marcada: Aproximadamente 40-60% de compactación en textos medianos/largos (>100 caracteres, gracias a zlib que comprime texto español en ratios de 2:1 a 5:1). Para textos cortos (<30 caracteres), ~0% o incluso aumento por overhead de compresión. Esto asegura traducción inversa 100% exacta.: Minimax (VSO, ·TAMpersonNQ), Kōmín (SOV, ᵖ/ᵒ Ⓟ[2s]̆？).
+## 🌐 ¿Qué hace esta app?
+Traduce entre **Español / Inglés** y dos lenguajes construidos:
+- **Minimax-ASCII** (compacto y solo ASCII)
+- **Kōmín-CJK** (estilo CJK con partículas)
+
+También **comprime sin perder información** si activas **Máx. Compresión Exacta** (`~...` guarda el original).  
+Al **decodificar**, si existe ese `~...`, recuperas el texto **exacto**.
+
+### 🧠 ¿Por qué me sirve?
+- Para **reducir** tamaño de mensajes/notas.
+- Para **codificar/decodificar** de forma legible y reversible.
+- Para jugar con **conlangs** simples.
+
+### ⚙️ Opciones (puedes ignorarlas al principio)
+- **Omitir artículos**: quita *el/la/los/las* o *a/an/the*. Ahorra ~10–15%.
+- **Cópula cero** (presente afirmativo): oculta *ser/estar/be* cuando suena natural. +~5–10%.
+- **Máx. Compresión Exacta**: añade `~...` con el original comprimido (mejor en textos medianos/largos).
 """
+FAQ_ES = """
+### ❓ Preguntas rápidas
+- **¿Se pierde info?** No, con **Máx. Compresión Exacta** el `~...` guarda el original.
+- **¿Sin spaCy?** Funciona igual (modo léxico). Con spaCy suena más natural.
+- **Privacidad**: todo corre dentro de este Space.
+"""
+TUTORIAL_ES = """
+### 🏁 Empezar (3 pasos)
+1. Elige **Fuente** y **Destino**.
+2. Escribe tu frase.
+3. Pulsa **Traducir**.
 
-ALL_LANGS = ["Español","English","Minimax-ASCII","Kōmín-CJK"]
+> Para recuperar **exactamente** el original más tarde, activa **Máx. Compresión Exacta**.
+"""
 
-with gr.Blocks(title="Universal Conlang Translator") as demo:
-    gr.Markdown("# Universal Conlang Translator · Max Compresión Exacta")
-    gr.Markdown(EXPLAIN_ES)
+# Texto EN
+EXPLAIN_EN = """
+## 🌐 What does this app do?
+It translates between **Spanish / English** and two constructed languages:
+- **Minimax-ASCII** (compact, ASCII-only)
+- **Kōmín-CJK** (CJK-style with particles)
 
-    # --- Traducir (universal) ---
-    with gr.Tab("Traducir"):
-        with gr.Row():
-            uni_src = gr.Dropdown(ALL_LANGS, value="Español", label="Fuente")
-            uni_tgt = gr.Dropdown(ALL_LANGS, value="Minimax-ASCII", label="Destino")
-        uni_text = gr.Textbox(lines=3, label="Texto", value="")
-        with gr.Row():
-            uni_drop = gr.Checkbox(value=True, label="Omitir artículos (ES/EN→conlang)")
-            uni_zero = gr.Checkbox(value=False, label="Cópula cero (presente afirm.) (ES/EN→conlang)")
-            uni_max_comp = gr.Checkbox(value=False, label="Max Compresión Exacta (sidecar oculto)")
-        uni_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
-        uni_out = gr.Textbox(lines=6, label="Traducción")
-        gr.Button("Traducir").click(
-            universal_translate,
-            [uni_text, uni_src, uni_tgt, uni_drop, uni_zero, uni_mode, uni_max_comp],
-            [uni_out]
-        )
-
-    # --- Construir frase (ES/EN → Conlang) ---
-    with gr.Tab("Construir frase (ES/EN → Conlang)"):
-        with gr.Row():
-            src_lang = gr.Dropdown(["Español","English"], value="Español", label="Fuente")
-            target = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
-        text_in = gr.Textbox(lines=3, label="Frase", value="")
-        with gr.Row():
-            drop_articles = gr.Checkbox(value=True, label="Omitir artículos")
-            zero_copula   = gr.Checkbox(value=False, label="Cópula cero (presente afirm.)")
