🔡X-SAMPA — Wells's ASCII encoding of the full IPA

X-SAMPA is John C. Wells's 1995 proposal for a keyboard-compatible ASCII coding of the entire IPA — in his words, “everything on the 1993 IPA Chart, including diacritics and tone marks” — designed so that IPA-transcribed material could be sent through e-mail and other 7-bit channels without loss.¹ Where SAMPA is a family of per-language tables, X-SAMPA is one universal mapping.¹² That difference is the whole point of the scheme, and it is why X-SAMPA, not SAMPA, became the ASCII-IPA that people actually use for general phonetic work.

The defining document is a roughly 7,000-word paper, the fetched version marked “Revised draft 1995 04 28”, hosted at UCL as a PDF, with an IPA-free HTML summary page that was last revised on 3 May 2000 when Unicode code points were added to the symbol listing.¹³ Note that Wells framed the paper explicitly as a proposal soliciting colleagues' reactions. X-SAMPA's status as a de-facto standard comes from adoption, not from any ratifying body.¹

1. The base layer

IPA symbols that are also lower-case Latin letters a–z keep their IPA values unchanged. Everything else is recoded within ASCII 33–126.¹ Case is significant: upper and lower case are different symbols. And the scheme's defining property, stated in §26 of the paper, is that symbol strings remain uniquely parsable even with no spaces between successive characters.¹

A sample of consonants X-SAMPA codes that the SAMPA basic table does not

2. The two special characters

Two ASCII characters do structural work, and understanding them is most of understanding X-SAMPA.

The backslash — the “universal diacritic”

Section 12 of the paper defines \ as meaning “the preceding character is to be interpreted in a special way”, which roughly doubles the available symbol space by giving two-place codes.¹ The syntax is fixed: the backslash attaches to the single immediately preceding character, and is written after it. Attaching it to more than one character is a syntax error, not a stylistic variant.

The backslash in action: base symbol versus backslash form

The underscore — the diacritic introducer

Section 13 assigns _ (ASCII 95) the role “interpret the following character as a diacritic”.¹ So t_h is aspirated t, t_w is labialized t, and d_n is d with nasal release. Ejectives are _> (alternatively _?) and implosives are _<.¹

Common underscore diacritics

The precedence rule. Where \ and _ co-occur, the backslash takes priority. t_?\ parses as t plus pharyngealization, because ?\ — the pharyngeal-fricative code — binds first.¹ Reversing this is one of the classic X-SAMPA errors.

3. Two SAMPA marks that X-SAMPA redefined

This is where reading old SAMPA files with X-SAMPA eyes goes wrong.¹

• The apostrophe (ASCII 39) becomes the palatalization diacritic. In SAMPA it had denoted rising tone — a use Wells notes was unpopular.
• The grave accent / backtick (ASCII 96) becomes rhoticity and retroflexion: t` is retroflex t, @` is r-coloured schwa. In SAMPA it had denoted falling tone.

Note in passing that the retroflex approximant is a composite of the two mechanisms: r\` is the alveolar approximant r\ plus rhoticity.¹³

4. Prosody

Tone and prosody can be handled two ways.¹ Either as underscore diacritics — _T extra-high through _B extra-low, _R rising, _F falling — or on a separate tier delimited by angle brackets < >, an approach that draws on the companion SAMPROSA proposals.¹⁴

5. The trap: ASCII schemes that look alike and are not

X-SAMPA is not the only ASCII coding of the IPA, and the others are not compatible with it. Kirshenbaum (used by eSpeak) and WorldBet assign different ASCII characters to the same IPA symbols, and — much worse — assign the same ASCII characters to different IPA symbols. The table below is drawn from this project's cross-system divergence dataset and shows only the rows where the schemes actually disagree.⁵

Where the ASCII schemes diverge — the rows to check before trusting a file

The practical consequence: an ASCII phonetic string is not self-identifying. Before parsing one, establish which scheme produced it. A file that renders as sense under X-SAMPA and as nonsense under Kirshenbaum has told you what it is; a file that renders as plausible but different sense under both has not.

6. Errors to avoid

• Mixing X-SAMPA with per-language SAMPA values. SAMPA is a set of language-specific tables; X-SAMPA is one universal mapping. The same ASCII glyph can differ between them.¹²
• Assuming any ASCII-IPA you meet online is X-SAMPA. Kirshenbaum and WorldBet are distinct schemes with conflicting assignments — see the table above.⁵
• Citing “CXS” (Conlang X-SAMPA) as standardized. CXS is community lore. It does not appear anywhere in Wells's defining document, and no authoritative specification exists. Label it explicitly as an unstandardized community variant, or leave it out.¹
• Reversing the _ / \ precedence, or attaching a backslash to more than the single preceding character. Both violate §12–13 syntax.¹
• Assuming eSpeak speaks X-SAMPA. It uses Kirshenbaum. “ASCII IPA tool” does not imply “X-SAMPA”.

7. Sources

• Wells, “Computer-coding the IPA: a proposed extension of SAMPA” (1995) — The defining document. Sections 1–27 are the source for the base layer, the backslash and underscore mechanisms, the precedence rule, the redefined SAMPA marks and the unique-parsability property.
• X-SAMPA — UCL summary page — The IPA-free HTML symbol listing, last revised 3 May 2000 when Unicode code points were added.
• SAMPA — official home page, University College London — The per-language scheme X-SAMPA generalizes.
• SAMPROSA — SAM Prosodic Transcription — The prosodic proposals X-SAMPA's angle-bracket tier draws on.
• Wikipedia — X-SAMPA — Secondary overview with a full symbol table.