Introduction
This article describes three approaches to parsing the sentences from a body of text; three approaches are shown as a means of describing the pros and cons for performing this task using each different approach. The demonstration application also describes an approach to generating sentence count, word count, and character count statistics on a body of text.
Figure 1: The test application running.
The three approaches to parsing out the sentences from the body of text include:
- Parse Reasonable: An approach based on splitting the text using typical sentence terminations where the sentence termination is retained
- Parse Best: An approach based on the use of splitting the text based upon the use of a regular expression and where the sentence termination is retained, and
- Parse Without Endings: An approach to splitting the text using typically sentence terminations where the terminations are not retained as part of the sentence.
The demonstration application contains some default text in a textbox
control; three buttons used to parse the text using one of the three approaches mentioned, and three label controls used to display the summary statistics generated on the body of text. Once the application is run, clicking on any of the three buttons will result in the display of each of the parsed sentences within the listbox
control at the bottom of the form, and will result in the display of the summary statistics using the three labels in the upper right hand side of the form.
Getting Started
In order to get started, unzip the included project and open the solution in the Visual Studio 2008 environment. In the solution explorer, you should note these files (Figure 2):
Figure 2: Solution Explorer.
As you can see from Figure 2, there is a single WinForms project containing a single form. All code required of this application is included in this form’s code.
The Main Form (Form1.cs)
The main form of the application, Form1
, contains all of the code necessary. The form contains default text within a textbox
control; the three buttons are used to execute each of the three functions used to parse the body of text into a collection of string
s; one per sentence. You may replace, remove, or add to the text contained in the textbox
control to run the methods against your own text. Three label controls are used to display summary statistics (sentence, word, and character counts) on the text contained in the textbox
control. These summary statistics are updated each time the text is parsed into sentences.
If you'd care to open the code view up in the IDE, you will see that the code file begins with the following library imports:
using System;
using System.Collections;
using System.ComponentModel;
using System.Data;
using System.Drawing;
using System.Text;
using System.Windows.Forms;
using System.Text.RegularExpressions;
Note that the defaults have been altered and now include the reference to the regular expressions library.
Following the imports, the namespace, class, and constructor are defined:
namespace SentenceParser
{
public partial class Form1 : Form
{
public Form1()
{
InitializeComponent();
}
Next up is a region entitled, “Best Sentence Parser”; this region contains a function entitled SplitSentences
which accepts a string
as an argument. This method tends to yield the best results in terms of parsing sentences but may issue inaccurate values if the text contains errors. The region also contains a button click event handler used to evoke the SplitSentences
function.
The code is annotated and reading through the notes will explain what is going on within the function.
#region Best Sentence Parser
private ArrayList SplitSentences(string sSourceText)
{
string sTemp = sSourceText;
ArrayList al = new ArrayList();
string[] splitSentences =
Regex.Split(sTemp, @"(?<=['""A-Za-z0-9][\.\!\?])\s+(?=[A-
Z])");
for (int i = 0; i < splitSentences.Length; i++)
{
string sSingleSentence =
splitSentences[i].Replace(Environment.NewLine,
string.Empty);
al.Add(sSingleSentence.Trim());
}
lblCharCount.Text = "Character Count: " +
GenerateCharacterCount(sTemp).ToString();
lblSentenceCount.Text = "Sentence Count: " +
GenerateSentenceCount(splitSentences).ToString();
lblWordCount.Text = "Word Count: " +
GenerateWordCount(al).ToString();
return al;
}
private void btnParseBest_Click(object sender, EventArgs e)
{
lstSentences.Items.Clear();
ArrayList al = SplitSentences(txtParagraphs.Text);
for (int i = 0; i < al.Count; i++)
lstSentences.Items.Add(al[i].ToString());
}
#endregion
Next up is a region entitled, “Reasonable Sentence Parser”; this region contains a function entitled ReasonableParser
which accepts a string
as an argument. This method tends to yield fair results in terms of parsing sentences but does not apply the proper sentence terminations if the input string
contains duplicate sentence with different terminations. This issue could be resolved by use of a recursive function to continue to move through each instance of the duplicate sentence, however it is less work to use the method indicated in the previous code region. The region also contains a button click event handler used to evoke the ReasonableParser
function.
The code is annotated and reading through the notes will explain what is going on within the function.
