How big is my book: Mashcat session

At Mashcat on 5 July in Cambridge I gave an afternoon session on getting computer readable information from the textual information held in MARC21 300 fields using Javascript and regular expressions. I intended this to be useful for cataloguers who might have done some of Codecademy‘s Code Year programme as well as an exploration of how data is entered into catalogue records, its problems, and potential solutions.

AACR2/MARC (and RDA) records store much quantitative information as text, usually as a number followed by units, e.g. “31 cm.” or “xi, 300 p”. This is not easy for computers to deal with. For instance, a computer programme cannot compare two sizes- e.g. “23 cm.” and “25 cm.”- without first extracting a number out of the string (23 and 25) as well as determining the units used (cm). In some cases, units might vary: in AARC2 books below 10 cm. are measured in mm., and non-book materials are often measured in inches (abbreviated to in.). Potential uses for better quantitative data in the 300$c include planning shelving for reclassification and more easily finding books by size or range.

Before the session, I sketched out a possible solution using Javascript and regular expressions to make this conversion for dimensions in the 300$c. I have a put up a version of A script to find the size of an item in mm. based on the 300$c, with the addition of an extra row which you can fill in to test your own examples without having to edit the script.

If you do want to look at how it works or try editing it yourself you can view source, copy all the HTML, then paste it into a text editor. Save it, then open the file using a browser to test it. Refresh the browser when you change the file.

The heart of the script looks like this:

var dollar_c = [
  "9 mm",
  "4 in.",
  "4 3/4 in.",
  "30 cm.",
  "1/2 in.",
  "20 x 40 cm."
];

// Convert text to mm
function text_to_mm (text) {
  // Convert fractions to decimals
  text = text.replace(/(\d*) (\d)\/(\d)/, function(str, p1,p2,p3) {return parseFloat(p1)+p2/p3});
  text = text.replace(/(\d)\/(\d)/, function(str, p1,p2) {return parseFloat(p1/p2)});
  // Extract the size of the book
  size = text.replace (/([\d\.]*).*/, "$1");
  // Extract the units
  units = text.replace(/.*([a-z]{2}).*/g, "$1");
  // Convert from various units to mm
  if (units === "mm") {
    var mm = size;
  }
  if (units === "cm") {
    var mm = size * 10;
  }
  if (units === "in") {
    var mm = size * 25.4;
  }
  mm=Math.floor(mm);
  return mm;
}

It starts with a declaration of an array of examples to be tested: you can alter this with your own if you prefer. text_to_mm is the function that does all the work. It takes in the text from a 300$c, converts fractions (e.g. 4 3/4) to decimals (4.75), finds a number, finds a unit, then performs calculations on the size depending on what the unit is to produce a figure to a standard figure in mm. At Mashcat, Owen Stephens managed to plug an adaptation of this script into Blacklight to create an index of book sizes. Using this he could do things like find the most common sizes or the largest book in a collection.

The main focus of my session, however, was on a similar script to figure out how many actual pages there are in a book, given the contents of a 300$a, e.g. “300 p.”, “ix, 350 p.”, “100 p., [45] leaves of plates”  (a page being one side of a sheet of paper; a leaf being a sheet of paper only printed on one side, so therefore counting as two pages). I have also published a version of A script to find the absolute no. of pages based on the 300$a with the similar addition of a row for easy user testing. Potential uses for recording page numbers rather than pagination include planning shelving space, easier to understand displays for users, and finding books of specified lengths.

