¿Cómo podemos escribir una función eficiente que produzca "equivalentes de homoglifos"de una cadena de entrada?

Ejemplo 1 (pseudocódigo):

homoglyphs_list = [
                     ["o", "0"], // "o" and "0" are homoglyphs
                     ["i", "l", "1"] // "i" and "l" and "1" are homoglyphs
                  ]

input_string    = "someinput"
output          = [
                   "someinput", "s0meinput", "somelnput",
                   "s0melnput", "some1nput", "s0me1nput"
                  ]

Ejemplo 2:

homoglyphs_list = [
                     ["rn", "m", "nn"],
                  ]

input_string    = "rnn"
output          = ["rnn", "rm", "mn", "rrn", "nnn", "nm", "nrn"]

Ejemplo 3:

homoglyphs_list = [
                     ["d", "ci", "a"], // "o" and "0" are homoglyphs
                     ["i", "l", "1"] // "i" and "l" and "1" are homoglyphs
                  ]
                  /*
                     notice that with the two rules above,
                     we can infer "d" = "ci" = "a" = "cl" = "c1"
                  */

input_string    = "pacerier"
output          = [
                   "pacerier",  "pacerler",  "pacer1er",  "pcicerier",
                   "pcicerler", "pcicer1er", "pclcerier", "pc1cerier",
                   "pclcerler", "pc1cerler", "pclcer1er", "pc1cer1er",
                   "pdcerier",  "pdcerler",  "pdcer1er"
                  ]

Nota: el orden de los miembros en la matriz de salida no es importante, y podemos suponer que se asume que las asignaciones de homoglyph dadas sonapropiado (Las entradas no nos darían un "bucle infinito").

Mi algoritmo actual funciona, pero el uso de fuerza bruta en bruto y el rendimiento es horrible. P.ej. una entrada de"mmmmm" con homoglifos["rn", "m", "nn"] tarda 38 segundos en ejecutarse:

// This is php code (non-pseudo just so we could test the running time), 
// but the question remains language-agnostic

public function Func($in, Array $mappings){
    $out_table = array();
    $out_table[$in] = null;
    while(true){
        $number_of_entries_so_far = count($out_table);
        foreach(array_keys($out_table) as $key){
            foreach($mappings as $mapping){
                foreach($mapping as $value){
                    for($a=0, $aa=strlen($key); $a<$aa; ++$a){
                        $pos = strpos($key, $value, $a);
                        if($pos === false){
                            continue;
                        }
                        foreach($mapping as $equal_value){
                            if($value === $equal_value){
                                continue;
                            }
                            $out_table[substr($key, 0, $pos) . $equal_value . substr($key, $pos + strlen($value))] = null;
                        }
                    }

                }
            }
        }
        if($number_of_entries_so_far === count($out_table)){
            // it means we have tried bruteforcing but added no additional entries,
            // so we can quit now
            break;
        }
    }
    return array_keys($out_table);
}

¿Cómo podemos implementar un algoritmo de expansión de homoglyph eficiente (rápido)?