diff --git a/gash/lexer.scm b/gash/lexer.scm
index cf17113..4eaa2e2 100644
--- a/gash/lexer.scm
+++ b/gash/lexer.scm
@@ -553,10 +553,11 @@ characters read."
   "Convert @var{proc} from a procedure that returns a token-value pair
 to a procedure that returns a lexical token to be consumed by the LALR
 module."
-  (lambda (port)
+  (lambda (port . rest)
     (let ((port-location (port->port-location port)))
       (receive (length token)
-          (call-with-metered-input-port port proc)
+          (call-with-metered-input-port port (lambda (mp)
+                                               (apply proc mp rest)))
         (match token
           ((category . value)
            (make-lexical-token
@@ -570,7 +571,15 @@ module."
   (get-token->get-lexical-token get-operator))
 
 (define get-word-lexical-token
-  (get-token->get-lexical-token get-word))
+  ;; The is a roundabout way to avoid putting more than one character
+  ;; on the 'unget-char' buffer for regular ports.  Mes only allows
+  ;; one.  Here, we delay adding a character until after we have
+  ;; initialized a soft port, which has an unbounded buffer.
+  (let ((get-word* (lambda* (port #:key start)
+                     (when start
+                       (unget-char port start))
+                     (get-word port))))
+    (get-token->get-lexical-token get-word*)))
 
 (define (get-newline-lexical-token port)
   "Get a newline as a lexical token to be consumed by the LALR module."
@@ -602,8 +611,7 @@ is a newline (or EOF)."
       (#\newline (get-newline-lexical-token port))
       (#\\ (match (next-char port)
              (#\newline (loop (next-char port)))
-             (_ (unget-char port #\\)
-                (get-word-lexical-token port))))
+             (_ (get-word-lexical-token port #:start #\\))))
       (_ (get-word-lexical-token port)))))