diff --git a/OpusCatMTEngine/App.xaml.cs b/OpusCatMTEngine/App.xaml.cs
index 4a95714..71c53cf 100644
--- a/OpusCatMTEngine/App.xaml.cs
+++ b/OpusCatMTEngine/App.xaml.cs
@@ -181,11 +181,19 @@ private async void CheckForUpdatesAsync()
 
         private void InitializePythonEngine()
         {
-            Environment.SetEnvironmentVariable("PYTHONNET_PYDLL", ".\\python-3.8.10-embed-amd64\\python38.dll");
-            Environment.SetEnvironmentVariable("PATH", ".\\python-3.8.10-embed-amd64");
-            Environment.SetEnvironmentVariable("PYTHONPATH", ".\\python-3.8.10-embed-amd64");
+            Environment.SetEnvironmentVariable("PYTHONNET_PYDLL", ".\\python-3.8.10-embed-amd64\\python38.dll;");
+            Environment.SetEnvironmentVariable("PATH", ".\\python-3.8.10-embed-amd64;");
+            Environment.SetEnvironmentVariable("PYTHONPATH", ".\\python-3.8.10-embed-amd64;");
+            Environment.SetEnvironmentVariable("PYTHONHOME", ".\\python-3.8.10-embed-amd64;");
+
+            var home = PythonEngine.PythonHome;
+
             PythonEngine.Initialize();
             PythonEngine.BeginAllowThreads();
+            using (Py.GIL())
+            {
+                PythonEngine.ImportModule("sacremoses");
+            }
         }
 
         /// <summary>
diff --git a/OpusCatMTEngine/MTModel.cs b/OpusCatMTEngine/MTModel.cs
index aef049c..5e8d785 100644
--- a/OpusCatMTEngine/MTModel.cs
+++ b/OpusCatMTEngine/MTModel.cs
@@ -24,7 +24,9 @@
 
 namespace OpusCatMTEngine
 {
+
     
+
     public enum MTModelStatus
     {
         OK,
@@ -66,6 +68,30 @@ public enum TagMethod
     
     public class MTModel : INotifyPropertyChanged
     {
+        private dynamic lemmatizer;
+
+        internal List<Tuple<int, int, string>> Lemmatize(string input)
+        {
+            List<Tuple<int, int, string>> lemmaList = new List<Tuple<int, int, string>>();
+            using (Py.GIL())
+            {
+                var output = new List<Tuple<int, int, string>>();
+
+                dynamic processed = this.lemmatizer(input);
+
+                foreach (var sentence in processed.sentences)
+                {
+                    foreach (var word in sentence.words)
+                    {
+                        output.Add(new Tuple<int, int, string>((int)word.start_char, (int)word.end_char, (string)word.lemma));
+                    }
+                }
+
+                return output;
+            }
+
+        }
+
         public object this[string propertyName]
         {
             get
@@ -168,6 +194,124 @@ internal void Shutdown()
             this.marianProcesses = null;
         }
 
