Count Sentences in Text Accurately

Sentence Counting and Boundary Detection

Counting sentences requires identifying where one sentence ends and another begins. While periods seem like obvious markers, real-world text is full of ambiguities.

The Naive Approach

A simple regex splits on sentence-ending punctuation:

function countSentences(text) {
  const matches = text.match(/[.!?]+/g);
  return matches ? matches.length : 0;
}

This works for simple text like "Hello world. How are you?" (2 sentences) but fails on many real-world inputs.

Why Sentence Splitting Is Hard

Consider these problematic cases:

• Abbreviations: "Dr. Smith went to Washington." — 1 sentence, not 2
• Decimal numbers: "The price is $3.50." — 1 sentence, not 2
• Ellipsis: "Wait... what happened?" — 1 sentence, not 4
• URLs: "Visit example.com." — 1 sentence, not 2
• Initials: "J.K. Rowling wrote the series." — 1 sentence, not 3
• Quotations: '"Really?" she asked.' — 1 or 2 depending on interpretation

A Better Algorithm

A more robust approach combines regex with heuristics:

function countSentences(text) {
  if (!text.trim()) return 0;
  // Split on sentence-ending punctuation followed by space + capital
  // or end of string
  const sentences = text
    .replace(/([.!?])\s*(?=[A-Z])/g, "$1|")
    .split("|");
  return sentences.filter(s => s.trim().length > 0).length;
}

This looks for a period, exclamation mark, or question mark followed by whitespace and a capital letter, which is a stronger signal of a sentence boundary.

The Intl.Segmenter Approach

Modern browsers support sentence-level segmentation:

const segmenter = new Intl.Segmenter("en", { granularity: "sentence" });
const sentences = [...segmenter.segment(text)];
const count = sentences.length;

This handles abbreviations, decimal points, and many edge cases correctly because it uses ICU (International Components for Unicode) rules internally.

Multi-Language Considerations

Different languages use different sentence terminators. Thai uses a space character, Japanese uses 。, and Greek uses ; for questions. The Intl.Segmenter API handles these conventions when the correct locale is specified.