Is it possible to automatically enumerate sentences in LaTeX?

Question

1.0.1 Is it possible to automatically enumerate sentences in LaTeX?

2.0.1 I want to thoroughly work through and criticize a longer text. 2.0.2 To facilitate quoting, I would like to typeset the text and enumerate the sentences in this way. 2.0.3 Some time ago, I saw this done in a complex game ruleset.

2.1.1 In the old ruleset, the enumeration was typeset in subscript. 2.1.2 The sentence enumeration was a continuation of the chapter enumeration. 2.1.3 In all aspects except adding the sentence number, the text should be typeset in the normal way. 2.1.4 The numbering should be added automagically without the use of a \beginsentence macro.

3.0.1 In case this works at all, is it possible to additionally reference sentences by their numbers [labels], and have the label automatically updated in case the numbering in the text changes? 3.0.2 Of course, 2.1.4 implies something along the lines of \begin{sentencenumbering} or so.

Answer 1

As other have pointed out, doing it completely automatically is probably quite difficult. If you want to use the \label-\ref mechanism, you would have to insert labels anyway. Let's pick some character which is usually not used in input, such as the vertical bar |. Ten minutes of hacking and we end up with:

\documentclass{article}

\newcounter{sentence}
\newcounter{para}

\makeatletter
\@addtoreset{sentence}{para}
\@addtoreset{para}{section}

\catcode`\|=\active
\def|{\@ifnextchar[%] to keep my editor happy
  \start@label\start@nolabel}

\def\start@label[#1]{\ifvmode \start@para@label[#1]\else \start@sent@label[#1]\fi}
\def\start@nolabel{\ifvmode \start@para@nolabel\else \start@sent@nolabel\fi}

\def\start@para@label[#1]{%
  \refstepcounter{para}%
  \label{#1}\leavevmode}

\def\start@sent@label[#1]{%
  \refstepcounter{sentence}%
  \label{#1}%
  \thesentence~}

\def\start@para@nolabel{%
  \stepcounter{para}\leavevmode}

\def\start@sent@nolabel{%
  \stepcounter{sentence}%
  \thesentence~}

\makeatother

\renewcommand{\thepara}{\thesection.\arabic{para}}
\renewcommand{\thesentence}{\thepara.\arabic{sentence}}

\begin{document}

\parindent=0pt
\parskip=1em

||These rules must be followed. |The end of a paragraph is indicated
as usual with a blank line.

||[parstart]A new paragraph must start with a vertical
bar. |[sentstart]Each sentence must also start with a vertical
bar. |It follows from~\ref{parstart} and~\ref{sentstart} that a new
paragraph actually starts with two vertical bars.

Without vertical bars, nothing special happens. This might be useful
to comment on the formal rules above or below.

|[parref]|Each vertical bar takes an optional argument. |If
given, it is used as a label. |[sentref]For example, this is
sentence~\ref{sentref} of paragraph~\ref{parref}.

\end{document}

I use the fact that after a \par we're in vertical mode to distinguish the two uses of |. But then we need to explicitly \leavevmode, since the beginning of a paragraph itself does not insert material causing us to switch to horizontal mode. If you want to be able to reference both whole paragraphs and single sentences, we need the double || at the start of each paragraph (they might both have an optional argument). If you never need to reference whole paragraphs, it's easy to change the syntax so that only a single | is needed at the start of a paragraph; in fact, it would be much simpler, since | could be implemented as a single macro with optional argument (which, when in \ifvmode, steps the paragraph counter before doing the other stuff).

Added I avoided using \everypar since it's not very reliable if lots of other things are happening. However, wrapping stuff in an environment might allow one to use \everypar and provide a simpler syntax. The biggest problem is really to allow the use of labels; we have to tell LaTeX when and how to look for a label.

Answer 2

Not easily¹ with TeX or LaTeX or anything else.

The problem is what is generally known in natural language processing as Sentence boundary disambiguation

Sentence boundary identification is difficult because punctuation marks are often ambiguous. A period may denote an abbreviation, decimal point, an ellipsis, or an email address - not the end of a sentence. In addition sentences can end by exclamation marks or question marks.

A better approach is to pre-process the file outside TeX. The NLTK written in Python might be a starting point.

¹ not easily meaning, if you devote a good chunk of your time you may be able to define a TeX parser to capture 95% of cases using TeX.

Answer 3

Obviously Yiannis is right.

However, if you can live with a trade-off then you can perhaps just redefine the macros \\ and \par (which is implicitly inserted whenever you leave an empty line) and write your sentences like this:

First sentence.\\
Second sentence.\\

Third sentence.

And end up with:

1.1 First sentence. 1.2 Second sentence.
2.1 Third sentence.

This requires two counters, one counting the sentences and one counting the paragraphs.

Answer 4

I have used automatically generated labels for generating a nomenclature with references to the place where a variable is introduced and explained in the main text.

You might be able to automatically construct meaningful labels from chapter, section, subsection, ... and so on ... snippets and end up with

\label{sentence:chapSnippet:sectionSnippet:subSectionSnippet:parSnippet}

At the beginning of each of these, redefine the snippet. Like:

\chapter{Insects}
\renewcommand{\chapSnippet}{insects}

You would then end up with labels like insects:humblebee:I for the first sentence (roman I) of the humblebee section in Insects. It's a kludge, but can still use custom labels inserted everywhere. A custom ref command can give you almost any reference format you like.

ADDED: Does anyone know a solution to this question: Building labels out of counters ? I didn't test the answer provided there (no latex on this machine).

Answer 5

The package scrjura implements enumeration of sentences

I gave an example here: http://tex.stackexchange.com/a/87344/4736

Answer 6

Also, paralist comes to mind.

\documentclass{scrartcl}
\pagestyle{empty}
\usepackage{paralist}

\begin{document}
  \begin{inparaenum}[1.]
    \item \label{first:sentence} This is a test.
    \item Second sentence referring to \ref{first:sentence}.
  \end{inparaenum}
\end{document}

See also working example on GitHub.

However, I couldn't get it to produce a nested numbering. But labeling works.