There are certain word which have different pronunciation in different languages. As a result , they end up being spelt differently . Example of such word include name of people , city name s of dishes, etc. take , for example , the capital of India- New Delhi. Delhi is also pronounced as Dilli in Hindi . Hence , it is not surprising to find both variants in an uncleaned text corpus . Similarly,
the surname ‘Agrawal’ has various spellings and pronunciations. Performing stemming or lemmatization to these words will not help us as much because the problem of redundant tokens will still be present. Hence, we need to reduce all the variations of a particular word to a common word.
To achieve this, you’ll need to know about what is called as the phonetic hashing technique.
Phonetic hashing buckets all the similar phonemes (words with similar sound or pronunciation) into a single bucket and gives all these variations a single hash code. Hence, the word ‘Dilli’ and ‘Delhi’ will have the same code.
Phonetic hashing is done using the Soundex algorithm. American Soundex is the most popular Soundex algorithm. It buckets British and American spellings of a word to a common code. It doesn’t matter which language the input word comes from – as long as the words sound similar, they will get the same hash code.
Now, let’s arrive at the Soundex of the word ‘Mississippi’. To calculate the hash code, you’ll make changes to the same word, in-place, as follows:
- Phonetic hashing is a four-letter code. The first letter of the code is the first letter of the input word. Hence it is retained as is. The first character of the phonetic hash is ‘M’. Now, we need to make changes to the rest of the letters of the word.
- Now, we need to map all the consonant letters (except the first letter). All the vowels are written as is and ‘H’s, ‘Y’s and ‘W’s remain unencoded (unencoded means they are removed from the word). After mapping the consonants, the code becomes MI22I22I11I.
- The third step is to remove all the vowels. ‘I’ is the only vowel. After removing all the ‘I’s, we get the code M222211. Now, you would need to merge all the consecutive duplicate numbers into a single unique number. All the ‘2’s are merged into a single ‘2’. Similarly, all the ‘1’s are merged into a single ‘1’. The code that we get is M21.
- The fourth step is to force the code to make it a four-letter code. You either need to pad it with zeroes in case it is less than four characters in length. Or you need to truncate it from the right side in case it is more than four characters in length. Since the code is less than four characters in length, you’ll pad it with one ‘0’ at the end. The final code is M210.
Since the process is fixed, we can simply create a function to create a Soundex code of any given input word. Learn how to make such function from professor Srinath as he explains this with a Jupyter notebook. Download the Jupyter notebook from the link given below to follow along:
NOTE: Please follow the notebook provided below, there is a slight change int he function to make it perform correctly.
Up next, you’ll learn how to identify and measure the ‘distance between words’ using the concept of edit distance which will help you build your own spell corrector.