-            max_comp_build = gr.Checkbox(value=False, label="Max Compresión Exacta (sidecar oculto)")
-        mode_build = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
-        out = gr.Textbox(lines=6, label="Salida")
-        gr.Button("Construir").click(
-            build_sentence,
-            [text_in, src_lang, target, drop_articles, zero_copula, mode_build, max_comp_build],
-            [out]
-        )
-
-    # --- Decodificar (Conlang → ES/EN) ---
-    with gr.Tab("Decodificar (Conlang → ES/EN)"):
+You can also **compress without losing information** by enabling **Max Exact Compression** (`~...` stores the original).  
+When **decoding**, if `~...` exists, you get the **exact original** back.
+"""
+FAQ_EN = """
+### ❓ Quick answers
+- **Any loss?** Not with **Max Exact Compression** — the `~...` keeps the original.
+- **No spaCy?** Still works (lexical mode). With spaCy it reads more naturally.
+- **Privacy**: everything runs inside this Space.
+"""
+TUTORIAL_EN = """
+### 🏁 Quick start (3 steps)
+1. Pick **Source** and **Target**.
+2. Type your sentence.
+3. Click **Translate**.
+
+> To recover the **exact** original later, enable **Max Exact Compression**.
+"""
+
+def make_group_es():
+    with gr.Group(visible=True) as group:
+        gr.Markdown("# 🌐 Universal Conlang Translator · Compresión Exacta (ES)")
         with gr.Row():
-            src_code = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Fuente")
-            tgt_lang = gr.Dropdown(["Español","English"], value="Español", label="Destino")
-        code_in = gr.Textbox(lines=3, label="Texto en conlang (incluye ~blob si procede)")
-        out3 = gr.Textbox(lines=6, label="Salida")
-
-        def decode_lossless_aware(text, src, tgt):
-            orig = extract_custom_sidecar(text)
-            if orig is not None: return orig
-            orig = extract_sidecar_b85(text)
-            if orig is not None: return orig
-            return decode_simple(strip_custom_sidecar(strip_sidecar_b85(text)), src, tgt)
-
-        gr.Button("Decodificar").click(
-            decode_lossless_aware, [code_in, src_code, tgt_lang], [out3]
-        )
-
-    # --- Round-trip ---
-    with gr.Tab("Prueba ida→vuelta"):
+            with gr.Column(scale=1):
+                with gr.Accordion("Resumen (ES)", open=True):
+                    gr.Markdown(EXPLAIN_ES)
+                with gr.Accordion("FAQ (ES)", open=False):
+                    gr.Markdown(FAQ_ES)
+            with gr.Column(scale=1):
+                with gr.Accordion("Tutorial (ES)", open=True):
+                    gr.Markdown(TUTORIAL_ES)
+                gr.Markdown("**Consejo:** Los mensajes muy cortos pueden no reducirse por la cabecera del `~...`.")
+        with gr.Tab("🔁 Traducir"):
+            with gr.Row():
+                uni_src = gr.Dropdown(ALL_LANGS, value="Español", label="Fuente")
+                uni_tgt = gr.Dropdown(ALL_LANGS, value="Minimax-ASCII", label="Destino")
+            uni_text = gr.Textbox(lines=3, label="Texto", placeholder="Ej.: Hola, ¿cómo estás?", show_copy_button=True)
+            with gr.Row():
+                uni_drop = gr.Checkbox(value=True, label="Omitir artículos (ES/EN → conlang)")
+                uni_zero = gr.Checkbox(value=False, label="Cópula cero (presente afirm.)")
+                uni_max_comp = gr.Checkbox(value=False, label="Máx. Compresión Exacta (sidecar `~...`)")
+            uni_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            with gr.Row():
+                btn_translate = gr.Button("🚀 Traducir", variant="primary")
+                btn_reset = gr.Button("🧹 Limpiar")
+            uni_out = gr.Textbox(lines=6, label="Traducción", show_copy_button=True)
+
+            btn_translate.click(
+                universal_translate,
+                [uni_text, uni_src, uni_tgt, uni_drop, uni_zero, uni_mode, uni_max_comp],
+                [uni_out]
+            )
+            btn_reset.click(lambda: "", None, [uni_text, uni_out])
+
+            gr.Markdown("### 🔎 Ejemplos (clic para autocompletar)")
+            ex1 = gr.Button("ES→Minimax: «Hola, ¿cómo estás?»")
