How is this pronounced?

Why word-level lookup beats character-level

Around a thousand common Chinese characters have more than one reading, and which one applies depends on the word the character is sitting in. 行 is xíng in 行李 and háng in 银行. 乐 is lè in 快乐 and yuè in 音乐. 长 is cháng when it means long and zhǎng when it means to grow.

A converter that walks through your text one character at a time has to guess, and it usually guesses the most common reading. This one segments the text into words first and takes each word's reading from the dictionary, so the ambiguity never comes up. Every reading is checked to produce exactly one syllable per character, which is what lets the pinyin sit above the right character rather than merely near it.

Tone sandhi, and what is actually applied

Some tones change depending on what follows. Inside a word the dictionary already accounts for it — 一起 is stored as yìqǐ, not yīqǐ. Between words nothing can, because the pair only exists in your text. 不是 is not a dictionary entry, so a per-word lookup gives bù shì where every speaker says bú shì.

So two rules are applied across word boundaries. 不 becomes bú before a fourth tone. 一 becomes yí before a fourth tone and yì before a first, second or third — except in ordinals like 第一, where it keeps its citation tone. Everything that changed is listed under the result, rather than silently corrected.

Third-tone sandhi is off by default

Two third tones in a row are said as second-then-third: 你好 comes out as níhǎo. But it is conventionally not written that way, and a learner reading along with a textbook expects 你好 to say nǐhǎo. It is a toggle rather than a default, and in longer chains of third tones the grouping depends on phrasing in a way no automatic rule settles — so treat that output as a hint, not an authority.

The limits

Words outside the 11,643-entry dictionary fall back to character-by-character readings, which is exactly where the guessing returns — names in particular. Neutral tones follow the dictionary's judgement, which is the standard-Mandarin one; speakers in different regions neutralise differently. And segmentation itself can be wrong on rare compounds, which will show up as a strange-looking reading rather than as an error message.

If you want to see the words behind the readings, the text analyzer shows the same segmentation with meanings attached, and the reader lets you tap any word for its definition and audio.