+        private Dictionary<int, List<TermMatch>> GetTermMatches(string input)
+        {
+            var termMatches = new Dictionary<int, List<TermMatch>>();
+
+            foreach (var term in this.Terminology.Terms)
+            {
+                var thisTermMatches = term.SourcePatternRegex.Matches(input);
+                foreach (Match termMatch in thisTermMatches)
+                {
+                    if (termMatches.ContainsKey(termMatch.Index))
+                    {
+                        termMatches[termMatch.Index].Add(
+                            new TermMatch(term, termMatch));
+                    }
+                    else
+                    {
+                        termMatches[termMatch.Index] = new List<TermMatch>() {
+                            new TermMatch(term, termMatch)};
+                    }
+                }
+                
+            }
+
+            return termMatches;
+        }
+
+
+        private Dictionary<int,List<TermMatch>> GetLemmaMatches(string input, Dictionary<int, List<TermMatch>> termMatches = null)
+        {
+            if (termMatches == null)
+            {
+                termMatches = new Dictionary<int, List<TermMatch>>();
+            }
+            
+
+            //Get lemmatized input and find lemmatized term matches. Prioritize normal term matches
+            //in case of overlap
+            //var lemmatizedInput = PythonNetHelper.Lemmatize(this.sourceLanguages.First().ShortestIsoCode, input);
+            var lemmatizedInput = this.Lemmatize(input);
+
+            var lemmaToPositionDict = new Dictionary<string, List<int>>();
+            int lemmaCounter = 0;
+            foreach (var lemma in lemmatizedInput.Select(x => x.Item3))
+            {
+                if (lemmaToPositionDict.ContainsKey(lemma))
+                {
+                    lemmaToPositionDict[lemma].Add(lemmaCounter);
+                }
+                else
+                {
+                    lemmaToPositionDict[lemma] = new List<int>() { lemmaCounter };
+                }
+                lemmaCounter++;
+            }
+
+            foreach (var term in this.Terminology.Terms.Where(x => x.MatchSourceLemma))
+            {
+
+                if (term.SourceLemmas == null)
+                {
+                    term.SourceLemmas = this.Lemmatize(term.SourcePattern).Select(x => x.Item3).ToList();
+                }
+                var sourceLemmas = term.SourceLemmas;
+
+                //Check if first lemma in term found in sentence
+
+                bool termLemmaFound;
+
+                if (lemmaToPositionDict.ContainsKey(sourceLemmas[0]))
+                {
+                    var firstLemmaPositions = lemmaToPositionDict[sourceLemmas[0]];
+
+                    //Then check if the other lemmas of the term follow in the source sentence
+                    foreach (var startPos in firstLemmaPositions)
+                    {
+                        int sourceSentencePos = startPos;
+                        termLemmaFound = true;
+                        //start looking at second lemma
+                        for (var termLemmaIndex = 1; termLemmaIndex < sourceLemmas.Count; termLemmaIndex++)
+                        {
+                            sourceSentencePos = startPos + termLemmaIndex;
+                            if (sourceSentencePos < lemmatizedInput.Count)
+                            {
+                                if (lemmatizedInput[sourceSentencePos].Item3 != sourceLemmas[termLemmaIndex])
+                                {
+                                    termLemmaFound = false;
+                                    break;
+                                }
+                            }
+                            else
+                            {
+                                termLemmaFound = false;
+                                break;
+                            }
+                        }
+
+                        if (termLemmaFound)
+                        {
+                            var startChar = lemmatizedInput[startPos].Item1;
+                            var endChar = lemmatizedInput[sourceSentencePos].Item2;
+                            if (termMatches.ContainsKey(startChar))
+                            {
+                                termMatches[startChar].Add(
+                                    new TermMatch(term, startChar, endChar - startChar, true));
+                            }
+                            else
+                            {
+                                termMatches[startChar] = new List<TermMatch>() {
+                                            new TermMatch(term, startChar, endChar-startChar, true)};
+                            }
+                        }
+                    }
+                }
+            }
+
+            return termMatches;
+        }
+
         public Task<TranslationPair> Translate(
             string input, 
             IsoLanguage sourceLang, 
@@ -197,64 +341,12 @@ public Task<TranslationPair> Translate(
                 //Use a simple method of removing overlapping matches of different terms:
                 //For each position record only the longest term match, then when annotating term data,
                 //start from the term closest to edge and skip overlapping terms.
-                var termMatches = new Dictionary<int, List<TermMatch>>();
+                var termMatches = this.GetTermMatches(input);
 