The script starts with a similar array of examples to be tested:

// An array of test examples
var dollar_a = [
  "9 p.",
  "9p",
  "30 leaves",
  "30 p., 20 leaves",
  "xiv, 20 p.",
  "20, 30 p.",
  "20, 30, 40 p.",
  "xv, 20, 30, 40 p., 5, 5 leaves of plates",
  "clviii, ii, 4, vi p."
];

The main function is called text_to_pages. The first thing it does is convert any Roman numerals to Arabic ones. The heavy lifting for this is a function by Stephen Levithan which does the actual number conversion. However, we still need to identify and extract the Roman numerals from the pagination in order to convert them. This line does the extraction and makes a list of the Roman numerals:

var roman_texts=text.match(/[ivxlc]*[, ]/g);

The session I gave concentrated on regular expressions (a bit like the wildcards you use on library databases but turned up to eleven) which in all cases here are contained within slashes, and I made a simple introductory guide to regular expressions (.docx). There are many guides to regular expressions on the web too, and useful testers to play with such as this one. The regular expression in the line above can be broken down as follows:

• [ivxlc] uses square brackets to look for any one of the characters listed within them.
• The following * means to look for any number of these in a row
• [, ] any of a comma or a space, again using square brackets. Obviously these characters are not used in Roman numerals but they are a convenient method of isolating these characters as numbers rather, say, the “l” in leaves which would also match otherwise.

The next few lines work through the list, replace any instances of [, ] with “” (i.e. nothing) to leave the bear Roman numerals, convert all the numbers in the list using Stephen Levithan’s functions, then do the replacements on the pagination given in text:

if (roman_texts) {
    for (i=0; i<roman_texts.length; i++) {
      // Remove space
      roman_texts[i]=roman_texts[i].replace(/[, ]/,"");
      var arabic_text = deromanize(roman_texts[i])+" ";
      text = text.replace(roman_texts[i],arabic_text+" ");
    }
  }
}

Like the size script above, the rest of conversion needs to do two things: find the numbers and find the units. To do this we need to find the sequences involved. While this is easy with something like “24 p.” (number is 24, unit is p) or even “xv leaves” (number is 15, unit is leaves), it becomes troublesome when you get something like “23, 100 p.”: the first number is 23 but there is no unit associated with it, only a comma to signify that it is a sequence at all. The following lines try and get round this problem but looking for sequences where the comma appears to be the unit and then looking ahead to find the next unit. In the “23, 100 p.” example the script would keep looking forward past the 100 until it gets to the “p”.

// Convert 20, 30 p. to 20 p. 30 p
  while (text.match(/\d*,/)) {
    text = text.replace(/(\d*),(.*?(p|leaves))/, "$1 $3 $2");
  }

The first regular expression in the while line looks for:

• \d* any number of digits. \d is any digit and * looks for any number of them, followed by
• , a comma

So as long as the script finds any sequences of numbers followed by a comma, it will carry on making the replacement underneath it. The replacement line itself looks for

• \d* any number of digits again, followed by
• , a comma
• .*? which is . any character * any number of times. The ? makes sure that the smallest matching group of characters is matched; otherwise the expression will think that the units corresponding to the number 15 in the pagination “15, 25 p., 50 leaves” is “leaves” rather than “p”.
• p|leaves either p or leaves. The pipe means either match on the left of it or the right of it. Because this is in a set of round brackets, the pipe only applies there, rather than the whole expression.

Brackets also capture subsets which is really useful here: the first set of () brackets captures the number of pages and stores it as $1, the second set captures everything between the comma and the end of the units as $2, the third  set captures the units only, either “p” or “leaves”, and stores it as $3. So in the example “15, 25 p., 50 leaves”, $1 is “15”, $2 is ” 25 p”, and $3 is “p”. The replacement puts these back in a different order, i.e. “$1 $3 $2” which would be “15 p 25 p”.