#region Reasonable Sentence Parser
private ArrayList ReasonableParser(string sTextToParse)
{
ArrayList al = new ArrayList();
string sTemp = sTextToParse;
sTemp = sTemp.Replace(Environment.NewLine, " ");
char[] arrSplitChars = { '.', '?', '!' };
sentence
string[] splitSentences = sTemp.Split(arrSplitChars,
StringSplitOptions.RemoveEmptyEntries);
for (int i = 0; i < splitSentences.Length; i++)
{
int pos = sTemp.IndexOf(splitSentences[i].ToString());
char[] arrChars = sTemp.Trim().ToCharArray();
char c = arrChars[pos + splitSentences[i].Length];
al.Add(splitSentences[i].ToString().Trim() + c.ToString());
}
lblCharCount.Text = "Character Count: " +
GenerateCharacterCount(sTemp).ToString();
lblSentenceCount.Text = "Sentence Count: " +
GenerateSentenceCount(splitSentences).ToString();
lblWordCount.Text = "Word Count: " +
GenerateWordCount(al).ToString();
return al;
}
private void btnParseReasonable_Click(object sender, EventArgs e)
{
lstSentences.Items.Clear();
ArrayList al = ReasonableParser(txtParagraphs.Text);
for (int i = 0; i < al.Count; i++)
{
lstSentences.Items.Add(al[i].ToString());
}
}
#endregion
Next up is a region entitled, “Parse Without Sentence Terminations”; this region contains a function entitled IDontCareHowItEndsParser
which accepts a string
as an argument. This method tends to yield good results in terms of parsing sentences but does not add the termination to the parsed sentences; this is a good approach to use if you don't care what termination is used at the end of the sentence. The region also contains a button click event handler used to evoke the IDontCareHowItEndsParser
function.
The code is annotated and reading through the notes will explain what is going on within the function.
#region Parse Without Sentence Terminations
private ArrayList IDontCareHowItEndsParser(string sTextToParse)
{
string sTemp = sTextToParse;
sTemp = sTemp.Replace(Environment.NewLine, " ");
char[] arrSplitChars = { '.', '?', '!' };
sentence
string[] splitSentences = sTemp.Split(arrSplitChars,
StringSplitOptions.RemoveEmptyEntries);
ArrayList al = new ArrayList();
for (int i = 0; i < splitSentences.Length; i++)
{
splitSentences[i] = splitSentences[i].ToString().Trim();
al.Add(splitSentences[i].ToString());
}
lblCharCount.Text = "Character Count: " +
GenerateCharacterCount(sTemp).ToString();
lblSentenceCount.Text = "Sentence Count: " +
GenerateSentenceCount(splitSentences).ToString();
lblWordCount.Text = "Word Count: " +
GenerateWordCount(al).ToString();
return al;
}
private void btnParseNoEnding_Click(object sender, EventArgs e)
{
lstSentences.Items.Clear();
ArrayList al = IDontCareHowItEndsParser(txtParagraphs.Text);
for (int i = 0; i < al.Count; i++)
{
lstSentences.Items.Add(al[i].ToString());
}
}
#endregion
The final region is entitled, “Generate Statistics”. This region contains three functions which return the character count, word count, and sentence counts for a body of text. Again, this section is annotated; read through the annotation to get a description of how each function works.
#region Generate Statistics
public int GenerateCharacterCount(string allText)
{
int rtn = 0;
string sTemp = allText;
sTemp = sTemp.Replace(Environment.NewLine, string.Empty);
sTemp = sTemp.Trim();
string[] splitSentences =
Regex.Split(sTemp,
@"(?<=['""A-Za-z0-9][\.\!\?])\s+(?=[A-Z])");
for(int cnt=0; cnt<splitSentences.Length; cnt++)
{
string sSentence = splitSentences[cnt].ToString();
sSentence = sSentence.Trim();
char[] sentence = sSentence.ToCharArray();
for (int i = 0; i < sentence.Length; i++)
{
if (char.IsLetterOrDigit(sentence[i]) ||
char.IsPunctuation(sentence[i]) ||
char.IsWhiteSpace(sentence[i]))
rtn += 1;
}
}
return rtn;
}
public int GenerateWordCount(ArrayList allSentences)
{
int rtn = 0;
foreach (string sSentence in allSentences)
{
char[] arrSplitChars = {' '};
string[] arrWords = sSentence.Split(arrSplitChars,
StringSplitOptions.RemoveEmptyEntries);
rtn += arrWords.Length;
}
return rtn;
}
public int GenerateSentenceCount(string[] allSentences)
{
int rtn = 0;
rtn = allSentences.Length;
return rtn;
}
#endregion
Summary
This article is intended to describe several approaches for parsing the sentences out of a body of text. Further, the article describes three functions which may be used to generate summary statistics on a body of text. There are of course, other ways that may be used to do each of these things. In general, the best approach to parsing out the sentences appears to be through the use of a regular expression. Modifications to the regular expression may yield different results which might work better with the sort of text you are working with; however, I have found that this approach works well with even complicated bodies of text so long as the text is properly formatted into proper sentences.
History
- 3rd June, 2008: Initial version