-                //Get lemmatized input and find lemmatized term matches. Prioritize normal term matches
-                //in case of overlap
-                var lemmatizedInput = PythonNetHelper.Lemmatize(this.sourceLanguages.First().ShortestIsoCode, input);
-
-                //Make dicts out of 
-                var lemmaToPositionDict = new Dictionary<string, List<int>>();
-                int lemmaCounter = 0;
-                foreach (var lemma in lemmatizedInput.Select(x => x.Item3))
+                //Match term at lemma level, if specified
+                if (this.Terminology.Terms.Any(x => x.MatchSourceLemma))
                 {
-                    if (lemmaToPositionDict.ContainsKey(lemma))
-                    {
-                        lemmaToPositionDict[lemma].Add(lemmaCounter);
-                    }
-                    else
-                    {
-                        lemmaToPositionDict[lemma] = new List<int>() { lemmaCounter };
-                    }
-                    lemmaCounter++;
-                }
-
-                foreach (var term in this.Terminology.Terms)
-                {
-                    var thisTermMatches = term.SourcePatternRegex.Matches(input);
-                    foreach (Match termMatch in thisTermMatches)
-                    {
-                        
-                        if (termMatches.ContainsKey(termMatch.Index))
-                        {
-                            termMatches[termMatch.Index].Add(
-                                new TermMatch(term,termMatch));
-                        }
-                        else
-                        {
-                            termMatches[termMatch.Index] = new List<TermMatch>() {
-                            new TermMatch(term,termMatch)};
-                        }
-                    }
-
-                    //Match term at lemma level, if specified
-                    if (term.MatchSourceLemma)
-                    {
-                        var sourceLemma = term.SourceLemmas;
-
-                        //Check if first lemma in term found in sentence
-                        if (lemmaToPositionDict.ContainsKey(sourceLemma[0]))
-                        {
-                            var firstLemmaPositions = lemmaToPositionDict[sourceLemma[0]];
-
-                            //Then check if the other lemmas of the term follow in the source sentence
-                            foreach (var startPos in firstLemmaPositions)
-                            {
-
-                            }
-                        }
-
-                    }
+                    termMatches = this.GetLemmaMatches(input, termMatches);
                 }
 
                 int lastEditStart = input.Length;
@@ -768,6 +860,14 @@ public MTModel(
                     this.ModelConfig.TerminologyGuid = this.Terminology.TerminologyGuid;
                     this.SaveModelConfig();
                 }
+
+                using (Py.GIL())
+                {
+                    dynamic stanza = PythonEngine.ImportModule("stanza");
+                    this.lemmatizer = stanza.Pipeline(
+                            this.SourceLanguages[0].ShortestIsoCode, processors: "tokenize,mwt,pos,lemma");
+                }
+
             }
             
             this.ModelConfig.ModelTags.CollectionChanged += ModelTags_CollectionChanged;
diff --git a/OpusCatMTEngine/OpusCatMTEngine.csproj.user b/OpusCatMTEngine/OpusCatMTEngine.csproj.user
index 6cbe588..0bc87c2 100644
--- a/OpusCatMTEngine/OpusCatMTEngine.csproj.user
+++ b/OpusCatMTEngine/OpusCatMTEngine.csproj.user
@@ -3,4 +3,7 @@
   <PropertyGroup>
     <ProjectView>ProjectFiles</ProjectView>
   </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)' == 'Debug|x64'">
+    <EnableUnmanagedDebugging>true</EnableUnmanagedDebugging>
+  </PropertyGroup>
 </Project>
\ No newline at end of file
diff --git a/OpusCatMTEngine/PythonNetHelper.cs b/OpusCatMTEngine/PythonNetHelper.cs
index 4be53e3..690e8d1 100644
--- a/OpusCatMTEngine/PythonNetHelper.cs
+++ b/OpusCatMTEngine/PythonNetHelper.cs
@@ -10,37 +10,31 @@ namespace OpusCatMTEngine
     static class PythonNetHelper
     {
         private static Dictionary<string, dynamic> LemmatizerScopes = new Dictionary<string, dynamic>();
-
+        
         internal static List<Tuple<int,int,string>> Lemmatize(string lang, string input)
         {
             List<Tuple<int, int, string>> lemmaList = new List<Tuple<int, int, string>>();
             using (Py.GIL())
             {
-                if (!PythonNetHelper.LemmatizerScopes.ContainsKey(lang))
+                var output = new List<Tuple<int, int, string>>();
+                if (!LemmatizerScopes.ContainsKey(lang))
                 {
-                    //Initialize the lemmatizer
-                    using (var moduleScope = Py.CreateScope())
-                    {
-                        moduleScope.Exec(OpusCatMTEngine.Properties.Resources.StanzaWrapperCode);
-                        // create a Python scope
-                        using (PyScope scope = Py.CreateScope())
-                        {
-                            scope.Import(moduleScope, "stanza_wrapper");
-
-                            PythonNetHelper.LemmatizerScopes[lang] = scope.Eval<dynamic>(
-                                $"stanza_wrapper.StanzaWrapper('{lang}', processors='tokenize, pos, lemma, depparse')");
-                        }
-                    }
+                    dynamic stanza = Py.Import("stanza");
+                    LemmatizerScopes[lang] = stanza.Pipeline(lang, processors: "tokenize, pos, lemma, depparse");
                 }
 
