Archive
Matching context sensitive rules and generating output using regular expressions
I have previously written about generating words that sound like an input word. My interest in reimplementing this project from many years ago was fueled by a desire to find out exactly how flexible modern regular expression libraries are (the original used a bespoke tool). I had a set of regular expressions describing a mapping from one or more letters to one or more phonemes and I wanted to use someone else’s library to do all the heavy duty matching.
The following lists some of the mapping rules. The letters between [] are the ones that are ‘consumed’ by the match and any letters/characters either side are the context required for the rule to match. The characters between // are phonemes represented using the Arpabet phonetic transcription.
@[ew]=/UW/ [giv]=/G IH V/ [g]i^=/G/ [ge]t=/G EH/ su[gges]=/G JH EH S/ [gg]=/G/ b$[g]=/G/ [g]+=/JH/ # space - start of word # $ - one or more vowels # ! - two or more vowels # + - one front vowel # : - zero or more consonants # ^ - one consonant # ; - one or more consonants # . - one voiced consonant # @ - one special consonant # & - one sibilant # % - one suffix |
After some searching I settled on using the PCRE (Perl Compatible Regular Expressions) library, which contains more functionality than you can shake a stick at.
My plan was to translate each of the 300+ rules, using awk, into a regular expression, concatenate all of these together using alternation and let PCRE handle all of the matching details; which is what I did and it worked. Along the way a few problems had to be solved…
How can the appropriate phoneme(s) be generated when a rule matches? The solution is to use what PCRE calls callouts. During matching if the sequence (?C77) is encountered in the pattern a developer defined function (set up prior to calling pcre_execute) is called with information about the current state of the match. In this example the information would include the value 77 (values between 0 and 255 are supported). By embedding a unique number in the subpattern for each rule (and writing the appropriate phoneme sequence out to a configuration file that is read on program startup) it is possible to generate the appropriate output (because there are more than 255 rules a pair of callouts are needed to specify larger values).
How can the left/right letter context be handled? Most regular expression matching works by consuming all of the matched characters, making them unavailable for matching by other parts of the regular expression during that match. PCRE supports what it calls left and right assertions, which require a pattern to match but don’t consume the matched characters, leaving them to be matched by some other part of the pattern. So the rule [ge]t is mapped to the regular expression ge(?=t) which consumes a ge followed by a t but leaves the t for matching by another part of the pattern.
One problem occurs for backward assertions, which are restricted to matching the same number of characters for all alternatives of the pattern. For example the backward assertion (?<=(a|ab)e) is not supported because one path through the pattern is two characters long while the other is three characters long. The rule @[ew] cannot be matched using a backward assertion because @ includes letter sequences of different length (e.g., N, J, TH). The solution is to use a callout to perform a special left context match (specified by the callout number) which works by reversing the word being matched and the left context pattern and performing a forward (rather than backward) match.
The final pattern is over 10,000 characters long and looks something like (notice that everything is enclosed in () and terminated by a + to force the longest possible match, i.e., the complete word:
(((a(?=$)(?C51)))|((?<=^)(are(?=$)(?C52)))| ... |(z(?C106)(?C55)))+ |
Now we need a method of using the letter to phoneme rules to map phonemes to letters. In some cases a phoneme sequence can be mapped to multiple letter sequences and I wanted to generate all of the possible letter sequences (e.g., cat -> K AE T -> cat, kat, qat). PCRE does support a matching function capable of returning all possible matches. However this function does not support some of the functionality required, so I decided to 'force' the single match function to generate all possible sequences by using a callout to make it unconditionally fail as the last operation of every otherwise successful match, causing the matching process to backtrack and try to find an alternative match. Not the most efficient of solutions but it saved me having to learn a lot more about the functionality supported by PCRE.
For a given sequence of phonemes it is simple enough to match it using a regular expression created from the existing rules. However, any match also needs to meet any left/right letter context requirements. Because we are generating letters left to right we have a left context that can be matched, but no right context.
The left context is matched by applying the technique used for variable length left contexts, described above, i.e., the letters generated so far are reversed and these are matched using a reversed left context pattern.
An efficient solution to matching right context would be very fiddly to implement. I took the simple approach of ignoring the right letter context until the complete phoneme sequence had been matched; the generated letter sequence out of this matching process is feed as input to the letter-to-phoneme function and the returned phoneme sequence compared against the original generating phoneme sequence. If the two phoneme sequences are identical the generated letter sequence is included in the final set of returned letter sequences. Not very computer efficient, but an efficient use of my time.
I could not resist including some optimzations. For instance, if a letter sequence only matches at the start or end of a word then the corresponding phonemes can only match at the start/end of the sequence.
I have skipped some of the minor details, which you can read about in the source of the tool.
I would be interested to hear about the libraries/tools used by readers with experience matching patterns of this complexity.