+            ex2 = gr.Button("EN→Kōmín: «This system keeps messages compact.»")
+            ex3 = gr.Button("ES→Minimax (con compresión): «El clima hoy es excelente para pasear.»")
+            ex4 = gr.Button("EN→Kōmín (con compresión): «Please decode this later with the sidecar.»")
+
+            ex1.click(lambda: ("Hola, ¿cómo estás?", "Español", "Minimax-ASCII"), None, [uni_text, uni_src, uni_tgt])
+            ex2.click(lambda: ("This system keeps messages compact.", "English", "Kōmín-CJK"), None, [uni_text, uni_src, uni_tgt])
+            ex3.click(lambda: ("El clima hoy es excelente para pasear.", "Español", "Minimax-ASCII"), None, [uni_text, uni_src, uni_tgt])
+            ex4.click(lambda: ("Please decode this later with the sidecar.", "English", "Kōmín-CJK"), None, [uni_text, uni_src, uni_tgt])
+
+        with gr.Tab("🛠️ Construir (ES/EN → Conlang)"):
+            with gr.Row():
+                src_lang = gr.Dropdown(["Español","English"], value="Español", label="Fuente")
+                target = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
+            text_in = gr.Textbox(lines=3, label="Frase", show_copy_button=True)
+            with gr.Row():
+                drop_articles = gr.Checkbox(value=True, label="Omitir artículos")
+                zero_copula   = gr.Checkbox(value=False, label="Cópula cero (presente afirm.)")
+                max_comp_build = gr.Checkbox(value=False, label="Máx. Compresión Exacta")
+            mode_build = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            with gr.Row():
+                btn_build = gr.Button("🏗️ Construir", variant="primary")
+                btn_build_clear = gr.Button("🧹 Limpiar")
+            out = gr.Textbox(lines=6, label="Salida", show_copy_button=True)
+
+            btn_build.click(
+                build_sentence,
+                [text_in, src_lang, target, drop_articles, zero_copula, mode_build, max_comp_build],
+                [out]
+            )
+            btn_build_clear.click(lambda: "", None, [text_in, out])
+
+        with gr.Tab("🗝️ Decodificar (Conlang → ES/EN)"):
+            with gr.Row():
+                src_code = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Fuente")
+                tgt_lang = gr.Dropdown(["Español","English"], value="Español", label="Destino")
+            code_in = gr.Textbox(lines=3, label="Texto en conlang (puede incluir `~...`)", show_copy_button=True)
+            out3 = gr.Textbox(lines=6, label="Salida", show_copy_button=True)
+
+            def decode_lossless_aware(text, src, tgt):
+                orig = extract_custom_sidecar(text)
+                if orig is not None: return orig
+                orig = extract_sidecar_b85(text)
+                if orig is not None: return orig
+                return decode_simple(strip_custom_sidecar(strip_sidecar_b85(text)), src, tgt)
+
+            with gr.Row():
+                btn_decode = gr.Button("🔓 Decodificar", variant="primary")
+                btn_decode_clear = gr.Button("🧹 Limpiar")
+
+            btn_decode.click(decode_lossless_aware, [code_in, src_code, tgt_lang], [out3])
+            btn_decode_clear.click(lambda: "", None, [code_in, out3])
+
+            gr.Markdown("> **Tip:** si ves `~...`, la decodificación será 100% exacta.")
+
+        with gr.Tab("🔄 Prueba ida→vuelta"):
+            with gr.Row():
+                rt_src = gr.Dropdown(["Español","English"], value="Español", label="Fuente")
+                rt_tgt = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
+            rt_text = gr.Textbox(lines=3, label="Frase", show_copy_button=True)
+            rt_max_comp = gr.Checkbox(value=False, label="Máx. Compresión Exacta")
+            rt_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            rt_out_conlang = gr.Textbox(lines=3, label="Conlang (ida)", show_copy_button=True)
+            rt_out_back    = gr.Textbox(lines=3, label="Vuelta", show_copy_button=True)
+            with gr.Row():
+                btn_rt = gr.Button("▶️ Probar", variant="primary")
+                btn_rt_clear = gr.Button("🧹 Limpiar")
+
+            btn_rt.click(round_trip, [rt_text, rt_src, rt_tgt, rt_mode, rt_max_comp], [rt_out_conlang, rt_out_back])
+            btn_rt_clear.click(lambda: "", None, [rt_text, rt_out_conlang, rt_out_back])
+
+        gr.Markdown("---")
+        gr.Markdown("Hecho con ❤️ · **spaCy** (opcional) · Todo se ejecuta en este Space.")
