Faster MEM-finding in $O (r + \bar{r} + g)$ space

Suppose we are given a text $T [1..n]$, a straight-line program with $g$ rules for $T$ and an assignment of tags to the characters in $T$ such that the Burrows-Wheeler Transform of $T$ has $r$ runs, the Burrows-Wheeler Transform of the reverse of $T$ has $\bar{r}$ runs and the tag array -- the list of tags in the lexicographic order of the suffixes starting at the characters the tags are assigned to -- has $t$ runs. If the alphabet size is at most polylogarithmic in $n$ then there is an $O (r + \bar{r} + g + t)$-space index for $T$ such that when we are given a pattern $P [1..m]$ we can compute the maximal exact matches (MEMs) of $P$ with respect to $T$ in $O (m)$ time plus $O (\log n)$ time per MEM and then list the distinct tags assigned to the first characters of occurrences of that MEM in constant time per tag listed, all correctly with high probability.