Our first thought is to try all possible paths that Bessie can take. For small $N$, this works out well, but it turns out that for a grid of size $N$, the number of different paths is $\frac{(2N-2)!}{(N-1)!^2}$. When $N$ is 18, this quantity is way too big.

However, we are looking for palindromes, namely, strings where the first half of the string is the reverse of the second half of the string.

If we look at the grid more carefully, we note that every palindrome has its middle character on the diagonal starting at the top-right corner of the grid and ending at the bottom-left corner of the grid.

Therefore, for a given square on that diagonal, we can keep track of all possible strings that we can generate going from the top-left corner to that square going only right and down, as well as all strings that we can generate from the bottom-right corner going to that square going only up and left. If, for a given square, a string appears in both of those sets, then that string is a prefix to a valid palindrome.

The total number of paths from a corner to the diagonal is $2^{N-1}$, when $N=18$, this is $131072$. We therefore generate only $262144$ strings of length $17$ in the worst case, which is fast enough.

Here is my Java code simulating this process:

import java.io.*; import java.util.*; public class palpathB { static int n; public static void main(String[] args) throws IOException { BufferedReader br = new BufferedReader(new FileReader("palpath.in")); PrintWriter pw = new PrintWriter(new BufferedWriter(new FileWriter("palpath.out"))); n = Integer.parseInt(br.readLine()); char[][] grid = new char[n][n]; // rows1[a] stores all strings from the top-left corner to // the square on the diagonal at row a rows1 = new HashSet[n]; // rows2[a] stores all strings from the bottom-right corner to // the square on the diagonal at row a rows2 = new HashSet[n]; for(int a = 0; a < n; a++) { rows1[a] = new HashSet<String>(); rows2[a] = new HashSet<String>(); } for(int i = 0; i < n; i++) { String s = br.readLine(); for(int j = 0; j < n; j++) { grid[i][j] = s.charAt(j); } } // by rotating the grid twice, I could reuse the dfs function // instead of having to write a second one that goes up and left dfs(grid, 0, 0, rows1, ""); transpose(grid); dfs(grid, 0, 0, rows2, ""); Set<String> ans = new HashSet<String>(); for(int a = 0; a < n; a++) { for(String s: rows1[a]) { // check if a string can be generated from both corners if(rows2[a].contains(s)) { ans.add(s); } } } pw.println(ans.size()); pw.close(); } public static void dfs(char[][] grid, int x, int y, Set<String>[] sets, String curr) { if(x + y == n-1) { sets[x].add(curr + grid[x][y]); } else { dfs(grid, x+1, y, sets, curr + grid[x][y]); dfs(grid, x, y+1, sets, curr + grid[x][y]); } } // this makes column n of the grid row (n+1-i) public static void transpose(char[][] grid) { for(int i = 0; i < n; i++) { for(int j = 0; j < n; j++) { if(i + j >= n-1) continue; char t = grid[i][j]; grid[i][j] = grid[n-1-j][n-1-i]; grid[n-1-j][n-1-i] = t; } } } static Set<String>[] rows1, rows2; }