From 93c2ab15447a283355f4df681b2cd323fd3e4be3 Mon Sep 17 00:00:00 2001
From: Michael Ringgaard <michael@ringgaard.com>
Date: Tue, 22 Jan 2019 10:35:43 +0100
Subject: [PATCH] Alias transfer (#326)

---
 app/lib/docview.js                            |  45 +-
 data/wiki/calendar.sling                      |   2 +-
 doc/guide/install.md                          |   2 +-
 doc/guide/myelin.md                           |   6 +-
 python/run.py                                 |  73 +--
 python/task/download.py                       |   1 +
 python/task/wiki.py                           |   7 +-
 python/task/workflow.py                       |  42 ++
 sling/frame/object.h                          |   3 +-
 sling/frame/store.cc                          |   5 +-
 sling/frame/store.h                           |  10 +-
 sling/nlp/document/BUILD                      |   1 +
 sling/nlp/document/document.cc                |  49 +-
 sling/nlp/document/document.h                 |  18 +-
 sling/nlp/kb/BUILD                            |  63 ++-
 sling/nlp/kb/app/index.html                   |  15 +-
 sling/nlp/kb/app/kb.css                       |   4 +
 sling/nlp/kb/calendar.cc                      |  37 +-
 sling/nlp/kb/calendar.h                       |  19 +-
 sling/nlp/kb/facts.cc                         |  42 +-
 sling/nlp/kb/facts.h                          |   7 +
 sling/nlp/kb/knowledge-service.cc             |  28 +-
 sling/nlp/kb/knowledge-service.h              |   3 +-
 sling/nlp/{wiki => kb}/name-table-builder.cc  |   0
 sling/nlp/{wiki => kb}/name-table.cc          |   2 +-
 sling/nlp/{wiki => kb}/name-table.h           |   0
 sling/nlp/kb/phrase-table-builder.cc          | 466 ++++++++++++++++++
 sling/nlp/{wiki => kb}/phrase-table.cc        |  75 +--
 sling/nlp/{wiki => kb}/phrase-table.h         |  24 +-
 sling/nlp/wiki/BUILD                          |  64 +--
 .../wiki/{profile-aliases.cc => aliases.cc}   |  54 +-
 sling/nlp/wiki/phrase-table-builder.cc        | 249 ----------
 sling/nlp/wiki/wiki-annotator.cc              |  39 +-
 sling/nlp/wiki/wiki-annotator.h               |   4 +-
 sling/nlp/wiki/wiki-macros.cc                 |  12 +-
 sling/nlp/wiki/wiki.cc                        |   1 +
 sling/nlp/wiki/wiki.h                         |   3 +-
 sling/nlp/wiki/wikidata-converter.cc          |  12 +-
 sling/nlp/wiki/wikidata-converter.h           |   1 +
 sling/nlp/wiki/wikipedia-documents.cc         |   6 -
 sling/pyapi/BUILD                             |   9 +-
 sling/pyapi/pyphrase.h                        |   2 +-
 sling/pyapi/pytask.cc                         |   4 +-
 sling/task/frames.cc                          |  20 +-
 sling/task/frames.h                           |   3 -
 sling/util/unicode.cc                         |   9 +
 sling/util/unicode.h                          |  11 +-
 47 files changed, 1013 insertions(+), 539 deletions(-)
 rename sling/nlp/{wiki => kb}/name-table-builder.cc (100%)
 rename sling/nlp/{wiki => kb}/name-table.cc (98%)
 rename sling/nlp/{wiki => kb}/name-table.h (100%)
 create mode 100644 sling/nlp/kb/phrase-table-builder.cc
 rename sling/nlp/{wiki => kb}/phrase-table.cc (57%)
 rename sling/nlp/{wiki => kb}/phrase-table.h (88%)
 rename sling/nlp/wiki/{profile-aliases.cc => aliases.cc} (85%)
 delete mode 100644 sling/nlp/wiki/phrase-table-builder.cc

diff --git a/app/lib/docview.js b/app/lib/docview.js
index 61f0b7b7..c56ba904 100644
--- a/app/lib/docview.js
+++ b/app/lib/docview.js
@@ -451,7 +451,7 @@ export class DocumentViewer extends Component {
   }
 
   BuildPanel(phrase, fidx) {
-    let mention = this.document.frames[fidx];
+    let frame = this.document.frames[fidx];
     let panel = document.createElement("div");
     panel.className = "panel";
     panel.id = "p" + next_panel++;
@@ -462,9 +462,11 @@ export class DocumentViewer extends Component {
 
     let title = document.createElement("span");
     title.className = "panel-title";
-    title.appendChild(document.createTextNode(phrase));
-    titlebar.appendChild(title);
-    this.AddTypes(titlebar, mention.types);
+    if (phrase) {
+      title.appendChild(document.createTextNode(phrase));
+      titlebar.appendChild(title);
+      this.AddTypes(titlebar, frame.types);
+    }
 
     let icon = document.createElement("span");
     icon.className = "panel-icon";
@@ -476,19 +478,24 @@ export class DocumentViewer extends Component {
     let contents = document.createElement("div");
     contents.className = "panel-content"
 
-    let slots = mention.slots;
-    if (slots) {
-      for (let i = 0; i < slots.length; i += 2) {
-        let n = slots[i];
-        let v = slots[i + 1];
-        if (this.document.frames[n].id == "evokes" ||
-            this.document.frames[n].id == "is") {
-          let avm = this.BuildAVM(v, {});
-          contents.appendChild(avm);
+    if (phrase) {
+      let rendered = {};
+      let slots = frame.slots;
+      if (slots) {
+        for (let i = 0; i < slots.length; i += 2) {
+          let n = slots[i];
+          let v = slots[i + 1];
+          if (this.document.frames[n].id == "evokes" ||
+              this.document.frames[n].id == "is") {
+            let avm = this.BuildAVM(v, rendered);
+            contents.appendChild(avm);
+          }
         }
       }
+    } else {
+      let avm = this.BuildAVM(fidx, {});
+      contents.appendChild(avm);
     }
-
     panel.appendChild(contents);
     return panel;
   }
@@ -504,7 +511,11 @@ export class DocumentViewer extends Component {
     let span = e.currentTarget;
     let phrase = span.getAttribute("phrase");
     let fidx = parseInt(span.getAttribute("frame"));
-    this.AddPanel('"' + phrase + '"', fidx);
+    if (phrase) {
+      this.AddPanel('"' + phrase + '"', fidx);
+    } else {
+      this.AddPanel(null, fidx);
+    }
   }
 
   ClosePanel(e) {
@@ -519,7 +530,6 @@ export class DocumentViewer extends Component {
     chip.className = "chip";
     chip.id = "t" + fidx;
     chip.setAttribute("frame", fidx);
-    chip.setAttribute("phrase", name);
     chip.appendChild(document.createTextNode(name));
 
     return chip;
@@ -548,13 +558,14 @@ export class DocumentViewer extends Component {
 
     let fidx = parseInt(span.getAttribute("frame"))
     let mention = this.document.frames[fidx];
+    let rendered = {};
     let slots = mention.slots;
     if (slots) {
       for (let i = 0; i < slots.length; i += 2) {
         let n = slots[i];
         let v = slots[i + 1];
         if (this.document.frames[n].id == "evokes") {
-          let avm = this.BuildAVM(v, {});
+          let avm = this.BuildAVM(v, rendered);
           callout.appendChild(avm);
         }
       }
diff --git a/data/wiki/calendar.sling b/data/wiki/calendar.sling
index d33bb939..790d93df 100644
--- a/data/wiki/calendar.sling
+++ b/data/wiki/calendar.sling
@@ -211,7 +211,7 @@
     625: Q2662
     626: Q2661
     627: Q2664
-    628: Q2664
+    628: Q2660
     629: Q2659
     630: Q2657
 
diff --git a/doc/guide/install.md b/doc/guide/install.md
index 3586a175..89a2f462 100644
--- a/doc/guide/install.md
+++ b/doc/guide/install.md
@@ -43,7 +43,7 @@ SLING uses [Bazel](https://bazel.build/) as the build system, so you need to
 to build the SLING parser.
