diff --git a/chrome/content/zotero/xpcom/rdf/serialize.js b/chrome/content/zotero/xpcom/rdf/serialize.js
index 0de4b2859..267662750 100644
--- a/chrome/content/zotero/xpcom/rdf/serialize.js
+++ b/chrome/content/zotero/xpcom/rdf/serialize.js
@@ -580,9 +580,35 @@ __Serializer.prototype.statementsToXML = function(sts) {
     	var type = null;
     	
         var results = []
-        Zotero.debug(sz.toStr(subject));
         var sts = subjects[sz.toStr(subject)]; // relevant statements
-        sts.sort();
+		// Sort only on the predicate, leave the order at object
+		// level undisturbed.  This leaves multilingual content in
+		// the order of entry (for partner literals), which helps
+		// readability.
+		//
+		// For the predicate sort, we attempt to split the uri
+		// as a hint to the sequence, as sequenced items seems
+		// to be of the form http://example.com#_1, http://example.com#_2,
+		// et cetera.  Probably not the most optimal of fixes, but
+		// it does work.
+        sts.sort(function(a,b){
+			 var aa = a.predicate.uri.split('#_');
+			 var bb = a.predicate.uri.split('#_');
+			 if (aa[0] > bb[0]) {
+				 return 1;
+			 } else if (aa[0] < bb[0]) {
+				 return -1;
+			 } else {
+				 if ("undefined" !== typeof aa[1] && "undefined" !== typeof bb[1]) {
+					 if (parseInt(aa[1], 10) > parseInt(bb[1], 10)) {
+						 return 1;
+					 } else if (parseInt(aa[1], 10) < parseInt(bb[1], 10)) {
+						 return -1;
+					 }
+				 }
+			 }
+			 return 0;
+		});
         for (var i=0; i<sts.length; i++) {
             var st = sts[i];