Sets is_reliable for CLD3 if below a minimum byte threshold of 50 bytes.

extensions/renderer/i18n_custom_bindings.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "extensions/renderer/i18n_custom_bindings.h"
 
 #include <stddef.h>
 #include <stdint.h>
 
 #include <vector>
@@ skipping 23 matching lines @@
 
 namespace extensions {
 
 using namespace v8_helpers;
 
 namespace {
 
 // Max number of languages to detect.
 const int kCldNumLangs = 3;
 
+// CLD3 minimum reliable byte threshold. Predictions for inputs below this size
+// in bytes will be considered unreliable.
+const int kCld3MinimumByteThreshold = 50;
+
 struct DetectedLanguage {
   DetectedLanguage(const std::string& language, int percentage)
       : language(language), percentage(percentage) {}
   ~DetectedLanguage() {}
 
   // Returns a new v8::Local<v8::Value> representing the serialized form of
   // this DetectedLanguage object.
   std::unique_ptr<base::DictionaryValue> ToDictionary() const;
 
   std::string language;
@@ skipping 249 matching lines @@
   }
 
   LanguageDetectionResult result(is_reliable);
   // populate LanguageDetectionResult with languages and percents
   InitDetectedLanguages(languages, percents, &result.languages);
   args.GetReturnValue().Set(result.ToValue(context()));
 
 #elif BUILDFLAG(CLD_VERSION) == 3
   chrome_lang_id::NNetLanguageIdentifier nnet_lang_id(/*min_num_bytes=*/0,
                                                       /*max_num_bytes=*/512);
-  const std::vector<chrome_lang_id::NNetLanguageIdentifier::Result>
-      lang_results = nnet_lang_id.FindTopNMostFreqLangs(text, kCldNumLangs);
+  std::vector<chrome_lang_id::NNetLanguageIdentifier::Result> lang_results =
+      nnet_lang_id.FindTopNMostFreqLangs(text, kCldNumLangs);
+
+  // is_reliable is set to false if we believe the input is too short to be
+  // accurately identified by the current model.
+  //
+  // Note that when is_reliable is false, the TranslateExtension .js code

[Devlin, 2017/04/03 20:46:10]

This is a public API, so we shouldn't include implementation details of a
specific extension (even in comments)

+  // gathers additional surrounding context and will try the prediction once
+  // more.
+  if (text.size() < kCld3MinimumByteThreshold) {
+    for (auto& result : lang_results) {
+      result.is_reliable = false;
+    }
+  }
+
   LanguageDetectionResult result;
 
   // Populate LanguageDetectionResult with prediction reliability, languages,
   // and the corresponding percentages.
   InitDetectedLanguages(lang_results, &result);
   args.GetReturnValue().Set(result.ToValue(context()));
 #else
 # error "CLD_VERSION must be 2 or 3"
 #endif
 }
 
 }  // namespace extensions