 
 ```shell
-sudo apt-get install pkg-config zip g++ zlib1g-dev unzip python
+sudo apt-get install pkg-config zip g++ zlib1g-dev unzip python2.7 python2.7-dev
 wget -P /tmp https://github.com/bazelbuild/bazel/releases/download/0.13.0/bazel-0.13.0-installer-linux-x86_64.sh
 chmod +x /tmp/bazel-0.13.0-installer-linux-x86_64.sh
 sudo /tmp/bazel-0.13.0-installer-linux-x86_64.sh
diff --git a/doc/guide/myelin.md b/doc/guide/myelin.md
index c3e8e593..a382970b 100644
--- a/doc/guide/myelin.md
+++ b/doc/guide/myelin.md
@@ -377,7 +377,7 @@ flow = "flow" <version>
 
 var = <#flags> (IN=1, OUT=2, REF=4, LEARNABLE=8 UNIQUE=16, from version 5)
       <name$>
-      <#aliases> <alias$>
+      <#aliases> <alias$>*
       <dtype$>
       <shape>
       <#bytes> value
@@ -397,11 +397,11 @@ blob = <#flags> (unused, from version 5)
 
 func = <#flags> (TRAINING=1, from version 5)
        <name$>
-       <#ops> <op$>
+       <#ops> <op$>*
 
 cnx = <#flags> (unused, from version 5)
       <name$>
-      <#vars> <var$>
+      <#vars> <var$>*
 
 shape = <#dims> <size>*
 
diff --git a/python/run.py b/python/run.py
index ac5385f8..9323ce2e 100755
--- a/python/run.py
+++ b/python/run.py
@@ -120,37 +120,6 @@
              default=False,
              action='store_true')
 
-flags.define("--dryrun",
-             help="build worflows but do not run them",
-             default=False,
-             action='store_true')
-
-flags.define("--monitor",
-             help="port number for task monitor (0 means no monitor)",
-             default=6767,
-             type=int,
-             metavar="PORT")
-
-flags.define("--logdir",
-             help="directory where workflow logs are stored",
-             default="local/logs",
-             metavar="DIR")
-
-def run_workflow(wf):
-  # In dryrun mode the workflow is just dumped without running it.
-  if flags.arg.dryrun:
-    print wf.wf.dump()
-    return
-
-  # Start workflow.
-  log.info("start workflow")
-  wf.wf.start()
-
-  # Wait until workflow completes. Poll every second to make the workflow
-  # interruptible.
-  done = False
-  while not done: done = wf.wf.wait(1000)
-
 def download_corpora():
   if flags.arg.download_wikidata or flags.arg.download_wikipedia:
     wf = download.DownloadWorkflow("wiki-download")
@@ -164,7 +133,7 @@ def download_corpora():
       for language in flags.arg.languages:
         wf.download_wikipedia(language=language)
 
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
 def import_wiki():
   if flags.arg.import_wikidata or flags.arg.import_wikipedia:
@@ -180,7 +149,7 @@ def import_wiki():
         log.info("Import " + language + " wikipedia")
         wf.wikipedia(language=language)
 
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
 def parse_wikipedia():
   # Convert wikipedia pages to SLING documents.
@@ -189,7 +158,7 @@ def parse_wikipedia():
       log.info("Parse " + language + " wikipedia")
       wf = wiki.WikiWorkflow(language + "-wikipedia-parsing")
       wf.parse_wikipedia(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
 def fuse_items():
   # Merge categories from wikipedias.
@@ -197,28 +166,28 @@ def fuse_items():
     log.info("Merge wikipedia categories")
     wf = wiki.WikiWorkflow("category-merging")
     wf.merge_wikipedia_categories()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Invert categories.
   if flags.arg.invert_categories:
     log.info("Invert categories")
     wf = wiki.WikiWorkflow("category-inversion")
     wf.invert_wikipedia_categories()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Compute item popularity.
   if flags.arg.compute_item_popularity:
     log.info("Compute item popularity")
     wf = wiki.WikiWorkflow("item-popularity")
     wf.compute_item_popularity()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Fuse items.
   if flags.arg.fuse_items:
     log.info("Fuse items")
     wf = wiki.WikiWorkflow("fuse-items")
     wf.fuse_items()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
 
 def build_knowledge_base():
@@ -227,7 +196,7 @@ def build_knowledge_base():
     log.info("Build knowledge base repository")
     wf = wiki.WikiWorkflow("knowledge-base")
     wf.build_knowledge_base()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Extract item names from wikidata and wikipedia.
   if flags.arg.extract_names:
@@ -235,7 +204,7 @@ def build_knowledge_base():
       log.info("Extract " + language + " names")
       wf = wiki.WikiWorkflow(language + "-name-extraction")
       wf.extract_names(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
   # Build name table.
   if flags.arg.build_nametab:
@@ -243,7 +212,7 @@ def build_knowledge_base():
       log.info("Build " + language + " name table")
       wf = wiki.WikiWorkflow(language + "-name-table")
       wf.build_name_table(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
   # Build phrase table.
   if flags.arg.build_phrasetab:
@@ -251,7 +220,7 @@ def build_knowledge_base():
       log.info("Build " + language + " phrase table")
       wf = wiki.WikiWorkflow(language + "-phrase-table")
       wf.build_phrase_table(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
 def train_embeddings():
   # Extract vocabulary for word embeddings.
@@ -260,7 +229,7 @@ def train_embeddings():
       log.info("Extract " + language + " vocabulary")
       wf = embedding.EmbeddingWorkflow(language + "-vocabulary")
       wf.extract_vocabulary(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
   # Train word embeddings.
   if flags.arg.train_word_embeddings:
@@ -268,28 +237,28 @@ def train_embeddings():
       log.info("Train " + language + " word embeddings")
       wf = embedding.EmbeddingWorkflow(language + "-word-embeddings")
       wf.train_word_embeddings(language=language)
-      run_workflow(wf)
+      workflow.run(wf.wf)
 
   # Extract vocabulary for fact and category embeddings.
   if flags.arg.extract_fact_lexicon:
     log.info("Extract fact and category lexicons")
     wf = embedding.EmbeddingWorkflow("fact-lexicon")
     wf.extract_fact_lexicon()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Extract facts from knowledge base.
   if flags.arg.extract_facts:
     log.info("Extract facts from knowledge base")
     wf = embedding.EmbeddingWorkflow("fact-extraction")
     wf.extract_facts()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
   # Train fact and category embeddings.
   if flags.arg.train_fact_embeddings:
     log.info("Train fact and category embeddings")
     wf = embedding.EmbeddingWorkflow("fact-embeddings")
     wf.train_fact_embeddings()
-    run_workflow(wf)
+    workflow.run(wf.wf)
 
 
 if __name__ == '__main__':
@@ -309,20 +278,16 @@ def train_embeddings():
     flags.arg.build_nametab = True
     flags.arg.build_phrasetab = True
 
-  # Start task monitor.
-  if flags.arg.monitor > 0: workflow.start_monitor(flags.arg.monitor)
-
   # Run workflows.
+  workflow.startup()
   download_corpora()
   import_wiki()
   parse_wikipedia()
   fuse_items()
   build_knowledge_base()
   train_embeddings()
-
-  # Stop task monitor.
