A Guide to the Hakka Language Family Across Dialects, Tones and Regions

Hakka is a dialect cluster, not a language with a few regional accents, and the differences concentrate in exactly the places that break conversation: tone count, sandhi rules, and entering-tone retention. A phrasebook bought for a Meizhou trip fails in a Miaoli village. A Taiwan Hakka TV broadcast sounds alien to a Hailu-speaking grandmother. Understanding this before you commit to a course or a textbook saves months of wasted study.

This map covers the entire language section. It does not teach you to speak Hakka. It tells you which variety you are dealing with, what makes it distinct from the next one, and where to go next depending on your goal. Comparing Sixian and Hailu? Head to the dialect comparison guide. Trying to make sense of those abrupt, clipped syllables your teacher calls entering tones? A separate guide pulls apart -p, -t, and -k finals. Deciding whether to sit the Hakka Affairs Council certification in Taiwan? That has its own page, linked by the question it answers. Every guide in this section names the variety under discussion and uses linguistic sources. None of them presents a single pronunciation as universally Hakka.

Two varieties dominate Taiwan Hakka media and education: Sixian, with six tones, and Hailu, with seven. The Meizhou (Meixian) variety is the prestige baseline in mainland China but is not the most widely spoken variant globally. Confuse those three and you will waste months of study on the wrong tone inventory.

• Prestige baseline (mainland): Meizhou Meixian dialect, six citation tones, retains -p -t -k entering tones
• Dominant Taiwan variety: Sixian dialect (six tones), default for Hakka TV and school textbooks
• Second Taiwan variety: Hailu dialect (seven tones), stronger literary-layer retention, distinct sandhi
• Tone inventory range: Six to seven citation tones across major varieties; mutual intelligibility drops sharply when tone count differs by two or more
• Entering-tone retention: All major varieties retain -p, -t, -k finals; a speaker who has lost them cannot reproduce classical poetry or surname pronunciations the same way
• Language vitality (Taiwan): EGIDS score masks a generational cliff: self-identified speakers far outnumber fluent speakers under 40

Every guide here names the variety. Every guide cites linguistic evidence. When you see a tone chart, it is labelled Sixian, Hailu, Dabu, Raoping, or Meixian, not ‘Hakka’. Learn to tell them apart before you commit to one.

Sixian: The Taiwan Default

Sixian is the variety you hear on Hakka TV standardised broadcasts and in most Taiwan public-sector language resources. It originated with migrants from Jiaying prefecture (now Meizhou) and carries six citation tones. Its sandhi rule count is low compared to Hailu. That makes it the pragmatic choice for a first variety. The Hakka Affairs Council testing offers Sixian as one of its core examination varieties. Do not assume it is identical to modern Meixian. A century of separation and Mandarin contact has shifted its lexicon. A learner notices the gap within the first month.

Hailu: Seven Tones and a Literary Spine

Hailu comes from the Haifeng-Lufeng region of Guangdong and is the second major variety in Taiwan, concentrated in Hsinchu and pockets of Miaoli. Its tone inventory is seven, not six, and the extra tone is not cosmetic. It changes the sandhi environments. A Sixian speaker cannot follow a Hailu conversation without deliberate exposure. Hailu preserves a stronger literary layer, which surfaces in formal registers and ancestral rites. If your goal is to reconnect with a Hailu-speaking lineage, do not study Sixian first. The mutual intelligibility gap is real and it frustrates diaspora learners every year.

Meizhou Meixian: The Prestige Baseline That Is Not the Standard

Meixian is the reference variety for standardisation in mainland China, but it is not the one spoken by the largest number of Hakka people worldwide. Its phonology is well documented and it retains the full set of entering tones (-p, -t, -k). Do not treat Meixian as the ‘original’ from which others devolved. Urban Meixian speech has absorbed Cantonese loanwords and Mandarin neologisms. Treat it as a research baseline.

Dabu and Raoping: The Comparison Varieties That Rewrite Sandhi

Dabu phonology is distinct enough to matter. Originating in the Meizhou region, Dabu has its own migration stream to Taiwan, particularly the Taichung area, and its tone sandhi patterns do not map neatly onto either Sixian or Hailu. Raoping, from northeastern Guangdong, adds another layer of complexity. Its tone sandhi environments are among the most extensive in the family, and Raoping speakers in Hsinchu and Miaoli pockets can be difficult for other Hakka speakers to follow. A comparison that omits Dabu and Raoping is not a comparison. It is a tour of the two largest car parks and a claim to have mapped the city.

The decision path for a new learner runs through tone count, sandhi complexity, and entering-tone realisation. Start with the number of citation tones. A Sixian six-tone inventory is the baseline for Taiwan media. Hailu seven tones adds a contour that alters sandhi rule count and sandhi environments across whole phrases. Raoping tone sandhi is still more elaborate, with more environments triggering a change than in either Sixian or Hailu. Compare Hakka to Hokkien: Hokkien has seven tones and a near-universal sandhi rule. Hakka sandhi varies dramatically by variety, from almost none in some Meixian analyses to pervasive in Raoping. The practical consequence? Mutual intelligibility between two varieties is not predicted by shared vocabulary alone. Tone count and sandhi rule count are better predictors. Two varieties that differ by two or more citation tones will feel like separate languages to someone who has not trained on the other.

