# C, 623 <del>673</del> <del>707</del> bytes

Source code:
<!-- language-all: lang-c -->

    j,c,p[256][256]={0},r;char X[999]="j,c,p[256][256]={0},r;char X[999]=\"@\";main(){char*a=X,*b=X,*Y=X+315,*y;for(y=Y;*a!=64;)*y++=*a++;for(;*b;*y++=*b++)if(*b==34)*y++=92;for(a++;*a;)*y++=*a++;*y=0;for(a=Y;*a;a++)p[*a][*(a+1)]++;c=*Y;do{putchar(c);for(r=j=0;j<256;j++)r+=p[c][j];r=rand()%r;for(j=0;j<256;j++){r-=p[c][j];if(r<0)break;}c=j;}while(c);}";main(){char*a=X,*b=X,*Y=X+315,*y;for(y=Y;*a!=64;)*y++=*a++;for(;*b;*y++=*b++)if(*b==34)*y++=92;for(a++;*a;)*y++=*a++;*y=0;for(a=Y;*a;a++)p[*a][*(a+1)]++;c=*Y;do{putchar(c);for(r=j=0;j<256;j++)r+=p[c][j];r=rand()%r;for(j=0;j<256;j++){r-=p[c][j];if(r<0)break;}c=j;}while(c);}

With newlines and whitespace added for legibility:

	01	j,c,p[256][256]={0},r;
	02	char X[999]="j,c,p[256][256]={0},r;char X[999]=\"@\";main(){char*a=X,*b=X,*Y=X+315,*y;for(y=Y;*a!=64;)*y++=*a++;for(;*b;*y++=*b++)if(*b==34)*y++=92;for(a++;*a;)*y++=*a++;*y=0;for(a=Y;*a;a++)p[*a][*(a+1)]++;c=*Y;do{putchar(c);for(r=j=0;j<256;j++)r+=p[c][j];r=rand()%r;for(j=0;j<256;j++){r-=p[c][j];if(r<0)break;}c=j;}while(c);}";
	03	main(){
	04		char*a=X,*b=X,*Y=X+315,*y;
	05		for(y=Y;*a!=64;)*y++=*a++;
	06		for(;*b;*y++=*b++)if(*b==34)*y++=92;
	07		for(a++;*a;)*y++=*a++;*y=0;
	08		for(a=Y;*a;a++)p[*a][*(a+1)]++;
	09		c=*Y;
	10		do{ putchar(c);
	11			for(r=j=0;j<256;j++)r+=p[c][j];
	12			r=rand()%r;
	13			for(j=0;j<256;j++){
	14				r-=p[c][j];
	15				if(r<0)break;
	16			} c=j;
	17		}while(c);
	18	}

The variable `p[][]` will contain the occurrences of each character following another. `X` contains the entire source, with `'@'` substituted for the value of `X`, in quotes. The buffer `Y` will point start at `X[315]`, one character past the end of the initial string, and will use up 624 bytes. (315 + 624 < 999.)

The for-loops on lines `05`, `06`, and `07` replace the `'@'` in `X` with the content of `X`, escaping double quotes, and stores it in `Y`.

Line `08` tallies the occurrences of one character following another. And line `09` starts the Markov model at the first character of the program (`'j'`).

The do/while loop from `10` to `17` first outputs the current state of the Markov model, then finds the next state using a random number within the range of all successive occurrences.

Sample output:

`j<256;r(;)*Y=0}";j;}whilea=Y;a++31)r;a++=92;dor(a++1564;f(*b=X+;)*b=Y=j+;j;char(r=";*Y=0;*a+;for*a;*a;}chifor*a++)]++156;)]=0)%r(ak;j;j]++){0}c);}whar<0},cha=p[c]=X,*y;}";}c][j<256][j<256;*aifo{c)%r(chin(r X,r=r*b++++=j][*br+=0;forak;*b=X++=0;*Y;for(y={r X[*y;*a;for(;for*y+315,r X,cha+;j=0},*Y;);for(a=0;r++;*y=0}c][*ar=0;}`

Edit: Cut down 34 bytes to go from 707 to 673 bytes.

Edit: Cut down 50 bytes to go from 673 to 623 bytes.