Sam Clark of T&T Productions, the copy editor for the third edition of MATLAB Guide (co-authored with Des Higham and to be published by SIAM in December 2016), recently asked whether we would like to change “row-wise” to “rowwise”.
A search of my hard disk reveals that I have always used the hyphen, probably because I don’t like consecutive w’s. Indeed, in 1999 I published a paper Row-Wise Backward Stable Elimination Methods for the Equality Constrained Least Squares Problem .
A bit more searching found recent SIAM papers containing “rowwise”, so it is clearly acceptable usage to omit the hyphen..
My dictionaries and usage guides don’t provide any guidance as far as I can tell. Here is what some more online searching revealed.
- The Oxford English Dictionary does not contain either form (in the entry for “row” or elsewhere), but the entry for “column” contains “column-wise” but not “columnwise”.
- The Google Ngram Viewer shows a great prevalence of the hyphenated form, which was about three time as common as the unhyphenated form in the year 2000.
- A search for “row-wise” and “rowwise” at google.co.uk finds about 724,000 and 248,00 hits, respectively.
- A Google Scholar search for “row-wise” and “rowwise” finds 31,600 and 18,900 results, respectively. For each spelling, there are plenty of papers with that form in the title. The top hit for “rowwise” is a 1993 paper The rowwise correlation between two proximity matrices and the partial rowwise correlation, which manages to include the word twice for good measure!
Since the book is about MATLAB, it also seemed appropriate to check how the MATLAB documentation hyphenates the term. I could only find the hyphenated form:
doc flipdim: When the value of dim is 1, the array is flipped row-wise down
But for columnwise I found that MATLAB R2016b is inconsistent, as the following extracts illustrate, the first being from the documentation for the Symbolic Math Toolbox version of the function.
doc reshape: The elements are taken column-wise from A ... Reshape a matrix row-wise by transposing the result. doc rmmissing: 1 for row-wise (default) | 2 for column-wise doc flipdim: When dim is 2, the array is flipped columnwise left to right. doc unwrap: If P is a matrix, unwrap operates columnwise.
So what is our conclusion? We’re sticking to “row-wise” because we think it is easier to parse, especially for those whose first language is not English.