PolyglotMan project takes man pages from most of the popular flavors of UNIX and transforms them into any of a number of text source formats. In contrast various man2html filters, which essentially translate bold and italic text and otherwise wrap the entire manual page in PRE tags, PolyglotMan tries to interpret the page and produce good HTML text that can be reflowed.
Better translations are possible when working from source and some aspects of pages, like tables, require source input, as for instance the page for tbl itself. Several have written cgi programs for WWW to format man pages on the fly; these are collected in the contrib directory of the distribution.
PolyglotMan was formerly known as RosettaMan. The name of the binary is still called rman, for scripts that depend on that name (mnemonically, think "reverse man"). Previously PolyglotMan required pages to be formatted by nroff prior to its processing. With version 3.0, it prefers [tn]roff source and usually produces results that are better yet. And source processing is the only way to translate tables. Source format translation is not as mature as formatted, however, so try formatted translation as a backup.