RPR1 · YLK1 · query path
The Franklin machines took a query like God? love?: a trailing question mark expanded a
word against the words actually in the lexicon, matching regardless of case, and several terms were
an AND over verses. Two sets, and a verse has to contain a member of each. That is the query
this page runs, against both containers, counting every byte each one has to read to answer it.
A double question mark went one step further: it looked the word up in a small on-device
thesaurus first, and expanded every synonym it found by the same stemming rule. So
love?? becomes love, charity, affection, delight, and then each of those becomes its own
family of forms. The thesaurus here is 239 groups over words this corpus actually contains, in
1,006 bytes — laid out after US
5,551,049, filed the same year as the link codec, which stores a group as a run of dense word
numbers and infers its length from where it sits.
The comparison is the whole point of YLK1's design. RPR1 stores text and only text, so it has no choice but to read all of it. YLK1 stores each word as a pointer to its own next occurrence, which means the chains are the posting lists — the index is not a second structure to be read, it is the same bytes as the text.
| word | lexicon row | occurrences | chain bytes | bits each |
|---|