Generating sounds-like and accented words
I have always been surprised by the approaches other people have taken to generating words that sound like a particular word or judging whether two words sound alike. The aspell program letter sequence is in its dictionary; the Soundex algorithm is often used to compare whether two words sound alike and has the advantage of being very simple and delivers results that many people seem willing to accept. Over 25 years ago I wrote some software that used a phoneme based approach and while sorting through a pile of papers recently I came across an old report used as the basis for that software. I decided to implement a word sounds-like tool to show people how I think sounds-like should be done. To reduce the work involved this initial tool is based on what I already know, which in some cases is very out of date.
Phonemes are the basic units of sound and any sounds-like software needs to operate on a word’s phoneme sequence, not its letter sequence. The way to proceed is to convert a word’s letter sequence to its phoneme sequence, manipulate the phoneme sequence to create other sequences that have a spoken form similar to the original word and then convert these new sequences back to letter sequences.
A 1976 report by Elovitz, Johnson, McHugh and Shore contains a list of 329 rules for converting a word’s letter sequence into a phoneme sequence. It seemed to me that this same set of rules could be driven in reverse to map a phoneme sequence back to a letter sequence (the complications involved in making this simple idea work will be discussed in another article).
Once we have a phoneme sequence how might it be modified to create similar sounding words?
The distinctive feature theory assigns ten or so features to every phoneme, these denote details such as such things as manner and place of articulation. I decided to use these features as the basis of a distance metric between two phonemes (e.g., the more features two phonemes had in common the more similar they sounded). The book “Phonology theory and analysis” by Larry M. Hyman contains the required table of phoneme/distinctive features. Yes, I am using a theory from the 1950s and a book from the 1970s, but to start with I want to recreate what I know can be done before moving on to use more modern theories/data.
In practice this approach generates too many letter sequences that just don’t look like English words. The underlying problem is that the letter/phoneme rules were not designed to be run in reverse. Rather than tune the existing rules to do a better job when run in reverse I used the simpler method of filtering using letter bigrams to remove non-English letter sequences (e.g., ‘ck’ is not acceptable at the start of a word letter sequence). In preInternet times word bigram information was obtained from specialist cryptographic publishers, but these days psychologists researching human reading are a very good source of reliable information (or at least one I am familiar with).
I have implemented this approach and the system currently supports the generation of:
- letter sequences that sound the same as the input word, e.g., cowd, coad, kowd, koad.
- letter sequences that sound similar to the input word, e.g., bite, dight, duyt, gight, guyt, might, muyt, pight, puyt, bit, byt, bait, bayt, beight, beet, beat, beit, beyt, boyt, boit, but, bied, bighd, buyd, bighp, buyp, bighng, buyng, bighth, buyth, bight, buyt
- letter sequences that sound like the input word said with a German accent, e.g., one, vun and woven, voughen, vuphen.
The output can be piped through a spell checker to remove nondictionary letter sequences.
How accurate are the various sequence translations? Based on a comparison against manual translation of several thousand words from the Brown corpus Elovitz et al claim around 90% of words in random text will be correctly translated to phonemes. I have not done any empirical analysis of the performance of the rules when used to convert phoneme sequences to letters; it will obviously be less than 90%.
The source code of the somewhat experimental tool is available for download. Please note that the code has only been built on Linux, is likely to be fragile in various places and needs a recent copy of the pcre library. Bug reports welcome.
Some of the uses for a word’s phoneme sequence include:
- matching names contained in information transcribed using different conventions or by different people (i.e., slight spelling differences).
- better word splitting at the end of line in LaTeX. Word splitting decisions are best made using sound units.
- better spell checking, particularly for non-native English speakers when coupled with a sound model of common mistakes made by speakers of other languages.
- aid to remembering partially forgotten words whose approximate sound can be remembered.
- inventing trendy spellings for words.
Where next?
Knowledge of the written and spoken word had moved forward in the last 25 years and various other techniques that might improve the performance of the tool are now available. My interest in the written, rather than the spoken, form of code means I have only followed written/sound conversion at a superficial level; reader suggestions on more modern theories, models and data sources that might be used to improve the tools performance are most welcome. A few of my own thoughts:
- As I understand it modern text to speech systems are driven by models derived through machine learning (i.e., some learning algorithm has processed lots of data). There might be existing models out there that can be used, otherwise the MRC Psycholinguistic Database is a good source for lots for word phoneme sequences and perhaps might be used to learn rules for both letter to phoneme and phoneme to letter conversion.
- Is Distinctive feature theory the best basis for a phoneme sounds-like metric? If not which theory should be used and where can the required detailed phoneme information be found? Hyman gives yes/no values for each feature while the first edition of Ladeforded’s “A Course in Phonetics” gives percentage contribution values for the distinctive features of some phonemes; subsequent editions don’t include this information. Is a complete table of percentage contribution of each feature to every phoneme available somewhere?
- A more sophisticated approach to sounds-like would take phoneme context into account. A slightly less crude approach would be to make use of phoneme bigram information extracted from the MRC database. What is really needed is a theory of sounds-like and some machine usable rules; this would hopefully support the adding and removal of phonemes and not just changing existing ones.
As part of my R education I plan to create an R sounds-like package.
In the next article I will talk about how I used and abused the PCRE (Perl Compatible Regular Expressions) library to recognize a context dependent set of rules and generate corresponding output.
Recent Comments