Hebrew Nikud

Hebrew Nikud (ניקוד) Reference

A reference guide for adding selective nikud to Hebrew text. Designed for AI agents that need accurate pronunciation hints (e.g., for TTS).

Golden Rule

Only add nikud when you're 100% certain it's correct. Wrong nikud is worse than no nikud — the TTS model will read your mistake literally instead of guessing correctly from context.

When to Add Nikud

1. Ambiguous consonants (dagesh in בכ"פ)
2. Gender-specific suffixes
3. Homographs (same spelling, different pronunciation)
4. Foreign names and loanwords
5. Stress placement that changes meaning

When in doubt — don't nikud. Let the TTS model guess from context.

1. Vowel Symbols Reference

Shva Rules (שְׁוָא)

• Start of word → vocal (na): בְּרֵאשִׁית (bereshit)
• End of word → silent (nach): כָּתַבְתְּ (katavt)
• Two consecutive → first silent, second vocal: יִשְׁמְרוּ (yishmeru)
• After long vowel → vocal: כּוֹתְבִים (kotvim)
• After short vowel → silent: מַלְכָּה (malka)

2. Dagesh (דגש) — Hard vs Soft Consonants

Begedkefet (בגדכפ"ת)

Six letters historically changed sound with dagesh. In modern Hebrew, only three still have audible differences:

For TTS purposes, only בכ"פ matter (B/V, K/Kh, P/F).

When does dagesh appear?

Dagesh Lene (light) — hardening, in begedkefet letters:

• At the start of a word (after pause): בַּיִת (bayit)
• After a silent shva: מִסְפָּר (mispar - the פ has dagesh)

Dagesh Forte (strong) — doubling, in any letter except gutturals (אהחע"ר):

• After the definite article הַ: הַבַּיִת (habayit)
• In Pi'el/Pu'al/Hitpa'el verb patterns: סִפֵּר, דִּבֵּר
• After prepositions with article: בַּבַּיִת (babayit)

Common dagesh examples for TTS

Pe/Fe (פּ/פ) — most error-prone:

• פִּיצָה (pizza), פִּייר (Pierre), פַּעַם (pa'am)
• פּוֹלִיטִיקָה (politika), פָּרִיז (Paris)
• אוֹפֶּרָה (opera), קָפּוּצִ'ינוֹ (cappuccino)

Bet/Vet (בּ/ב):

• בְּסֵדֶר (b'seder), בְּדִיוּק (bediyuk), בְּרָכָה (brakha)
• בּוֹסְטוֹן (Boston), בֵּירָה (bira - beer)

Kaf/Khaf (כּ/כ):

• כּוֹס (kos), כַּמָּה (kama), כּוֹכָב (kokhav)
• כְּרִיסְטִינָה (Christina)

3. Verb Conjugations (בניינים)

Hebrew has 7 verb patterns. This is the hardest part — if unsure of the binyan, don't nikud the verb.

פָּעַל (Pa'al / Qal) — Basic active

פִּעֵל (Pi'el) — Intensive active

הִפְעִיל (Hif'il) — Causative active

הִתְפַּעֵל (Hitpa'el) — Reflexive

נִפְעַל (Nif'al) — Passive of Pa'al

פֻּעַל (Pu'al) — Passive of Pi'el

הֻפְעַל (Huf'al) — Passive of Hif'il

⚠️ Common Verb Confusions

Rule of thumb:

• Simple action → Pa'al (כָּתַב wrote, שָׁמַר guarded)
• Intensive / caused action → Pi'el (סִפֵּר told, דִּבֵּר spoke, לִמֵּד taught)
• Made someone do → Hif'il (הִסְבִּיר explained, הִזְמִין invited)
• Was done to → Nif'al/Pu'al/Huf'al (נִכְתַּב was written)

4. Gender Suffixes

Examples

מה שלומְךָ? (to male)
מה שלומֵךְ? (to female)
יש לְךָ זמן? (to male)
יש לָךְ זמן? (to female)
אני אוהב אוֹתְךָ (male object)
אני אוהב אוֹתָךְ (female object)

5. Common Homographs

Words spelled the same but pronounced differently:

6. Foreign Names & Loanwords

The model often mispronounces foreign words. Add dagesh for P/B/K sounds:

7. Preposition Nikud Rules

Prepositions בְּ (be-), כְּ (ke-), לְ (le-) change nikud in certain situations:

8. Quick Decision Tree

Should I add nikud to this word?
│
├─ Is it a common word with obvious pronunciation?
│  └─ YES → Don't nikud (מה, יש, אני, הוא, שלום, טוב, etc.)
│
├─ Is it a בכ"פ letter that could go either way?
│  └─ YES → Add dagesh if it's P/B/K (not F/V/Kh)
│
├─ Is it a gender suffix (ך, את)?
│  └─ YES → Add nikud for the correct gender
│
├─ Could it be read as a different word (homograph)?
│  └─ YES → Add nikud to disambiguate
│
├─ Is it a foreign name or loanword?
│  └─ YES → Add dagesh for P/B/K sounds
│
├─ Is it a verb where the binyan matters?
│  │
│  ├─ Am I certain of the binyan?
│  │  └─ YES → Add nikud per the binyan table above
│  │
│  └─ Am I NOT certain?
│     └─ DON'T nikud it! Let TTS guess.
│
├─ Am I 100% sure of the correct nikud?
│  ├─ YES → Add it
│  └─ NO → Don't add it!
│
└─ Default: Don't nikud.

9. Common Mistakes to Avoid

1. Over-nikuding — Adding nikud to every word makes TTS worse, not better
2. Wrong binyan — סָפַר (counted) instead of סִפֵּר (told) is a classic
3. Forgetting dagesh in loanwords — פיצה sounds like "fitza" without dagesh
4. Kamatz vs Kamatz Katan — Both look like ָ but kamatz katan sounds "o" (כָּל = kol)
5. Mixing gender suffixes — ְךָ (male) vs ֵךְ (female) are easy to confuse
6. Nikuding when unsure — If you're not 100% certain, DON'T. The TTS model's contextual guess is usually better than wrong nikud.