1: \begin{abstract}

2:   Let $\I_{g,\ast}$ denote the {\em Torelli group} of the genus

3:   $g\geq 2$ surface $S_g$ with one marked point.    This

4:   is the group of homotopy classes (rel basepoint) of homeomorphisms of $S_g$ fixing the basepoint and acting trivially on  $H\coloneq H_1(S_g,\Q)$.  In 1983 Johnson constructed a beautiful

5:   family of invariants

6:   \[\tau_i\colon H_i(\I_{g,\ast},\Q)\to \bwedge^{i+2}H\] 

7:   for $0\leq i\leq 2g-2$, using a kind of Abel--Jacobi map for

8:   families.  He used these invariants to detect nontrivial cycles in $\I_{g,\ast}$.

9:   Johnson proved that $\tau_1$ is an isomorphism, and asked

10:   if the same is true for $\tau_i$ with $i>1$.

11:   

12:   The goal of this paper is to introduce various methods for computing

13:   $\tau_i$; in particular we prove that $\tau_i$ is not injective 

14:   for any $2\leq i<g$, answering Johnson's question in the negative.  We also show that $\tau_2$ is

15:   surjective.  For $g\geq 3$, we find many classes in the image of

16:   $\tau_i$ and use them to deduce that $H_i(\I_{g,\ast},\Q)\neq 0$ for each $1\leq

17:   i< g$.  This is in contrast with the case of mapping class groups.  Many of our classes are

18:   stable, so we can deduce that $H_i(\I_{\infty,1},\Q)$ is

19:   infinite-dimensional for each $i\geq 1$.  Finally, we conjecture a

20:   new kind of ``representation-theoretic stability'' for the homology

21:   of the Torelli group, for which our results provide evidence.

22: \end{abstract}