-                var lemmatized = PythonNetHelper.LemmatizerScopes[lang].lemmatize(input);
-                var output = new List<Tuple<int, int, string>>();
-                foreach (var lemma in lemmatized)
+                dynamic processed = LemmatizerScopes[lang](input);
+                foreach (var sentence in processed.sentences)
                 {
-                    output.Add(new Tuple<int,int,string>((int)lemma[0],(int)lemma[1],(string)lemma[2]));
+                    foreach (var word in sentence.words)
+                    {
+                        output.Add(new Tuple<int, int, string>((int)word.start_char, (int)word.end_char, (string)word.lemma));
+                    }
                 }
+                
                 return output;
             }
+            
         }
     }
 }
diff --git a/OpusCatMTEngine/Terminology/Term.cs b/OpusCatMTEngine/Terminology/Term.cs
index b71e02d..9bbf076 100644
--- a/OpusCatMTEngine/Terminology/Term.cs
+++ b/OpusCatMTEngine/Terminology/Term.cs
@@ -53,18 +53,19 @@ public bool MatchSourceLemma
             }
         }
 
+        [YamlIgnore]
         public List<string> SourceLemmas
         {
             get
             {
-                if (_sourceLemmas == null)
+                if (_sourceLemmas == null || _sourceLemmas.Count == 0)
                 {
-                    if (this.SourceLanguageCode != null)
+                    /*if (this.SourceLanguageCode != null)
                     {
                         _sourceLemmas = PythonNetHelper.Lemmatize(
                             this.SourceLanguageCode,
                             this.SourcePattern).Select(x => x.Item3).ToList();
-                    }
+                    }*/
                 }
                 return _sourceLemmas;
             }
diff --git a/OpusCatMTEngine/Terminology/TermMatch.cs b/OpusCatMTEngine/Terminology/TermMatch.cs
index 89df545..35f6c4b 100644
--- a/OpusCatMTEngine/Terminology/TermMatch.cs
+++ b/OpusCatMTEngine/Terminology/TermMatch.cs
@@ -11,6 +11,9 @@ public class TermMatch
         public TermMatch(Term term, Match termMatch)
         {
             this.Term = term;
+            this.Start = termMatch.Index;
+            this.Length = termMatch.Length;
+            this.LemmaMatch = false;
 
         }
 
diff --git a/OpusCatMTEngine/UI/TerminologyView.xaml b/OpusCatMTEngine/UI/TerminologyView.xaml
index 1a8b24b..f9efb47 100644
--- a/OpusCatMTEngine/UI/TerminologyView.xaml
+++ b/OpusCatMTEngine/UI/TerminologyView.xaml
@@ -38,6 +38,7 @@
                                 <DataGridTextColumn Header="Target lemma" Binding="{Binding TargetLemma}" />
                                 <DataGridCheckBoxColumn Header="Regex?" Binding="{Binding SourcePatternIsRegex}" />
                                 <DataGridCheckBoxColumn Header="Case sensitive?" Binding="{Binding SourcePatternIsCaseSensitive}" />
+                                <DataGridCheckBoxColumn Header="Match lemmas?" Binding="{Binding MatchSourceLemma}" />
                             </DataGrid.Columns>
                         </DataGrid>
                     </DockPanel>
diff --git a/OpusCatMTEngine/UI/TerminologyView.xaml.cs b/OpusCatMTEngine/UI/TerminologyView.xaml.cs
index f87e86a..6f55b3c 100644
--- a/OpusCatMTEngine/UI/TerminologyView.xaml.cs
+++ b/OpusCatMTEngine/UI/TerminologyView.xaml.cs
@@ -49,11 +49,18 @@ public TerminologyView(
             this.Terminologies = terminologies;
             
             InitializeComponent();
+            this.TermList.InitializingNewItem += InitializeNewTerm;
             this.TermList.ItemsSource = this.Model.Terminology.Terms;
             
