Auto-completion and spell checkers are everywhere: word processors, phones, etc. Educators complain
that students don’t know how to spell because spell checkers are so ubiquitous. Let’s write a simple
program that does completion and spell checking. It isn’t efficient enough to be used in a commercial
application, but you’ll get the idea of how you might write one.
Word completion is often based on the user’s typing history, and the suggestions are usually ranked by
how frequently the word is used. Also, many tools will give context-based word completion. For this
project, we will be writing a very simple word-completor. We simply use a collection of valid words, and
return all possible completions of some prefix.
A good example
Peter Norvig, a very well known computer scientist (https://en.wikipedia.org/wiki/Peter_Norvig) has a
very popular blog where, one day, he tackled the problem of writing a computer spelling corrector in
Python (https://norvig.com/spell-correct.html on a plane trip in around 3-4 hours). We’re going to use
some of his ideas and write a spell corrector in the same way.
Making spelling mistakes
We will consider 4 kinds of mistakes one might make in typing a word. Let’s use the word something
● a delete : the typist accidentally forgets to type a letter: somthing
● an insert : the typist “fat fingers” the keys and adds an extra letter: somethjing
● a transpose : the typist switches the order of two letters : somehting
● a replacement: the typist types the wrong letter : somrthing
Of course, a typist could make more than one of those kinds of mistakes on a word, but the chances are
lower that this occurs. However, we’ll at least consider the possibility of two such mistakes in the same
We are going to build functions for those 4 kinds of mistakes. The possibilities they generate could,
however, be large. Sticking with the word something, what are the counts for the 4 different errors:
● deletes : 9 (there are 9 letters)
● transpose : 8 (there are 8 pairs)
● replacement : 225, 9 locations where 25 possible other letters could be substituted (not counting the
letter already there, 226 if we do count those letters)
● inserts : 260 , 10 locations (including the locations before and after the word) * 26 (all letters of the
If we take all the possibilities (9+8+225+260 = 502), then we must evaluate whether any of those 502
might be the intended word. We can go one step further. For each of the 502 possibilities we can run each
one through the 4 mistakes generating a yet much larger set, allowing for the possibility of having two
errors in a word. I generated the following numbers for the double deletion error somthng
1 level possibilities: 390
2nd level Dels: 2905
2nd level Reps: 73490
2nd level Trans: 2534
2nd level Inserts: 83240
2nd level possibilities: 70184
You may notice that thelevel possibilities don’t add up. That’s because there was repetition in the
generation of the possibility words. We ignore duplicates. Nonetheless, that is a lot of possible words, far
more than we can present to a user. How to sort that out? We can take those 70,000+ words and see if any
of them show up in a big list of English words. If we did that, we would get only 4 positive results which
I call the reasonables
Reasonables : something,soothing,sorting,southing
A good spell checker would offer up those 4, reasonable, possibilities.
You probably knew this, but efficiency is going to matter a bit here. To go from somthng to the 4
reasonable replacements should, practically, occur pretty quickly. We probably can’t be faster than 2-3
seconds but it can’t be 30 seconds. We have to consider efficiency
To that end we are going to use sets and maps. These are data structures that can do fast
search/merge/inserts, all of which we are going to need. We will also pass all the sets/maps as references,
including the result set/map. That is, we will pass in a set/map which is to contain the answer. In this way
our functions don’t have to return a result by copy.
Complete the Project 7 by writing code for the following functions. Details of type for the functions can
be found in proj7_unctions.h (provided for you, see details below). The
● void deletes (const string & word, set
o creates all the single letter deletes possible from the argument word and places them in the
o void return (meaning no return value.) All deletes are inserted into the set
● void replaces(const string &word, set
o creates all the single letter replaces possible from the argument word and places them in the
▪ We allow the original letters to occur in the replacement set, making it potentially +1
against the size of just the replacement set.
o void return (meaning no return value.) All replacements are placed in the set
● void transposes(const string &word, set
o creates all the single letter transpositions possible from the argument word and places them
in the argument result.