Now that all the sequences will be in number-unit pairs, we can get on with making a list of them to work through:

 // Find sequences
  var sequences = text.match(/\d+.*?(,|p|leaves)/g);

This looks for:

• \d+ at least one digit
• .*? any number of any characters, although not being greedy
• (,|p|leaves) any of a comma, “p”, or “leaves”. Obviously, if the while loop above has worked, then the comma isn’t needed, but I’ll confess this is a hangover from a previous version of the script…

The next section goes through each of the sequences found and extracts the number and then the unit:

// go through sequences
  var pages = 0;
  for (var i=0; i<sequences.length; i++) {
    // Extract no
    var number = parseFloat(sequences[i].match(/\d+/g)[0]);
    var units = sequences[i].match(/(p|leaves)/g)[0];
    if (units == "p") {
      pages+=number;
    }
    if (units == "leaves") {
      pages+=number*2;
    }
  }

The regular expression to find the number is straightforward:

• \d+ at least one digit

The parseFloat converts the digits as a string to a Javascript number. The regular expression to find the unit is also simple:

• (p|leaves) either “p” or “leaves”

If the units are “p”, then the variable pages is incremented by the value of the number found; if “leaves”, then pages is incremented by twice that number.

The programme should cope with the loss of abbreviations in RDA as “p.” is expanded to “pages” but the regular expression to find the units will still find the “p” at the beginning much as it isn’t put off by the full stop after the “p”. It could be expanded to look for other variations and I will do so if I can:

• “S.” for German “Seite” or “Seiten”.
• “leaf”, as in “1 leaf of plates”
• sequences which start in the middle of larger ones, like journal issues with “xii, p. 546-738”. This one will be the most complicated as it goes against the basic flow of the existing code.

I also haven’t properly tested folded sheets or multiple volume works. Other improvements are needed in failing more gracefully when it doesn’t find what it’s expecting: the programme should really test the existence of the arrays it makes before looping through them, but this would make it harder to understand at a glance or demonstrate on screen so I didn’t do it.

The scripts are written in Javascript for several reasons: it is the language that Codecademy focusses on for beginners; it requires no specialist environment, server, or even a web connection: you just need a basic text editor and a browser; it is easy to adapt for a web page if you do manage to build something; and, it is the language I am most confident working in. It would be fairly easy to port to other languages though, and Owen changed the size script with some other modifications to work in Beanscript/Java in Blacklight.

I can’t speak for the attendees, but I learnt a lot, and much was made more clear, from playing around with these scripts and talking to people at Mashcat:

• Quite how depended AARC2 and RDA (and consequently MARC21) are on textual information, even for what appears to be quantitative data.
• That even for what appears to be standard number-unit data, there are too many complications that make it non trivial to extract data:
  • fractions (not even decimals) in 300$c
  • differing units: book sizes in mm. or cm. depending on how big the book is; disc sizes in in.; extent in pages or leaves (or volumes or atlases or sheets…)
  • sequences with implied units, such as those with commas.
• there is frequently a lack of clarity and ambiguity of what is actually being measured:
  • for books the dimension recorded is normally height (although this is not explicit from a user’s point of view,  sometimes it’s height and width, and for a folded sheet it could be all sorts of things); for a disc it’s the diameter.
  • For the 300$a what’s being recorded is pagination, something entirely different from number of pages. Although important for things like rare books, how important is complete pagination for most users compared to a robust idea of how large a book is? Amazon provide a number of pages. More importantly, how understandable is pagination? During my demonstration, some of my audience of librarians were left cold by the meanings of square brackets for example (and square brackets can mean any number of things depending on context). Perhaps there is room for both.

I suppose this latter point is a potential conclusion. Ed Chamberlain asked me what I thought should be done. I don’t know to be honest. I think, like much of the catalogue record, lots more research is needed to see what users (both human and computer) actually want or need. It should be said that entering pagination is in many ways easier for the cataloguer. However, I do think we need:

• quantitative data entered as numbers with clear and standard units. For instance, record all book heights as mm. and convert to cm. for display if needed.
• more data elements to properly make clear what is being recorded. Instead of a generic dimension, we need height, width, depth?, diameter, etc. Instead of pagination, we could have separate elements for pagination, number of pages, and number of volumes (50 volumes each of 10 pages is not the same as 4 volumes of 1000 pages each). Obviously all of them wouldn’t be needed for all items.

The research to enable us to choose what to record, why we’re recording it, and for whose benefit would be the best starting point for this as well as many other questions in cataloguing and metadata.