-  if flags.arg.monitor > 0: workflow.stop_monitor()
-  workflow.save_workflow_log(flags.arg.logdir)
+  workflow.shutdown()
 
   # Done.
   log.info("Done")
+
diff --git a/python/task/download.py b/python/task/download.py
index dc50fcb4..57ac719b 100644
--- a/python/task/download.py
+++ b/python/task/download.py
@@ -16,6 +16,7 @@
 
 import os
 import urllib2
+import _strptime
 import time
 
 from workflow import *
diff --git a/python/task/wiki.py b/python/task/wiki.py
index b4b6d97b..46cdef72 100644
--- a/python/task/wiki.py
+++ b/python/task/wiki.py
@@ -583,7 +583,7 @@ def extract_names(self, aliases=None, language=None):
     if aliases == None:
       # Get language-dependent aliases from Wikidata and Wikpedia.
       wikidata_aliases = self.wf.map(self.fused_items(),
-                                     "profile-alias-extractor",
+                                     "alias-extractor",
                                      params={
                                        "language": language,
                                        "skip_aux": True,
@@ -599,7 +599,7 @@ def extract_names(self, aliases=None, language=None):
     merged_aliases = self.wf.shuffle(aliases, len(names))
 
     # Filter and select aliases.
-    self.wf.reduce(merged_aliases, names, "profile-alias-reducer",
+    self.wf.reduce(merged_aliases, names, "alias-reducer",
                    params={"language": language})
     return names
 
@@ -648,8 +648,11 @@ def build_phrase_table(self, names=None, language=None):
     with self.wf.namespace("phrase-table"):
       builder = self.wf.task("phrase-table-builder")
       builder.add_param("language", language)
+      builder.add_param("transfer_aliases", True)
       self.wf.connect(self.wf.read(names, name="name-reader"), builder)
+      kb = self.knowledge_base()
       repo = self.phrase_table(language)
+      builder.attach_input("commons", kb)
       builder.attach_output("repository", repo)
     return repo
 
diff --git a/python/task/workflow.py b/python/task/workflow.py
index ebf5ae4a..b3408e78 100644
--- a/python/task/workflow.py
+++ b/python/task/workflow.py
@@ -21,8 +21,25 @@
 import time
 import sling
 import sling.pysling as api
+import sling.flags as flags
 import sling.log as log
 
+flags.define("--dryrun",
+             help="build worflows but do not run them",
+             default=False,
+             action='store_true')
+
+flags.define("--monitor",
+             help="port number for task monitor (0 means no monitor)",
+             default=6767,
+             type=int,
+             metavar="PORT")
+
+flags.define("--logdir",
+             help="directory where workflow logs are stored",
+             default="local/logs",
+             metavar="DIR")
+
 # Input readers.
 readers = {
   "records": "record-file-reader",
@@ -705,3 +722,28 @@ def save_workflow_log(path):
   log.info("workflow stats saved in " + logfn)
   return True
 
+def run(wf):
+  # In dryrun mode the workflow is just dumped without running it.
+  if flags.arg.dryrun:
+    print wf.dump()
+    return
+
+  # Start workflow.
+  wf.start()
+
+  # Wait until workflow completes. Poll every second to make the workflow
+  # interruptible.
+  done = False
+  while not done: done = wf.wait(1000)
+
+def startup():
+  # Start task monitor.
+  if flags.arg.monitor > 0: start_monitor(flags.arg.monitor)
+
+def shutdown():
+  # Stop task monitor.
+  if flags.arg.monitor > 0: stop_monitor()
+
+  # Save log to log directory.
+  save_workflow_log(flags.arg.logdir)
+
diff --git a/sling/frame/object.h b/sling/frame/object.h
index f39663a2..6a300cfd 100644
--- a/sling/frame/object.h
+++ b/sling/frame/object.h
@@ -69,7 +69,8 @@ class HandleSpace : public Space<Handle>, public External {
 };
 
 // Hash map and set keyed by handle.
-template<typename T> using HandleMap = std::unordered_map<Handle, T, HandleHash>;
+template<typename T> using HandleMap =
+  std::unordered_map<Handle, T, HandleHash>;
 typedef std::unordered_set<Handle, HandleHash> HandleSet;
 
 // Name with lazy lookup that can be initialized as static variables and
diff --git a/sling/frame/store.cc b/sling/frame/store.cc
index 8c8473a4..97d115a4 100644
--- a/sling/frame/store.cc
+++ b/sling/frame/store.cc
@@ -1088,7 +1088,10 @@ Handle Store::AllocateHandleSlow(Datum *object) {
   DCHECK(free_handle_ == nullptr);
 
   // Expand handle table.
-  handles_.reserve(handles_.size() * 2);
+  if (handles_.size() >= kMaxHandlesSize) LOG(FATAL) << "Handle overflow";
+  size_t newsize = handles_.size() * 2;
+  if (newsize > kMaxHandlesSize) newsize = kMaxHandlesSize;
+  handles_.reserve(newsize);
 
   // Update the pool pointer to handle table.
   pools_[store_tag_] = reinterpret_cast<Address>(handles_.base());
diff --git a/sling/frame/store.h b/sling/frame/store.h
index c24dd334..0feef8a5 100644
--- a/sling/frame/store.h
+++ b/sling/frame/store.h
@@ -204,8 +204,10 @@ inline Word Align(Word n) {
 // The handle class is implemented as a POD type to make it efficient to pass by
 // value.
 struct Handle {
-  static const int kIntShift = 2;  // integers are shifted two bits
-  static const int kTagBits  = 2;  // the two lowest bits are tag bits
+  static const int kIntShift     = 2;   // integers are shifted two bits
+  static const int kHandleBits   = 32;  // handles are 32-bit integers
+  static const int kTagBits      = 2;   // the two lowest bits are tag bits
+  static const int kRefTagBits   = 3;   // tag bits plus mark bit
 
   static const Word kTagMask     = 0x00000003;  // bit mask for handle tag
   static const Word kRefTagMask  = 0x00000007;  // object reference tag bit mask
@@ -240,6 +242,9 @@ struct Handle {
   static const int kMinInt       = -2147483648 >> kIntShift;
   static const int kMaxInt       = 2147483647 >> kIntShift;
 
+  // Maximum number of handles (local or global).
+  static const int kMaxHandles   = 1 << (kHandleBits - kRefTagBits);
+
   // Returns the tag bits for the value.
   Word tag() const { return bits & kTagMask; }
 
@@ -1355,6 +1360,7 @@ class Store {
   // objects go through the handle table, which provides a level of indirection
   // that allows object to move dynamically, e.g. during garbage collection and
   // when symbols are resolved.
+  static const size_t kMaxHandlesSize = Handle::kMaxHandles * sizeof(Reference);
   Reference *free_handle_;
   Space<Reference> handles_;
 
diff --git a/sling/nlp/document/BUILD b/sling/nlp/document/BUILD
index 42d6a709..a3e3a866 100644
--- a/sling/nlp/document/BUILD
+++ b/sling/nlp/document/BUILD
@@ -18,6 +18,7 @@ cc_library(
     "//sling/frame:object",
     "//sling/frame:store",
     "//sling/string:text",
+    "//sling/util:unicode",
   ],
 )
 
diff --git a/sling/nlp/document/document.cc b/sling/nlp/document/document.cc
index bfd23763..d68320a0 100644
--- a/sling/nlp/document/document.cc
+++ b/sling/nlp/document/document.cc
@@ -34,6 +34,18 @@ uint64 Token::Fingerprint() const {
   return fingerprint_;
 }
 
+CaseForm Token::Form() const {
+  if (form_ == CASE_INVALID) {
+    // Case for first token in a sentence is indeterminate.
+    if (index_ == 0 || brk_ >= SENTENCE_BREAK) {
+      form_ = CASE_NONE;
+    } else {
+      form_ = UTF8::Case(word_);
+    }
+  }
+  return form_;
+}
+
 void Span::Evoke(const Frame &frame) {
   mention_.Add(document_->names_->n_evokes, frame);
   document_->AddMention(frame.handle(), this);
@@ -123,11 +135,19 @@ uint64 Span::Fingerprint() const {
   return fp;
 }
 
+CaseForm Span::Form() const {
+  if (form_ == CASE_INVALID) form_ = document_->Form(begin_, end_);
+  return form_;
+}
+
 Document::Document(Store *store, const DocumentNames *names)
     : themes_(store), names_(names) {
   // Bind names.
-  if (names_ == nullptr) names_ = new DocumentNames(store);
-  names_->AddRef();
+  if (names_ == nullptr) {
+    names_ = new DocumentNames(store);
+  } else {
+    names_->AddRef();
+  }
 
   // Build empty document.
   Builder builder(store);
@@ -138,8 +158,11 @@ Document::Document(Store *store, const DocumentNames *names)
 Document::Document(const Frame &top, const DocumentNames *names)
     : top_(top), themes_(top.store()), names_(names) {
   // Bind names.