+    return group
+
+def make_group_en():
+    with gr.Group(visible=False) as group:
+        gr.Markdown("# 🌐 Universal Conlang Translator · Max Exact Compression (EN)")
         with gr.Row():
-            rt_src = gr.Dropdown(["Español","English"], value="Español", label="Fuente")
-            rt_tgt = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
-        rt_text = gr.Textbox(lines=3, label="Frase", value="")
-        rt_max_comp = gr.Checkbox(value=False, label="Max Compresión Exacta")
-        rt_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
-        rt_out_conlang = gr.Textbox(lines=3, label="Conlang (ida)")
-        rt_out_back    = gr.Textbox(lines=3, label="Vuelta")
-        gr.Button("Probar").click(
-            round_trip,
-            [rt_text, rt_src, rt_tgt, rt_mode, rt_max_comp],
-            [rt_out_conlang, rt_out_back]
-        )
+            with gr.Column(scale=1):
+                with gr.Accordion("Summary (EN)", open=True):
+                    gr.Markdown(EXPLAIN_EN)
+                with gr.Accordion("FAQ (EN)", open=False):
+                    gr.Markdown(FAQ_EN)
+            with gr.Column(scale=1):
+                with gr.Accordion("Tutorial (EN)", open=True):
+                    gr.Markdown(TUTORIAL_EN)
+                gr.Markdown("**Tip:** Very short messages may not shrink due to the `~...` header.")
+        with gr.Tab("🔁 Translate"):
+            with gr.Row():
+                uni_src = gr.Dropdown(ALL_LANGS, value="English", label="Source")
+                uni_tgt = gr.Dropdown(ALL_LANGS, value="Minimax-ASCII", label="Target")
+            uni_text = gr.Textbox(lines=3, label="Text", placeholder="e.g., Hello, how are you?", show_copy_button=True)
+            with gr.Row():
+                uni_drop = gr.Checkbox(value=True, label="Drop articles (ES/EN → conlang)")
+                uni_zero = gr.Checkbox(value=False, label="Zero copula (present affirmative)")
+                uni_max_comp = gr.Checkbox(value=False, label="Max Exact Compression (sidecar `~...`)")
+            uni_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            with gr.Row():
+                btn_translate = gr.Button("🚀 Translate", variant="primary")
+                btn_reset = gr.Button("🧹 Clear")
+            uni_out = gr.Textbox(lines=6, label="Translation", show_copy_button=True)
+
+            btn_translate.click(
+                universal_translate,
+                [uni_text, uni_src, uni_tgt, uni_drop, uni_zero, uni_mode, uni_max_comp],
+                [uni_out]
+            )
+            btn_reset.click(lambda: "", None, [uni_text, uni_out])
+
+            gr.Markdown("### 🔎 Examples (click to autofill)")
+            ex1 = gr.Button("EN→Minimax: “Hello, how are you?”")
+            ex2 = gr.Button("ES→Kōmín: “Este sistema mantiene los mensajes compactos.”")
+            ex3 = gr.Button("EN→Minimax (compressed): “The weather today is perfect for a walk.”")
+            ex4 = gr.Button("ES→Kōmín (compressed): “Por favor decodifica esto luego con el sidecar.”")
