Velvet: Construction from overlap information

[srb_bio]

Hi,

I am just reading the original VELVET paper and I find little difficulty in understanding the graph construction part. I hope someone will help me. Following are my understanding and doubts

"The reads are first hashed according to a pre-defined k-mer lengths."

read1: ACTCGTACTTGTA (say ID is R1)
read2: GTCCGTAGTCAGT (ID is R2)

Suppose k=3
So the first hash table should contain
from read1,
ACT(1), CTC, TCG, CGT(1),GTA(1), TAC, ACT(2), CTT, TTG, TGT, GTA(2)

from read2,
GTC, TCC, CCG, CGT(2), GTA(3), TAG, AGT(1), GTC, TCA, CAG, AGT(2)

"For each k-mer observed in the set of reads, the hash table records the ID of the first read encountered containing that k-mer and the position of its occurrence within that read."

So is this another hash table?
And this hash table should contain 3-mers like the follwoing
ACT should be stored as (ID, position) i.e. (R1, 1)
Similarly,

TAC should be stored as (R1, 6)
Here we don't have to store the second occurence of ACT, GTA etc??

"This first scan allows each
read to be re-written as a set of new k-mers combined to overlaps with
previously hashed reads. This new representation of the reads sequence is
called a roadmap."

I am little confused while reading the above. What would be the roadmaps for the example I have given?