-  if (names_ == nullptr) names_ = new DocumentNames(top.store());
-  names_->AddRef();
+  if (names_ == nullptr) {
+    names_ = new DocumentNames(top.store());
+  } else {
+    names_->AddRef();
+  }
 
   // Add document frame if it is missing.
   if (!top_.valid()) {
@@ -190,6 +213,7 @@ Document::Document(const Frame &top, const DocumentNames *names)
         t.brk_ = i == 0 ? NO_BREAK : SPACE_BREAK;
       }
       t.fingerprint_ = 0;
+      t.form_ = CASE_INVALID;
       t.span_ = nullptr;
     }
   }
@@ -311,6 +335,7 @@ void Document::AddToken(Text word, int begin, int end, BreakType brk) {
   t.word_.assign(word.data(), word.size());
   t.brk_ = brk;
   t.fingerprint_ = 0;
+  t.form_ = CASE_INVALID;
   t.span_ = nullptr;
   tokens_changed_ = true;
 }
@@ -412,6 +437,20 @@ uint64 Document::PhraseFingerprint(int begin, int end) {
   return fp;
 }
 
+CaseForm Document::Form(int begin, int end) {
+  CaseForm form = CASE_INVALID;
+  for (int t = begin; t < end; ++t) {
+    if (token(t).skipped()) continue;
+    CaseForm token_form = token(t).Form();
+    if (form == CASE_INVALID) {
+      form = token_form;
+    } else if (form != token_form) {
+      form = CASE_NONE;
+    }
+  }
+  return form;
+}
+
 string Document::PhraseText(int begin, int end) const {
   string phrase;
   for (int t = begin; t < end; ++t) {
@@ -466,7 +505,7 @@ Span *Document::Insert(int begin, int end) {
   bool crossing = false;
   Span *enclosing = EnclosingSpan(begin, end, &crossing);
   if (crossing) return nullptr;
-  
+
   // Check if span already exists.
   if (enclosing != nullptr && enclosing->begin() == begin &&
       enclosing->end() == end) {
diff --git a/sling/nlp/document/document.h b/sling/nlp/document/document.h
index 41185687..a120309a 100644
--- a/sling/nlp/document/document.h
+++ b/sling/nlp/document/document.h
@@ -24,6 +24,7 @@
 #include "sling/frame/store.h"
 #include "sling/nlp/document/token-breaks.h"
 #include "sling/string/text.h"
+#include "sling/util/unicode.h"
 
 namespace sling {
 namespace nlp {
@@ -89,6 +90,12 @@ class Token {
   // Token fingerprint.
   uint64 Fingerprint() const;
 
+  // Token case form.
+  CaseForm Form() const;
+
+  // Punctuation tokens etc. are skipped in phrase comparison.
+  bool skipped() const { return Fingerprint() == 1; }
+
  private:
   Document *document_;          // document the token belongs to
   Handle handle_;               // handle for token in the store
@@ -101,6 +108,8 @@ class Token {
   BreakType brk_;               // break level before token
 
   mutable uint64 fingerprint_;  // fingerprint for token text
+  mutable CaseForm form_;       // case form for token
+
   Span *span_;                  // lowest span covering the token
 
   friend class Document;
@@ -183,6 +192,9 @@ class Span {
   // Returns fingerprint for span phrase.
   uint64 Fingerprint() const;
 
+  // Returns case form for span phrase.
+  CaseForm Form() const;
+
  private:
   // Document that span belongs to.
   Document *document_;
@@ -203,8 +215,9 @@ class Span {
   Span *sibling_ = nullptr;   // first sibling to the right enclosed by parent
   Span *children_ = nullptr;  // left-most enclosed sub-span
 
-  // Span fingerprint. This is lazily initialized and cached.
+  // Span fingerprint and case form. This is lazily initialized and cached.
   mutable uint64 fingerprint_ = 0;
+  mutable CaseForm form_ = CASE_INVALID;
 
   friend class Document;
 };
@@ -295,6 +308,9 @@ class Document {
   // Returns the fingerprint for [begin, end).
   uint64 PhraseFingerprint(int begin, int end);
 
+  // Returns case form forphrase [begin, end).
+  CaseForm Form(int begin, int end);
+
   // Returns the phrase text for span.
   string PhraseText(int begin, int end) const;
 
diff --git a/sling/nlp/kb/BUILD b/sling/nlp/kb/BUILD
index 7212f270..3a52eb53 100644
--- a/sling/nlp/kb/BUILD
+++ b/sling/nlp/kb/BUILD
@@ -26,6 +26,64 @@ cc_library(
   ],
 )
 
+cc_library(
+  name = "name-table-builder",
+  srcs = ["name-table-builder.cc"],
+  deps = [
+    "//sling/base",
+    "//sling/file:repository",
+    "//sling/frame:object",
+    "//sling/task",
+    "//sling/task:frames",
+    "//sling/util:mutex",
+    "//sling/util:unicode",
+  ],
+  alwayslink = 1,
+)
+
+cc_library(
+  name = "name-table",
+  srcs = ["name-table.cc"],
+  hdrs = ["name-table.h"],
+  deps = [
+    "//sling/base",
+    "//sling/file:repository",
+    "//sling/string:text",
+    "//sling/util:unicode",
+  ],
+)
+
+cc_library(
+  name = "phrase-table-builder",
+  srcs = ["phrase-table-builder.cc"],
+  deps = [
+    ":facts",
+    "//sling/base",
+    "//sling/file:repository",
+    "//sling/frame:object",
+    "//sling/frame:serialization",
+    "//sling/nlp/document:phrase-tokenizer",
+    "//sling/nlp/wiki",
+    "//sling/task",
+    "//sling/task:frames",
+    "//sling/util:mutex",
+  ],
+  alwayslink = 1,
+)
+
+cc_library(
+  name = "phrase-table",
+  srcs = ["phrase-table.cc"],
+  hdrs = ["phrase-table.h"],
+  deps = [
+    "//sling/base",
+    "//sling/file:repository",
+    "//sling/frame:store",
+    "//sling/frame:object",
+    "//sling/string:text",
+  ],
+)
+
 embed_data(
   name = "app",
   srcs = [
@@ -41,15 +99,15 @@ cc_library(
   srcs = ["knowledge-service.cc"],
   hdrs = ["knowledge-service.h"],
   deps = [
-    ":calendar",
     ":app",
+    ":calendar",
+    ":name-table",
     "//sling/frame:object",
     "//sling/frame:serialization",
     "//sling/frame:store",
     "//sling/http:http-server",
     "//sling/http:static-content",
     "//sling/http:web-service",
-    "//sling/nlp/wiki:name-table",
   ],
 )
 
@@ -68,3 +126,4 @@ cc_binary(
     "//sling/string:strcat",
   ],
 )
+
diff --git a/sling/nlp/kb/app/index.html b/sling/nlp/kb/app/index.html
index 32c8b6e2..13b57c74 100644
--- a/sling/nlp/kb/app/index.html
+++ b/sling/nlp/kb/app/index.html
@@ -127,7 +127,12 @@ <h2><span>{{active.title}}</span></h2>
               </md-button>
             </div>
 
-            <div class="description">{{item.description}}</div>
+            <div class="description">
+              {{item.description}}
+            </div>
+            <div ng-show="item.type">
+              Data type: {{item.type}}
+            </div>
           </div>
 
           <!-- Item properties card -->
@@ -151,9 +156,12 @@ <h2><span>{{active.title}}</span></h2>
                 <td class="prop-values">
                   <div class="prop-value" ng-repeat="v in p.values">
                     <a ng-href="{{itemUrl(v.ref, v.url)}}">{{v.text}}</a>
+                    <span class="prop-lang" ng-show="v.lang" >
+                      <small>({{v.lang}})</small>
+                    </span>
 
                     <!-- Property qualifiers -->
-                    <table class="qual-tab" ngshow="v.qualifiers">
+                    <table class="qual-tab" ng-show="v.qualifiers">
                       <tr class="qual-row" ng-repeat="qp in v.qualifiers">
                         <td class="qprop-name">
                           <a ng-href="{{itemUrl(qp.ref)}}">{{qp.property}}</a>
@@ -163,6 +171,9 @@ <h2><span>{{active.title}}</span></h2>
                             <a ng-href="{{itemUrl(qv.ref, qv.url)}}">
                               {{qv.text}}
                             </a>
+                            <span class="prop-lang" ng-show="qv.lang" >
+                              <small>({{qv.lang}})</small>
+                            </span>
                           </div>
                         </td>
                       </tr>
diff --git a/sling/nlp/kb/app/kb.css b/sling/nlp/kb/app/kb.css
index 6ca887b8..47848503 100644
--- a/sling/nlp/kb/app/kb.css
+++ b/sling/nlp/kb/app/kb.css
@@ -91,6 +91,10 @@
   cursor: hand;
 }
 
+.prop-lang {
+  color: #808080;
+}
+
 .prop-value a {
   color: #0b0080;
   text-decoration: none;
diff --git a/sling/nlp/kb/calendar.cc b/sling/nlp/kb/calendar.cc
index bc803384..a3d3187e 100644
--- a/sling/nlp/kb/calendar.cc
+++ b/sling/nlp/kb/calendar.cc
@@ -125,6 +125,7 @@ void Date::ParseFromString(Text str) {
 
 void Date::ParseFromFrame(const Frame &frame) {
   // Try to get the 'point in time' property from frame and parse it.
