TROFF to pdf or docx conversion

I have my partially complete PhD thesis in TROFF format (?) – so it would laser print from the unix station back then. It has had a layer of Atari 1stWord encapsulation, which may still be at the header, but is essentially irrelevant as it was just a text editor to work on at home – away from the university's Sun workstations.

Can anyone help me convert this to wysiwyg format of any sort please? pdf? docx?

¬066010303050000132002006006010000
¬1¬¬¬1
¬2¬#¬¬1
¬F0110000001
¬9[............................................................]0010
 Ç
.\" pic
.\" tbl
.\" eqn
.\" roff -ms
.\"
.EQ
delim $$
global
.EN
.ND
.nr PO 1.5i
.nr LL 6.0i
.nr PS 9
.nr VS 18
.fp 1 H
.TL
.ps 14
CHAPTER 7
.AU
.ps 14
KNOWLEDGE ENGINEERING
.PP
.ps 9
.NH
Introduction
.PP
In the previous chapter we outlined a system which may be used to
handle uncertainty in a domain, using probabilities
on a causal network.
In this chapter we will to discuss how the knowledge within
a domain may be acquired and manipulated into the form
of a probabilistic causal network - with specific reference to
the two domains which represent
the process of dating 
timber-framed buildings, using
(i) architectural/historical
and
(ii) dendrochronological evidence.
Our ultimate goal here is to
ascertain, as best we may, the correct date of construction of
vernacular buildings.
As mentioned in Chapter 2,
one of the main problems we face is the resolution of
the potential conflict between
dendrochronological and archaeological predictions for the date of
a building.
If the two
bodies of knowledge produce results which are totally
in agreement, we are left with no other conclusion than that
the predicted date-range/period is the correct one,
however there is rarely complete agreement and this disagreement
may be very difficult to resolve.
.PP
We will derive networks on the two domains and give numerical
 Çexamples of propagation of evidence through these networks,
closing the chapter with a short exposition of
how we may use probabilistic methods in conjunction with
causal networks to resolve this mentioned conflict.
This use of probabilistic methods is contrary
to widely adhered to
notions concerning the unsuitability of this approach - from
the general concerns over the epistemological adequacy of
probabilistic methods in artificial intelligence expressed by
McCarthy and Hayes (1969 p.490), to recent doubts about the implementational
mechanics of probabilistic formalisms to be found in Graham and
Llewelyn-Jones (1988 p.88 et seqq.).