Overview of the phonics Package

James P. Howard, II

2026-08-11

The phonics package for R is designed to provide a variety of phonetic indexing algorithms in common and not-so-common use today. The algorithms generally reduce a string to a symbolic representation approximating the sound made by pronouncing the string. They can be used to match names, words, and as a proxy for assorted string distance algorithms.

Basic Usage

Every encoder accepts a character vector and returns one code per input. Match Rating Approach (MRA) additionally has a comparison stage. For example, soundex() and refinedSoundex() encode a vector directly.

library("phonics")

x1 <- "Catherine"
x2 <- "Kathryn"
x3 <- "Katrina"
x4 <- "William"

x <- c(x1, x2, x3, x4)

soundex(x1)

## [1] "C365"

soundex(x2)

## [1] "K365"

soundex(x)

## [1] "C365" "K365" "K365" "W450"

refinedSoundex(x1)

## [1] "C30609080"

refinedSoundex(x2)

## [1] "K3060908"

Both functions accept a maxCodeLen that limits the length of the returned code. All encoders except mra_encode() expose this bound. Cologne preserves its historical unbounded output by default (maxCodeLen = NULL); an explicit bound truncates it like the other encoders.

Beyond soundex, additional algorithms are available, as shown in the following table.

Algorithm	Function Name
Caverphone 1 and 2	caverphone()
Cologne Phonetic	cologne()
Lein Name Coding	lein()
Match Rating Approach	mra_encode(), mra_compare()
Metaphone	metaphone()
Original and USDA Modified NYSIIS	nysiis()
Oxford Name Compression Algorithm	onca()
Phonex	phonex()
Roger Root Name Coding Procedure	rogerroot()
Soundex and Apache Refined Soundex	soundex(), refinedSoundex()
Census Modified Statistics Canada	statcan()

Input and output contracts

Input is converted to a common case before encoding. With the default clean = TRUE, an input containing characters outside an encoder’s supported alphabet produces a warning and an NA result. With clean = FALSE, those characters are discarded before encoding. NA values remain NA, empty strings remain empty, and vector order is preserved.

The encoders implement different published normalization domains. In particular, Cologne accepts German umlauts and eszett, Phonex accepts their German equivalents, and Census Modified Statistics Canada normalizes its documented French letters.

Match Rating Approach

Unlike other algorithms described here, MRA is a two-stage algorithm with separate encoding and comparison routines. For instance, the results of Soundex on two different strings can be directly compared to test for equality:

soundex(x1) == soundex(x2)

## [1] FALSE

soundex(x2) == soundex(x3)

## [1] TRUE

However, the MRA encoding algorithm may return different encodings for similar strings that should match. So the second stage, for comparison, is used to compare to MRA-encoded strings. The encoding algorithm is provided by mra_encode and the comparison algorithm is provided by mra_compare.

(mra1 = mra_encode("Katherine"))

## [1] "KTHRN"

(mra2 = mra_encode("Catherine"))

## [1] "CTHRN"

(mra3 = mra_encode("Katarina"))

## [1] "KTRN"

mra_compare(mra1, mra2)

## [1] TRUE

mra_compare(mra1, mra3)

## [1] TRUE

mra_compare(mra2, mra3)

## [1] TRUE

The required similarity threshold gets smaller as the combined encoded length increases. The comparison removes aligned characters from left to right and then right to left, and scores the unmatched characters in the longer remainder.

Summary

This vignette has outlined the phonics package for R. Included in this package are several English-, German-, and French-language suitable algorithms for phonetically reducing names and strings. These can be used for comparison and indexing, as well as later record linkage.