+  if (frame.invalid()) return;
   Store *store = frame.store();
   Object time(store, store->Resolve(frame.GetHandle("P585")));
   if (time.invalid()) return;
@@ -256,19 +257,24 @@ void Calendar::Init(Store *store) {
   if (!cal.valid()) return;
 
   // Build calendar mappings.
-  BuildCalendarMapping(&weekdays_, cal.GetFrame("/w/weekdays"));
-  BuildCalendarMapping(&months_, cal.GetFrame("/w/months"));
-  BuildCalendarMapping(&days_, cal.GetFrame("/w/days"));
-  BuildCalendarMapping(&years_, cal.GetFrame("/w/years"));
-  BuildCalendarMapping(&decades_, cal.GetFrame("/w/decades"));
-  BuildCalendarMapping(&centuries_, cal.GetFrame("/w/centuries"));
-  BuildCalendarMapping(&millennia_, cal.GetFrame("/w/millennia"));
+  BuildCalendarMapping(&weekdays_, nullptr, cal.GetFrame("/w/weekdays"));
+  BuildCalendarMapping(&months_, &month_items_, cal.GetFrame("/w/months"));
+  BuildCalendarMapping(&days_, &day_items_, cal.GetFrame("/w/days"));
+  BuildCalendarMapping(&years_, nullptr, cal.GetFrame("/w/years"));
+  BuildCalendarMapping(&decades_, nullptr, cal.GetFrame("/w/decades"));
+  BuildCalendarMapping(&centuries_, nullptr, cal.GetFrame("/w/centuries"));
+  BuildCalendarMapping(&millennia_, nullptr, cal.GetFrame("/w/millennia"));
 };
 
-bool Calendar::BuildCalendarMapping(CalendarMap *mapping, const Frame &source) {
+bool Calendar::BuildCalendarMapping(CalendarMap *mapping,
+                                    CalendarItemMap *items,
+                                    const Frame &source) {
   if (!source.valid()) return false;
   for (const Slot &s : source) {
     (*mapping)[s.name.AsInt()] = s.value;
+    if (items != nullptr) {
+      (*items)[s.value] = s.name.AsInt();
+    }
   }
   return true;
 }
@@ -352,6 +358,21 @@ string Calendar::DateAsString(const Date &date) const {
   return "???";
 }
 
+bool Calendar::GetDayAndMonth(Handle item, Date *date) const {
+  auto f = day_items_.find(item);
+  if (f == day_items_.end()) return false;
+  date->day = f->second % 100;
+  date->month = f->second / 100;
+  return true;
+}
+
+bool Calendar::GetMonth(Handle item, Date *date) const {
+  auto f = month_items_.find(item);
+  if (f == month_items_.end()) return false;
+  date->month = f->second;
+  return true;
+}
+
 Handle Calendar::Day(const Date &date) const {
   if (date.precision < Date::DAY) return Handle::nil();
   return Day(date.month, date.day);
diff --git a/sling/nlp/kb/calendar.h b/sling/nlp/kb/calendar.h
index 834e96c4..a9640bcf 100644
--- a/sling/nlp/kb/calendar.h
+++ b/sling/nlp/kb/calendar.h
@@ -96,6 +96,12 @@ class Calendar {
     return DateAsString(Date(object));
   }
 
+  // Get day and month for calendar item.
+  bool GetDayAndMonth(Handle item, Date *date) const;
+
+  // Get month for item.
+  bool GetMonth(Handle item, Date *date) const;
+
   // Get item for day.
   Handle Day(const Date &date) const;
   Handle Day(int month, int day) const;
@@ -130,11 +136,16 @@ class Calendar {
   // Mapping from calendar item key to the corresponding calendar item.
   typedef std::unordered_map<int, Handle> CalendarMap;
 
+  // Mapping from calendar item to the corresponding key.
+  typedef HandleMap<int> CalendarItemMap;
+
   // Get name for item.
   Text ItemName(Handle item) const;
 
   // Build calendar mapping.
-  bool BuildCalendarMapping(CalendarMap *mapping, const Frame &source);
+  bool BuildCalendarMapping(CalendarMap *mapping,
+                            CalendarItemMap *items,
+                            const Frame &source);
 
   // Store with calendar.
   Store *store_ = nullptr;
@@ -170,6 +181,12 @@ class Calendar {
   // Millennia. The millennia are numbered as (year-1)/1000+1 for AD and
   // (year+1)/1000-1 for BC.
   CalendarMap millennia_;
+
+  // Mapping from calendar item to day of year (month*100+day).
+  CalendarItemMap day_items_;
+
+  // Mapping from calendar item to month.
+  CalendarItemMap month_items_;
 };
 
 // Date parser and generator based on language-dependent format, e.g.:
diff --git a/sling/nlp/kb/facts.cc b/sling/nlp/kb/facts.cc
index ec911be6..74873dc2 100644
--- a/sling/nlp/kb/facts.cc
+++ b/sling/nlp/kb/facts.cc
@@ -56,6 +56,7 @@ void FactCatalog::Init(Store *store) {
   SetExtractor(p_occupation_, &Facts::ExtractOccupation);
   SetExtractor(p_position_, &Facts::ExtractPosition);
   SetExtractor(p_member_of_sports_team_, &Facts::ExtractTeam);
+  SetExtractor(p_time_period_, &Facts::ExtractTimePeriod);
 
   // Set up items that stops closure expansion.
   static const char *baseids[] = {
@@ -115,7 +116,7 @@ Taxonomy *FactCatalog::CreateDefaultTaxonomy() {
     "Q186081",     // time interval
     "Q11563",      // number
     "Q17376908",   // languoid
-    "Q2198779",    // unit
+    "Q47574",      // unit of measurement
     "Q39875001",   // measure
     "Q3695082",    // sign
     "Q2996394",    // biological process
@@ -133,11 +134,7 @@ Taxonomy *FactCatalog::CreateDefaultTaxonomy() {
     "Q35120",      // entity
     nullptr,
   };
-  std::vector<Text> types;
-  for (const char **type = default_taxonomy; *type != nullptr; ++type) {
-    types.emplace_back(*type);
-  }
-  return new Taxonomy(this, types);
+  return new Taxonomy(this, default_taxonomy);
 }
 
 void Facts::Extract(Handle item) {
@@ -263,6 +260,26 @@ void Facts::ExtractDate(Handle value) {
   AddFact(catalog_->calendar_.Century(date));
 }
 
+void Facts::ExtractTimePeriod(Handle period) {
+  // Add fact for period.
+  ExtractSimple(period);
+
+  // Add facts for start and end time of period.
+  Frame f(store_, store_->Resolve(period));
+  Handle start = f.GetHandle(catalog_->p_start_time_);
+  if (!start.IsNil()) {
+    push(catalog_->p_start_time_);
+    ExtractDate(start);
+    pop();
+  }
+  Handle end = f.GetHandle(catalog_->p_end_time_);
+  if (!end.IsNil()) {
+    push(catalog_->p_end_time_);
+    ExtractDate(end);
+    pop();
+  }
+}
+
 void Facts::ExtractLocation(Handle location) {
   ExtractClosure(location, catalog_->p_located_in_.handle());
 }
@@ -326,6 +343,19 @@ Taxonomy::Taxonomy(const FactCatalog *catalog, const std::vector<Text> &types) {
   }
 }
 
+Taxonomy::Taxonomy(const FactCatalog *catalog, const char **types) {
+  catalog_ = catalog;
+  for (const char **type = types; *type != nullptr; ++type) {
+    Handle t = catalog->store_->LookupExisting(*type);
+    if (t.IsNil()) {
+      LOG(WARNING) << "Ignoring unknown type in taxonomy: " << *type;
+      continue;
+    }
+    int rank = typemap_.size();
+    typemap_[t] = rank;
+  }
+}
+
 Handle Taxonomy::Classify(const Frame &item) {
   // Get immediate types for item.