         }
 
-        
+        private void InitializeNewTerm(object sender, InitializingNewItemEventArgs e)
+        {
+            var term = (Term)e.NewItem;
+            term.SourceLanguageCode = this.model.SourceLanguages.First().ShortestIsoCode;
+            term.TargetLanguageCode = this.model.TargetLanguages.First().ShortestIsoCode;
+        }
+
         public MTModel Model { get => model; set => model = value; }
         public string Title { get; private set; }
         public ObservableCollection<Terminology> Terminologies { get; private set; }
diff --git a/Trados2019Plugin/FinetuneTransUnitExtractor.cs b/Trados2019Plugin/FinetuneTransUnitExtractor.cs
index e979ce5..a3324de 100644
--- a/Trados2019Plugin/FinetuneTransUnitExtractor.cs
+++ b/Trados2019Plugin/FinetuneTransUnitExtractor.cs
@@ -4,6 +4,7 @@
 using Sdl.LanguagePlatform.TranslationMemoryApi;
 using System;
 using System.Collections.Generic;
+using System.Globalization;
 using System.IO;
 using System.Linq;
 using System.Reflection;
@@ -143,7 +144,11 @@ public FinetuneTransUnitExtractor(
             int maxConcordanceWindow)
         {
             this.tms = tms;
+#if (TRADOS22)
+            this.sourceLanguage = new CultureInfo(tms.First().SourceLanguage.Name).TwoLetterISOLanguageName;
+#else
             this.sourceLanguage = tms.First().SourceLanguage.TwoLetterISOLanguageName;
+#endif
             //Shuffle the source segment to prevent focusing on the initial part of the job (in case
             //units needed value is reached before whole source has been processed).
             this.sourceSegments = sourceSegments.ToList();
diff --git a/Trados2019Plugin/OpusCatOptionControl.xaml.cs b/Trados2019Plugin/OpusCatOptionControl.xaml.cs
index 40bc851..c1eab2a 100644
--- a/Trados2019Plugin/OpusCatOptionControl.xaml.cs
+++ b/Trados2019Plugin/OpusCatOptionControl.xaml.cs
@@ -6,6 +6,7 @@
 using System.Collections.ObjectModel;
 using System.ComponentModel;
 using System.Diagnostics;
+using System.Globalization;
 using System.Linq;
 using System.Runtime.CompilerServices;
 using System.ServiceModel;
@@ -38,8 +39,15 @@ public OpusCatOptionControl(OpusCatOptionsFormWPF hostForm,
             this.DataContext = this;
             this.CredentialStore = credentialStore;
             this.Options = options;
+
+#if (TRADOS22)
+            this.projectLanguagePairs = languagePairs.Select(
+                x => $"{new CultureInfo(x.SourceCulture.Name).TwoLetterISOLanguageName}-" +
+                $"{new CultureInfo(x.TargetCulture.Name).TwoLetterISOLanguageName}").ToList();
+#else
             this.projectLanguagePairs = languagePairs.Select(
                 x => $"{x.SourceCulture.TwoLetterISOLanguageName}-{x.TargetCulture.TwoLetterISOLanguageName}").ToList();
+#endif
 
             InitializeComponent();
             this.ConnectionSelection.LanguagePairs = this.projectLanguagePairs;
diff --git a/Trados2019Plugin/OpusCatProviderLanguageDirection.cs b/Trados2019Plugin/OpusCatProviderLanguageDirection.cs
index acc8819..28dc2fb 100644
--- a/Trados2019Plugin/OpusCatProviderLanguageDirection.cs
+++ b/Trados2019Plugin/OpusCatProviderLanguageDirection.cs
@@ -14,6 +14,8 @@
 using System.Text.RegularExpressions;
 using System.Windows.Controls;
 using System.Runtime.InteropServices;
+using System.Globalization;
+using Sdl.Core.Globalization;
 
 namespace OpusCatTranslationProvider
 {
@@ -48,11 +50,17 @@ public OpusCatProviderLanguageDirection(OpusCatProvider provider, LanguagePair l
 
             _visitor = new OpusCatProviderElementVisitor();
 
+#if (TRADOS22)
+            var sourceCode = new CultureInfo(this._languageDirection.SourceCulture.Name).TwoLetterISOLanguageName;
+            var targetCode = new CultureInfo(this._languageDirection.TargetCulture.Name).TwoLetterISOLanguageName;
+#else
             var sourceCode = this._languageDirection.SourceCulture.TwoLetterISOLanguageName;
             var targetCode = this._languageDirection.TargetCulture.TwoLetterISOLanguageName;
+#endif
+
             this.langpair = $"{sourceCode}-{targetCode}";
 