+
+            ex1.click(lambda: ("Hello, how are you?", "English", "Minimax-ASCII"), None, [uni_text, uni_src, uni_tgt])
+            ex2.click(lambda: ("Este sistema mantiene los mensajes compactos.", "Español", "Kōmín-CJK"), None, [uni_text, uni_src, uni_tgt])
+            ex3.click(lambda: ("The weather today is perfect for a walk.", "English", "Minimax-ASCII"), None, [uni_text, uni_src, uni_tgt])
+            ex4.click(lambda: ("Por favor decodifica esto luego con el sidecar.", "Español", "Kōmín-CJK"), None, [uni_text, uni_src, uni_tgt])
+
+        with gr.Tab("🛠️ Build (ES/EN → Conlang)"):
+            with gr.Row():
+                src_lang = gr.Dropdown(["Español","English"], value="English", label="Source")
+                target = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
+            text_in = gr.Textbox(lines=3, label="Sentence", show_copy_button=True)
+            with gr.Row():
+                drop_articles = gr.Checkbox(value=True, label="Drop articles")
+                zero_copula   = gr.Checkbox(value=False, label="Zero copula (present affirmative)")
+                max_comp_build = gr.Checkbox(value=False, label="Max Exact Compression")
+            mode_build = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            with gr.Row():
+                btn_build = gr.Button("🏗️ Build", variant="primary")
+                btn_build_clear = gr.Button("🧹 Clear")
+            out = gr.Textbox(lines=6, label="Output", show_copy_button=True)
+
+            btn_build.click(
+                build_sentence,
+                [text_in, src_lang, target, drop_articles, zero_copula, mode_build, max_comp_build],
+                [out]
+            )
+            btn_build_clear.click(lambda: "", None, [text_in, out])
+
+        with gr.Tab("����️ Decode (Conlang → ES/EN)"):
+            with gr.Row():
+                src_code = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Source")
+                tgt_lang = gr.Dropdown(["Español","English"], value="English", label="Target")
+            code_in = gr.Textbox(lines=3, label="Conlang text (may include `~...`)", show_copy_button=True)
+            out3 = gr.Textbox(lines=6, label="Output", show_copy_button=True)
+
+            def decode_lossless_aware(text, src, tgt):
+                orig = extract_custom_sidecar(text)
+                if orig is not None: return orig
+                orig = extract_sidecar_b85(text)
+                if orig is not None: return orig
+                return decode_simple(strip_custom_sidecar(strip_sidecar_b85(text)), src, tgt)
+
+            with gr.Row():
+                btn_decode = gr.Button("🔓 Decode", variant="primary")
+                btn_decode_clear = gr.Button("🧹 Clear")
+
+            btn_decode.click(decode_lossless_aware, [code_in, src_code, tgt_lang], [out3])
+            btn_decode_clear.click(lambda: "", None, [code_in, out3])
+
+            gr.Markdown("> **Tip:** if you see `~...`, decoding will be bit-perfect.")
+
+        with gr.Tab("🔄 Round-trip"):
+            with gr.Row():
+                rt_src = gr.Dropdown(["Español","English"], value="English", label="Source")
+                rt_tgt = gr.Dropdown(["Minimax-ASCII","Kōmín-CJK"], value="Minimax-ASCII", label="Conlang")
+            rt_text = gr.Textbox(lines=3, label="Sentence", show_copy_button=True)
+            rt_max_comp = gr.Checkbox(value=False, label="Max Exact Compression")
+            rt_mode = gr.Dropdown(["Semi-lossless"], value="Semi-lossless", visible=False)
+            rt_out_conlang = gr.Textbox(lines=3, label="Outward (conlang)", show_copy_button=True)
+            rt_out_back    = gr.Textbox(lines=3, label="Back", show_copy_button=True)
+            with gr.Row():
+                btn_rt = gr.Button("▶️ Test", variant="primary")
+                btn_rt_clear = gr.Button("🧹 Clear")
+
+            btn_rt.click(round_trip, [rt_text, rt_src, rt_tgt, rt_mode, rt_max_comp], [rt_out_conlang, rt_out_back])
+            btn_rt_clear.click(lambda: "", None, [rt_text, rt_out_conlang, rt_out_back])
+
+        gr.Markdown("---")
+        gr.Markdown("Made with ❤️ · **spaCy** (optional) · Everything runs inside this Space.")
+    return group
+
+with gr.Blocks(title="Universal Conlang Translator", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("## 🌍 Idioma / Language")
+    lang_select = gr.Radio(choices=["ES","EN"], value="ES", label="Selecciona / Select")
+    group_es = make_group_es()
+    group_en = make_group_en()
+
+    def switch_lang(code):
+        if code == "EN":
+            return gr.update(visible=False), gr.update(visible=True)
+        return gr.update(visible=True), gr.update(visible=False)
+
+    lang_select.change(switch_lang, [lang_select], [group_es, group_en])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()