   Handles types(item.store());
diff --git a/sling/nlp/kb/facts.h b/sling/nlp/kb/facts.h
index 155e910c..79e18b8e 100644
--- a/sling/nlp/kb/facts.h
+++ b/sling/nlp/kb/facts.h
@@ -82,6 +82,9 @@ class FactCatalog {
   Name p_academic_degree_{names_, "P512"};
   Name p_member_of_sports_team_{names_, "P54"};
   Name p_league_{names_, "P118"};
+  Name p_time_period_{names_, "P2348"};
+  Name p_start_time_{names_, "P580"};
+  Name p_end_time_{names_, "P582"};
 
   Name n_time_{names_, "/w/time"};
   Name n_item_{names_, "/w/item"};
@@ -126,6 +129,9 @@ class Facts {
   // Extract date-valued fact with backoff to year, decade and century.
   void ExtractDate(Handle value);
 
+  // Extract time period.
+  void ExtractTimePeriod(Handle period);
+
   // Extract location of item with containment backoff.
   void ExtractPlacement(Handle item);
 
@@ -181,6 +187,7 @@ class Taxonomy {
  public:
   // Initialize taxonomy from a ranked type list.
   Taxonomy(const FactCatalog *catalog, const std::vector<Text> &types);
+  Taxonomy(const FactCatalog *catalog, const char **types);
 
   // Classify item according to taxonomy.
   Handle Classify(const Frame &item);
diff --git a/sling/nlp/kb/knowledge-service.cc b/sling/nlp/kb/knowledge-service.cc
index 840cdb5b..c7d25dfd 100644
--- a/sling/nlp/kb/knowledge-service.cc
+++ b/sling/nlp/kb/knowledge-service.cc
@@ -185,6 +185,13 @@ void KnowledgeService::HandleGetItem(HTTPRequest *request,
   }
   Builder b(ws.store());
   GetStandardProperties(item, &b);
+  Handle datatype = item.GetHandle(n_target_);
+  if (!datatype.IsNil()) {
+    Frame dt(kb_, datatype);
+    if (dt.valid()) {
+      b.Add(n_type_, dt.GetHandle(n_name_));
+    }
+  }
 
   // Fetch properties.
   Item info(ws.store());
@@ -253,8 +260,12 @@ void KnowledgeService::FetchProperties(const Frame &item, Item *info) {
       Handle value = h;
       bool qualified = false;
       if (kb_->IsFrame(h)) {
-        Handle qua = Frame(kb_, h).GetHandle(Handle::is());
-        if (!qua.IsNil()) {
+        // Handle the ambiguity between qualified frames and mono-lingual text
+        // by checking for the presence of a language slot.
+        Frame f(kb_, h);
+        Handle qua = f.GetHandle(Handle::is());
+        Handle lang = f.GetHandle(n_lang_);
+        if (!qua.IsNil() && lang.IsNil()) {
           value = qua;
           qualified = true;
         }
@@ -282,8 +293,17 @@ void KnowledgeService::FetchProperties(const Frame &item, Item *info) {
         // Add string value.
         v.Add(n_text_, value);
       } else if (property.datatype == n_text_type_) {
-        // Add text value.
-        v.Add(n_text_, value);
+        // Add text value with language.
+        if (kb_->IsFrame(value)) {
+          Frame monotext(kb_, value);
+          v.Add(n_text_, monotext.GetHandle(Handle::is()));
+          Frame lang = monotext.GetFrame(n_lang_);
+          if (lang.valid()) {
+            v.Add(n_lang_, lang.GetHandle(n_name_));
+          }
+        } else {
+          v.Add(n_text_, value);
+        }
       } else if (property.datatype == n_url_type_) {
         // Add URL value.
         v.Add(n_text_, value);
diff --git a/sling/nlp/kb/knowledge-service.h b/sling/nlp/kb/knowledge-service.h
index 6872453c..ce5e99fb 100644
--- a/sling/nlp/kb/knowledge-service.h
+++ b/sling/nlp/kb/knowledge-service.h
@@ -23,7 +23,7 @@
 #include "sling/http/http-server.h"
 #include "sling/http/static-content.h"
 #include "sling/nlp/kb/calendar.h"
-#include "sling/nlp/wiki/name-table.h"
+#include "sling/nlp/kb/name-table.h"
 
 namespace sling {
 namespace nlp {
@@ -117,6 +117,7 @@ class KnowledgeService {
   Name n_url_{names_, "url"};
   Name n_thumbnail_{names_, "thumbnail"};
   Name n_matches_{names_, "matches"};
+  Name n_lang_{names_, "lang"};
 
   Name n_xref_type_{names_, "/w/xref"};
   Name n_item_type_{names_, "/w/item"};
diff --git a/sling/nlp/wiki/name-table-builder.cc b/sling/nlp/kb/name-table-builder.cc
similarity index 100%
rename from sling/nlp/wiki/name-table-builder.cc
rename to sling/nlp/kb/name-table-builder.cc
diff --git a/sling/nlp/wiki/name-table.cc b/sling/nlp/kb/name-table.cc
similarity index 98%
rename from sling/nlp/wiki/name-table.cc
rename to sling/nlp/kb/name-table.cc
index b24c38a0..f2974d87 100644
--- a/sling/nlp/wiki/name-table.cc
+++ b/sling/nlp/kb/name-table.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "sling/nlp/wiki/name-table.h"
+#include "sling/nlp/kb/name-table.h"
 
 #include <algorithm>
 #include <string>
diff --git a/sling/nlp/wiki/name-table.h b/sling/nlp/kb/name-table.h
similarity index 100%
rename from sling/nlp/wiki/name-table.h
rename to sling/nlp/kb/name-table.h
diff --git a/sling/nlp/kb/phrase-table-builder.cc b/sling/nlp/kb/phrase-table-builder.cc
new file mode 100644
index 00000000..492075ee
--- /dev/null
+++ b/sling/nlp/kb/phrase-table-builder.cc
@@ -0,0 +1,466 @@
+// Copyright 2017 Google Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <algorithm>
+#include <string>
+#include <unordered_map>
+
+#include "sling/base/logging.h"
+#include "sling/base/types.h"
+#include "sling/file/repository.h"
+#include "sling/frame/object.h"
+#include "sling/nlp/document/phrase-tokenizer.h"
+#include "sling/nlp/kb/facts.h"
+#include "sling/nlp/wiki/wiki.h"
+#include "sling/task/frames.h"
+#include "sling/task/task.h"
+#include "sling/util/mutex.h"
+
+namespace sling {
+namespace nlp {
+
+// Build phrase table repository from aliases.
+class PhraseTableBuilder : public task::FrameProcessor {
+ public:
+  void Startup(task::Task *task) override {
+    // Get language for names.
+    string lang = task->Get("language", "en");
+    language_ = commons_->Lookup("/lang/" + lang);
+    task->Fetch("reliable_alias_sources", &reliable_alias_sources_);
+    task->Fetch("transfer_aliases", &transfer_aliases_);
+
+    // Set phrase normalization.
+    tokenizer_.set_normalization(
+        ParseNormalization(task->Get("normalization", "lcn")));
+
+    // Initialize alias transfer.
+    if (transfer_aliases_) InitAliasTransfer();
+
+    // Statistics.
+    num_aliases_ = task->GetCounter("aliases");
+    num_phrases_ = task->GetCounter("phrases");
+    num_entities_ = task->GetCounter("entities");
+    num_instances_ = task->GetCounter("instances");
+    num_transfers_ = task->GetCounter("alias_transfers");
+    num_zero_transfers_ = task->GetCounter("alias_zero_transfers");
+    num_instance_transfers_ = task->GetCounter("alias_instance_transfers");
+  }
+
+  void Process(Slice key, const Frame &frame) override {
+    MutexLock lock(&mu_);
+
+    // Get index for entity.
+    int index;
+    string id(key.data(), key.size());
+    auto fe = entity_mapping_.find(id);
+    if (fe == entity_mapping_.end()) {
+      index = entity_table_.size();
+      entity_table_.emplace_back(id);
+      num_entities_->Increment();
+      entity_mapping_[id] = index;
+    } else {
+      index = fe->second;
+    }
+
+    // Add aliases.
