diff --git a/parser/lstm-parse.cc b/parser/lstm-parse.cc
index adde5c9..e6d2f69 100644
--- a/parser/lstm-parse.cc
+++ b/parser/lstm-parse.cc
@@ -71,6 +71,7 @@ void InitCommandLine(int argc, char** argv, po::variables_map* conf) {
         ("lstm_input_dim", po::value<unsigned>()->default_value(60), "LSTM input dimension")
         ("train,t", "Should training be run?")
         ("maxit,M", po::value<unsigned>()->default_value(8000), "Maximum number of training iterations")
+        ("tolerance", po::value<double>()->default_value(0.0), "Tolerance on dev uas for stopping training")
         ("words,w", po::value<string>(), "Pretrained word embeddings")
         ("help,h", "Help");
   po::options_description dcmdline_options;
@@ -525,6 +526,8 @@ int main(int argc, char** argv) {
   assert(unk_prob >= 0.); assert(unk_prob <= 1.);
   const unsigned maxit = conf["maxit"].as<unsigned>();
   cerr << "Maximum number of iterations: " << maxit << "\n";
+  const double tolerance = conf["tolerance"].as<double>();
+  cerr << "Optimization tolerance: " << tolerance << "\n";
   ostringstream os;
   os << "parser_" << (USE_POS ? "pos" : "nopos")
      << '_' << LAYERS
@@ -607,9 +610,12 @@ int main(int argc, char** argv) {
     double llh = 0;
     bool first = true;
     unsigned iter = 0;
+    double uas = -1;
+    double prev_uas = -1;
     time_t time_start = std::chrono::system_clock::to_time_t(std::chrono::system_clock::now());
     cerr << "TRAINING STARTED AT: " << put_time(localtime(&time_start), "%c %Z") << endl;
-    while(!requested_stop && iter < maxit) {
+    while(!requested_stop && iter < maxit &&
+        (uas < 0 || prev_uas < 0 || abs(prev_uas - uas) > tolerance)) {
       for (unsigned sii = 0; sii < status_every_i_iterations; ++sii) {
            if (si == corpus.nsentences) {
              si = 0;
@@ -675,7 +681,9 @@ int main(int argc, char** argv) {
            total_heads += sentence.size() - 1;
         }
         auto t_end = std::chrono::high_resolution_clock::now();
-        cerr << "  **dev (iter=" << iter << " epoch=" << (tot_seen / corpus.nsentences) << ")\tllh=" << llh << " ppl: " << exp(llh / trs) << " err: " << (trs - right) / trs << " uas: " << (correct_heads / total_heads) << "\t[" << dev_size << " sents in " << std::chrono::duration<double, std::milli>(t_end-t_start).count() << " ms]" << endl;
+        prev_uas = uas;
+        uas = correct_heads / total_heads;
+        cerr << "  **dev (iter=" << iter << " epoch=" << (tot_seen / corpus.nsentences) << ")\tllh=" << llh << " ppl: " << exp(llh / trs) << " err: " << (trs - right) / trs << " uas: " << uas << "\t[" << dev_size << " sents in " << std::chrono::duration<double, std::milli>(t_end-t_start).count() << " ms]" << endl;
         if (correct_heads > best_correct_heads) {
           best_correct_heads = correct_heads;
           ofstream out(fname);
@@ -698,6 +706,8 @@ int main(int argc, char** argv) {
     }
     if (iter >= maxit) {
       cerr << "\nMaximum number of iterations reached (" << iter << "), terminating optimization...\n";
+    } else if (!requested_stop) {
+      cerr << "\nScore tolerance reached (" << tolerance << "), terminating optimization...\n";
     }
   } // should do training?
   if (true) { // do test evaluation