Entering Tones Are the Other Axis

All major varieties retain -p, -t, -k entering tones, but the exact phonetic realisation differs. A Sixian -p final is not identical to a Hailu -p final in every phonetic environment. Raoping entering tones have undergone shifts that affect syllable length. These entering tones are a conservative feature that Mandarin has lost entirely. That is why a Hakka speaker can reconstruct classical poetry rhyme schemes that a Mandarin monolingual cannot. When you compare varieties, ask not just ‘does it have entering tones’ but ‘how many entering-tone allotones does it distinguish’. A variety with two entering-tone contours is not the same as one with one, even if both retain the -p -t -k finals.

Start with the citation tone inventory: the tones a syllable carries in isolation. That number is what you see on a chart. It is not what you hear in speech. Sandhi, the tone change that occurs when syllables combine, can reduce or multiply the functional tone count. Sixian has a modest sandhi rule count. Hailu has more. Raoping has enough that a citation tone chart is almost a theoretical abstraction. When a teacher says ‘Hakka has six tones’, they are describing Sixian. When they say ‘seven’, they mean Hailu. Neither statement is wrong. Both are incomplete without naming the variety.

The tone sandhi environments are the specific contexts where a tone changes. These can be triggered by following tone, syntactic boundary, or speech tempo. A variety with extensive sandhi sounds more alien to a speaker of a sandhi-poor variety than the tone charts suggest. That is why a Sixian speaker can read a Hailu tone chart and still fail to understand spoken Hailu. The sandhi rules are not decorative. They are the grid the language runs on.

An entering tone is a syllable that ends in an unreleased stop consonant: -p, -t, or -k. These are not separate consonants you can hear in isolation. They are closures that cut the syllable short. All major varieties retain them, but the number of entering-tone allotones varies. Sixian distinguishes two entering-tone contours, one high and one low, depending on the historical tone category. Hailu may merge or split these differently. A learner who masters entering tones in Sixian will still need to re-tune for Hailu, because the pitch height and contour on the same -p -t -k syllable can differ.

Entering-tone retention is what lets a Hakka speaker recite Tang poetry with end-rhymes that a Mandarin speaker cannot reproduce. It is also what makes Hakka surname pronunciations distinctive. A surname that ends in -p in Hakka may end in a vowel or a different consonant in Cantonese or Mandarin. The entering tone is part of the name’s identity. If you are reconnecting with a lineage, get the entering tones right. Ancestral halls are ritual spaces where mispronouncing a name is not a minor error.

Taiwan offers the most developed environment for learning Hakka, built on a policy architecture that did not exist two decades ago. The Hakka Basic Act designates Hakka as a national language. The Hakka Affairs Council runs a standardised testing programme across multiple varieties, and the results are used for media hiring, civil-service postings, and school placement. The two primary test varieties are Sixian and Hailu, with Dabu and Raoping offered at some levels. Passing the certification at a high level is a genuine professional credential in Taiwan’s Hakka cultural regions.

Certification and Materials

The Hakka Affairs Council publishes its own preparation materials. Use them. They are the most reliable baseline for aligning with Taiwan standards. Do not mix mainland Meixian textbooks with Taiwan Sixian audio and expect them to match. They will not. The lexical divergence is real, particularly in modern vocabulary. Tone sandhi descriptions in mainland sources sometimes describe a more conservative or more Mandarin-influenced system than the one spoken in Taiwan.

Immersion and Romanisation

Hakka TV standardised broadcast uses Sixian as its primary variety, with Hailu programming in specific slots. This gives you a daily immersion source subtitled in Mandarin and sometimes in romanisation. The romanisation system is the Taiwan Hakka Romanization System, not the mainland scheme. Learn the one that matches your certification goal.

The Generational Cliff

The EGIDS vitality index for Hakka in Taiwan is deceptive. Official surveys report around 2.5 million ethnic Hakka, but intergenerational transmission data shows a steep drop after the under-40 cohort. A village designated as a Hakka cultural area under Taiwan law may have policy designation, signage, and funding without having fluent speakers in shops. Do not confuse the administrative map with the linguistic reality. The Hakka Affairs Council testing data and the council’s own language survey reports are the best public indicators of where transmission is holding.

A growing gap separates Hakka TV content from Hakka social media code-mixing. Broadcast Hakka is standardised, scripted, and subtitled. Social media Hakka is code-mixed with Mandarin, written in romanisation, ad-hoc characters, or a mixture of both, and uses forms that do not appear on television. A learner who can follow a Hakka TV news segment may be lost in a Hakka-language Instagram comment thread within seconds. This is not a failure of the learning method. It is two different registers, and both are real. If your goal is to participate in living Hakka-language communities online, supplement formal study with exposure to unscripted, mixed-code Hakka as early as possible. The broadcast standard gives you the structure. Social media gives you the language as it is used under forty.