+    for (const Slot &s : frame) {
+      if (s.name == n_alias_) {
+        // Check language.
+        Frame alias(frame.store(), s.value);
+        if (alias.GetHandle(n_lang_) != language_) continue;
+        num_aliases_->Increment();
+
+        // Compute phrase fingerprint.
+        Text name = alias.GetText(n_name_);
+        int count = alias.GetInt(n_count_, 1);
+        int sources = alias.GetInt(n_sources_, 0);
+        int form = alias.GetInt(n_form_, 0);
+        uint64 fp = tokenizer_.Fingerprint(name);
+        if (fp == 1) continue;
+
+        // Look up or add phrase for entity to phrase table.
+        Phrase *&phrase = phrase_table_[fp];
+        if (phrase == nullptr) {
+          phrase = new Phrase(fp);
+          num_phrases_->Increment();
+        }
+
+        // Add entity to phrase.
+        bool reliable = (sources & reliable_alias_sources_);
+        phrase->entities.emplace_back(index, count, form, reliable);
+
+        // Add alias count to entity frequency.
+        entity_table_[index].count += count;
+        num_instances_->Increment(count);
+      }
+    }
+  }
+
+  void Flush(task::Task *task) override {
+    // Prune phrase table by transfering unreliable aliases to reliable
+    // aliases for related items.
+    if (transfer_aliases_) {
+      LOG(INFO) << "Transfer aliases";
+      TransferAliases();
+    }
+
+    // Build phrase repository.
+    Repository repository;
+
+    // Add normalization flags to repository.
+    string norm = NormalizationString(tokenizer_.normalization());
+    repository.AddBlock("normalization", norm.data(), norm.size());
+
+    // Write entity map.
+    LOG(INFO) << "Build entity map";
+    File *entity_index_block = repository.AddBlock("EntityIndex");
+    File *entity_item_block = repository.AddBlock("EntityItems");
+    uint32 offset = 0;
+    for (Entity &entity : entity_table_) {
+      // Write entity index entry.
+      entity_index_block->WriteOrDie(&offset, sizeof(uint32));
+
+      // Write count and id to entity entry.
+      CHECK_LT(entity.id.size(), 256);
+      uint8 idlen = entity.id.size();
+      entity_item_block->WriteOrDie(&entity.count, sizeof(uint32));
+      entity_item_block->WriteOrDie(&idlen, sizeof(uint8));
+      entity_item_block->WriteOrDie(entity.id.data(), idlen);
+
+      // Compute offset of next entry.
+      offset += sizeof(uint32) + sizeof(uint8) + idlen;
+    }
+
+    // Write phrase map.
+    LOG(INFO) << "Build phrase map";
+    int num_phrases = phrase_table_.size();
+    int num_buckets = (num_phrases + 32) / 32;
+    std::vector<RepositoryMapItem *> items;
+    items.reserve(num_phrases);
+    for (auto &it : phrase_table_) {
+      Phrase *phrase = it.second;
+      items.push_back(phrase);
+
+      // Sort entities in decreasing order.
+      std::sort(phrase->entities.begin(), phrase->entities.end(),
+          [](const EntityPhrase &a, const EntityPhrase &b) {
+            return a.count() > b.count();
+          });
+    }
+    repository.WriteMap("Phrase", &items, num_buckets);
+
+    // Write repository to file.
+    const string &filename = task->GetOutput("repository")->resource()->name();
+    CHECK(!filename.empty());
+    LOG(INFO) << "Write phrase repository to " << filename;
+    repository.Write(filename);
+    LOG(INFO) << "Repository done";
+
+    // Clear collected data.
+    for (auto &it : phrase_table_) delete it.second;
+    phrase_table_.clear();
+    entity_table_.clear();
+    entity_mapping_.clear();
+  }
+
+ private:
+  // Entity with id and frequency.
+  struct Entity {
+    Entity(const string &id) : id(id) {}
+    string id;
+    uint32 count = 0;
+  };
+
+  // Entity phrase with index and frequency. The count_and_flags field contains
+  // the count in the lower 29 bit. Bit 29 and 30 contain the case form, and
+  // bit 31 contains the reliable source flag.
+  struct EntityPhrase {
+    EntityPhrase() = default;
+    EntityPhrase(int index, uint32 count, uint32 form, bool reliable)
+        : index(index),
+          count_and_flags(count | (form << 29) | (reliable ? (1 << 31) : 0)) {}
+    uint32 index;
+    uint32 count_and_flags;
+
+    // Phrase frequency.
+    int count() const { return count_and_flags & ((1 << 29) - 1); }
+    void set_count(uint32 count) {
+      count_and_flags = (count_and_flags & ~((1 << 29) - 1)) | count;
+    }
+
+    // Alias reliability.
+    bool reliable() const { return count_and_flags & (1 << 31); }
+
+    // Phrase form.
+    int form() const { return (count_and_flags >> 29) & 3; }
+  };
+
+  // Phrase with fingerprint and entity distribution.
+  struct Phrase : public RepositoryMapItem {
+    // Initialize new phrase.
+    Phrase(uint64 fingerprint) : fingerprint(fingerprint) {}
+
+    // Write phrase to repository.
+    int Write(File *file) const override {
+      file->WriteOrDie(&fingerprint, sizeof(uint64));
+      uint32 count = entities.size();
+      file->WriteOrDie(&count, sizeof(uint32));
+      for (const EntityPhrase &ep : entities) {
+        file->WriteOrDie(&ep, sizeof(EntityPhrase));
+      }
+      return sizeof(uint64) + sizeof(uint32) + count * sizeof(EntityPhrase);
+    }
+
+    // Use phrase fingerprint as the hash code.
+    uint64 Hash() const override { return fingerprint; }
+
+    uint64 fingerprint;                  // phrase fingerprint
+    std::vector<EntityPhrase> entities;  // list of entities for name phrase
+  };
+
+  // Transfer alias counts from source to target.
+  bool Transfer(EntityPhrase *source, EntityPhrase *target) {
+    // Check for conflicting case forms.
+    int source_form = source->form();
+    int target_form = target->form();
+    if (source_form != CASE_NONE &&
+        target_form != CASE_NONE &&
+        source_form != target_form) {
+      return false;
+    }
+
+    // Check for zero transfers.
+    int source_count = source->count();
+    int target_count = target->count();
+    if (source_count == 0) {
+      num_zero_transfers_->Increment();
+      return false;
+    }
+
+    // Transfer alias counts from source to target.
+    target->set_count(target_count + source_count);
+    source->set_count(0);
+    num_transfers_->Increment();
+    num_instance_transfers_->Increment(source_count);
+    return true;
+  }
+
+  // Exchange aliases between items.
+  bool Exchange(EntityPhrase *a, EntityPhrase *b) {
+    if (a->reliable() && !b->reliable()) {
+      return Transfer(b, a);
+    } else if (b->reliable() && !a->reliable()) {
+      return Transfer(a, b);
+    } else {
+      return false;
+    }
+  }
+
+  void TransferAliases() {
+    // Run over all phrases in phrase table.
+    for (auto &it : phrase_table_) {
+      Phrase *phrase = it.second;
+
+      // There must be more than one entity for any transfers to take place.
+      if (phrase->entities.size() < 2) continue;
+
+      // Build mappings between entity items and entity indices.
+      Store store(commons_);
+      int num_items = phrase->entities.size();
+      Handles entity_item(&store);
+      HandleMap<int> entity_index;
+      entity_item.resize(num_items);
+      for (int i = 0; i < num_items; ++i) {
+        const EntityPhrase &e = phrase->entities[i];
+        const Entity &entity = entity_table_[e.index];
+        Handle item = store.Lookup(entity.id);
+        entity_item[i] = item;
+        entity_index[item] = i;
+      }
+
+      // Find potential targets for alias transfer.
+      bool pruned = false;
+      std::vector<int> numbers;
+      std::vector<int> years;
+      for (int source = 0; source < num_items; ++source) {
+        // Get set of facts for item.
+        Facts facts(&catalog_, &store);
+        facts.Extract(entity_item[source]);
+        for (Handle h :  facts.list()) {
+          Array fact(&store, h);
+          DCHECK_GE(fact.length(), 2);
+
+          // Get head property and target value.