-            #endregion
+#endregion
         }
 
 
@@ -79,7 +87,7 @@ public ITranslationProvider TranslationProvider
         /// <param name="settings"></param>
         /// <param name="segment"></param>
         /// <returns></returns>
-        #region "SearchSegment"
+#region "SearchSegment"
         public SearchResults SearchSegment(SearchSettings settings, Segment segment)
         {
            
@@ -90,15 +98,21 @@ public SearchResults SearchSegment(SearchSettings settings, Segment segment)
             }
             
 
-            #region "SearchResultsObject"
+#region "SearchResultsObject"
             SearchResults results = new SearchResults();
             results.SourceSegment = segment.Duplicate();
 
-            #endregion
+#endregion
             string sourceText = _visitor.PlainText;
-           
+
+#if (TRADOS22)
+            var sourceCode = new CultureInfo(this._languageDirection.SourceCulture.Name).TwoLetterISOLanguageName;
+            var targetCode = new CultureInfo(this._languageDirection.TargetCulture.Name).TwoLetterISOLanguageName;
+#else
             var sourceCode = this._languageDirection.SourceCulture.TwoLetterISOLanguageName;
             var targetCode = this._languageDirection.TargetCulture.TwoLetterISOLanguageName;
+#endif
+
             var langpair = $"{sourceCode}-{targetCode}";
 
             List<SearchResult> systemResults = this.GenerateSystemResult(sourceText, settings.Mode,segment,sourceCode,targetCode);
@@ -108,7 +122,7 @@ public SearchResults SearchSegment(SearchSettings settings, Segment segment)
             }
  
             return results;
-            #endregion
+#endregion
         }
 
         private List<SearchResult> GenerateSystemResult(
@@ -175,7 +189,7 @@ private List<SearchResult> GenerateSystemResult(
             }
             return systemResults;
         }
-        #endregion
+#endregion
 
 
 
@@ -188,7 +202,7 @@ private List<SearchResult> GenerateSystemResult(
         /// <param name="sourceSegment"></param>
         /// <param name="formattingPenalty"></param>
         /// <returns></returns>
-        #region "CreateSearchResult"
+#region "CreateSearchResult"
         private SearchResult CreateSearchResult(Segment searchSegment, Segment translation,
             bool formattingPenalty,string mtSystem)
         {
@@ -233,7 +247,7 @@ private SearchResult CreateSearchResult(Segment searchSegment, Segment translati
 
             return searchResult;
         }
-        #endregion
+#endregion
 
 
         public bool CanReverseLanguageDirection
@@ -243,6 +257,11 @@ public bool CanReverseLanguageDirection
 
         public static Segment CurrentTranslation { get; private set; }
 
+#if (TRADOS22)
+        CultureCode ITranslationProviderLanguageDirection.SourceLanguage => new CultureCode(this.SourceLanguage);
+        CultureCode ITranslationProviderLanguageDirection.TargetLanguage => new CultureCode(this.TargetLanguage);
+#endif
+
         public SearchResults[] SearchSegments(SearchSettings settings, Segment[] segments)
         {
             SearchResults[] results = new SearchResults[segments.Length];
@@ -325,7 +344,7 @@ public SearchResults[] SearchTranslationUnitsMasked(SearchSettings settings, Tra
         }
 
 
-        #region "NotForThisImplementation"
+#region "NotForThisImplementation"
         /// <summary>
         /// Not required for this implementation.
         /// </summary>
@@ -403,6 +422,6 @@ public ImportResult[] AddOrUpdateTranslationUnits(TranslationUnit[] translationU
         {
             throw new NotImplementedException();
         }
-        #endregion
+#endregion
     }
 }
diff --git a/Trados2019Plugin/TagRestorer.cs b/Trados2019Plugin/TagRestorer.cs
index 0092c71..acd2cc1 100644
--- a/Trados2019Plugin/TagRestorer.cs
+++ b/Trados2019Plugin/TagRestorer.cs
@@ -120,7 +120,7 @@ private void GetInitialPositions()
             {
                 Type elementType = segElement.GetType();
 
-                if (elementType == typeof(Text))
+                if (elementType == typeof(Text) && sourceSubwordQueue.Count > 0)
                 {
                     //Only check for numbers and letters, as they should be identical with both BPE
                     //and SentencePiece subwords.