o void return (meaning no return value.) All transpositions are placed in the set
● void inserts(const string &word, set
o creates all the single letter inserstions possible from the argument word and places them in
the argument result.
o void return (meaning no return value.) All inserstions are placed in the set
● void read_words(string fname, set
o reads a file, the file name indicated by fname, into set
file named words.txt with about 112,000 words, each word on a separate line. Write the
function to read files with multiple words on the same line.
o each word is converted to lower case before it is added to the set.
o while I believe that all the words in the provided file are alphabetic, you must only store
alphabetic characters. Thus non-alphabetic character are deleted: Two4Me twome
o void return (meaning no return value.) All insertions are placed in the set
● void find_completions(const string &w,
o takes in a word ‘w’ as a string, and a set of valid words.
o void return. All the words in word_list that have ‘w’ as a prefix are added to result.
This might be a very large set!
● void find_corrections(const string &word, set
o takes in a word as a string, and calculates the deletes, inserts, transposes, and replaces sets for
o void return. All the dels, reps, trans and ins for each word are added to result. This
might be a very large set!
● void find_2step_corrections(const string &word, set
o takes in a word as a string, and calculates 2 levels of the deletes, inserts, transposes, and
replaces sets for the word. i.e. for each word in 4 sets, all corrections are applied again.
o void return. All the dels, reps, trans and ins for each word are added to result. This
might be a very, very large set!
● void find_reasonables(const set
o for every word in possibles, check if that word exists in the word_list. If it does, it is
added to result
o void return
1. proj07_functions.cpp — your completion of the functions.
2. You are given proj07_functions.h and a list of words in words.txt Remember to include
your section, the date, project number and comments.
3. Turn it into Mimir as always. Remember, no main!
1. If you are going to use algorithms, and you probably should, you have to be careful about
constant parameters. You cannot iterate form a .begin() to .end() on a const
constant-ness. However, you can use .cbegin() to .cend(), which yields constant pointers
and that will work.
2. The set_union algorithm creates a union of two STL data structures, placing the result in a
third data structure. The first 4 arguments are the .begin() and .end() of the two structures
being unioned. However, the 5 argument, for this exercise, should be an inserter iterator to a
set. This allows for inserting to a set and removing a duplicate if it occurs. The inserter
iterator takes two arguments: the set we are inserting into and the location to do the insertion. The
location is really irrelevant as the set will re-order itself anyway, but it must be provided. We use
the .end() iterator as a place for the location (because it’s easy).
inserter(result, result.end() );
See the example inserter.cpp in the project directory.
3. You should never check for duplicates when you are adding something to a set. The set will
automatically stop a duplicate from being added. Just insert everything in and let the set sort it
4. We pass everything by constant reference except the parameter that will contain the result. We
pass that as a reference, not const, so we can modify it. This assumes that you pass the result with
the correct values in place (empty perhaps, a union of 2 other sets, whatever is appropriate).
5. To find if something is in a set (such as a word from the possibles in the word_list), you can do
three things, one of them bad:
a. myset.count(the_word) 0 if not there, 1 if it is
b. myset.find(the_word) myset.end() if not there, an iterator to the word if it
c. find(myset.begin(), myset.end(), the_word);
The first and the second are fast algorithms. They use the alphabetical order of the set to do a
binary search. This is (for lack of a better analogy), like the Price is Right Clock Game
(http://priceisright.wikia.com/wiki/Clock_Game ). You guess a price, the host says “higher” or
“lower” and you guess again. The strategy, to us anyway, is obvious. Guess the middle value. If
“lower”, then guess in the middle of the lower range, otherwise in the higher range. Repeat.
The last one does a linear search (looking at each element in order). That is much slower!!!
6. Overall, algorithms are going to be the way to solve this. Look at the algorithms we mentioned in
the videos and notes and see if they apply. If you have questions on Piazza, questions about
algorithms (which to use, which is most appropriate, how to use it) would help you the most.