+          Handle p = fact.get(0);
+          Handle t = fact.get(fact.length() - 1);
+
+          // Collect numbers and years.
+          if (p == n_instance_of_) {
+            if (t == n_natural_number_) {
+              numbers.push_back(source);
+            }
+            if (t == n_year_ || t == n_year_bc_ || t == n_decade_) {
+              years.push_back(source);
+            }
+          }
+
+          // Check for property exceptions.
+          if (transfer_exceptions_.count(p) > 0) continue;
+
+          // Check if target has the phrase as an alias.
+          auto f = entity_index.find(t);
+          if (f == entity_index.end()) continue;
+          int target = f->second;
+          if (target == source) continue;
+
+          // Transfer alias from unreliable to reliable alias.
+          auto &src = phrase->entities[source];
+          auto &tgt = phrase->entities[target];
+          if (Exchange(&src, &tgt)) pruned = true;
+        }
+      }
+
+      // Transfer aliases for years.
+      if (!years.empty()) {
+        for (int source = 0; source < years.size(); ++source) {
+          for (int target = 0; target < years.size(); ++target) {
+            if (source == target) continue;
+            auto &src = phrase->entities[years[source]];
+            auto &tgt = phrase->entities[years[target]];
+            if (Exchange(&src, &tgt)) pruned = true;
+          }
+        }
+      }
+
+      // Transfer aliases for numbers.
+      if (!numbers.empty()) {
+        for (int source = 0; source < numbers.size(); ++source) {
+          for (int target = 0; target < numbers.size(); ++target) {
+            if (source == target) continue;
+            auto &src = phrase->entities[numbers[source]];
+            auto &tgt = phrase->entities[numbers[target]];
+            if (Exchange(&src, &tgt)) pruned = true;
+          }
+        }
+      }
+
+      // Prune aliases with zero count.
+      if (pruned) {
+        int j = 0;
+        for (int i = 0; i < num_items; ++i) {
+          if (phrase->entities[i].count() == 0) continue;
+          if (i != j) phrase->entities[j] = phrase->entities[i];
+          j++;
+        }
+        phrase->entities.resize(j);
+      }
+    }
+  }
+
+  void InitAliasTransfer() {
+    // Initialize alias transfer exceptions.
+    static const char *exceptions[] = {
+      "P1889",  // different from
+      "P460",   // said to be the same as
+      "P1533",  // identical to this given name
+      "P138",   // named after
+      "P2959",  // permanent duplicated item
+      "P734",   // family name
+      "P735",   // given name
+      "P112",   // founded by
+      "P115",   // home venue
+      "P144",   // based on
+      "P1950",  // second family name in Spanish name
+      "P2359",  // Roman nomen gentilicium
+      "P2358",  // Roman praenomen
+      "P2365",  // Roman cognomen
+      "P2366",  // Roman agnomen
+      "P941",   // inspired by
+      "P629",   // edition or translation of
+      "P37",    // official language
+      "P103",   // native language
+      "P566",   // basionym
+      nullptr
+    };
+    for (const char **p = exceptions; *p != nullptr; ++p) {
+      transfer_exceptions_.insert(commons_->LookupExisting(*p));
+    }
+
+    // Initialize fact catalog.
+    catalog_.Init(commons_);
+  }
+
+  // Symbols.
+  Name n_lang_{names_, "lang"};
+  Name n_name_{names_, "name"};
+  Name n_alias_{names_, "alias"};
+  Name n_count_{names_, "count"};
+  Name n_form_{names_, "form"};
+  Name n_sources_{names_, "sources"};
+
+  Name n_instance_of_{names_, "P31"};
+  Name n_natural_number_{names_, "Q21199"};
+  Name n_year_{names_, "Q577"};
+  Name n_year_bc_{names_, "Q29964144"};
+  Name n_decade_{names_, "Q39911"};
+
+  // Language for aliases.
+  Handle language_;
+
+  // Reliable alias sources.
+  int reliable_alias_sources_ =
+    (1 << SRC_WIKIDATA_LABEL) |
+    (1 << SRC_WIKIDATA_ALIAS) |
+    (1 << SRC_WIKIDATA_NAME) |
+    (1 << SRC_WIKIDATA_DEMONYM);
+
+  // Phrase tokenizer.
+  PhraseTokenizer tokenizer_;
+
+  // Sorted name table mapping phrase fingerprints to entities.
+  std::unordered_map<uint64, Phrase *> phrase_table_;
+
+  // Entity table with id and frequency count.
+  std::vector<Entity> entity_table_;
+
+  // Mapping of entity id to entity index in entity table.
+  std::unordered_map<string, int> entity_mapping_;
+
+  // Alias transfer.
+  bool transfer_aliases_ = false;
+
+  // Fact catalog for alias transfer.
+  FactCatalog catalog_;
+
+  // Property exceptions for alias transfer.
+  HandleSet transfer_exceptions_;
+
+  // Statistics.
+  task::Counter *num_phrases_ = nullptr;
+  task::Counter *num_entities_ = nullptr;
+  task::Counter *num_aliases_ = nullptr;
+  task::Counter *num_instances_ = nullptr;
+  task::Counter *num_transfers_ = nullptr;
+  task::Counter *num_zero_transfers_ = nullptr;
+  task::Counter *num_instance_transfers_ = nullptr;
+
+  // Mutex for serializing access to repository.
+  Mutex mu_;
+};
+
+REGISTER_TASK_PROCESSOR("phrase-table-builder", PhraseTableBuilder);
+
+}  // namespace nlp
+}  // namespace sling
+
diff --git a/sling/nlp/wiki/phrase-table.cc b/sling/nlp/kb/phrase-table.cc
similarity index 57%
rename from sling/nlp/wiki/phrase-table.cc
rename to sling/nlp/kb/phrase-table.cc
index b8e746c0..4c67adda 100644
--- a/sling/nlp/wiki/phrase-table.cc
+++ b/sling/nlp/kb/phrase-table.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "sling/nlp/wiki/phrase-table.h"
+#include "sling/nlp/kb/phrase-table.h"
 
 namespace sling {
 namespace nlp {
@@ -39,7 +39,7 @@ void PhraseTable::Load(Store *store, const string &filename) {
   entity_table_->resize(entity_index_.size());
 }
 
-Handle PhraseTable::GetEntityHandle(int index) {
+Handle PhraseTable::GetEntityHandle(int index) const {
   Handle handle = (*entity_table_)[index];
   if (handle.IsNil()) {
     const EntityItem *entity = entity_index_.GetEntity(index);
@@ -52,47 +52,56 @@ Handle PhraseTable::GetEntityHandle(int index) {
   return handle;
 }
 
-void PhraseTable::Lookup(uint64 fp, Handles *matches) {
-  matches->clear();
+const PhraseTable::Phrase *PhraseTable::Find(uint64 fp) const {
   int bucket = fp % phrase_index_.num_buckets();
   const PhraseItem *phrase = phrase_index_.GetBucket(bucket);
   const PhraseItem *end = phrase_index_.GetBucket(bucket + 1);
   while (phrase < end) {
-    if (phrase->fingerprint() == fp) {
-      const EntityPhrase *entities = phrase->entities();
-      for (int i = 0; i < phrase->num_entities(); ++i) {
-        int index = entities[i].index;
-        Handle handle = GetEntityHandle(index);
-        matches->push_back(handle);
-      }
-      break;
-    }
+    if (phrase->fingerprint() == fp) return phrase;
     phrase = phrase->next();
   }
+  return nullptr;
 }
 
-void PhraseTable::Lookup(uint64 fp, MatchList *matches) {
-  matches->clear();
-  int bucket = fp % phrase_index_.num_buckets();
-  const PhraseItem *phrase = phrase_index_.GetBucket(bucket);
-  const PhraseItem *end = phrase_index_.GetBucket(bucket + 1);
-  while (phrase < end) {
-    if (phrase->fingerprint() == fp) {
-      const EntityPhrase *entities = phrase->entities();
-      for (int i = 0; i < phrase->num_entities(); ++i) {
-        int index = entities[i].index;
-        Text id = entity_index_.GetEntityId(index);
-        Handle handle = GetEntityHandle(index);
-        int count = entities[i].count();
-        int form = entities[i].form();
-        bool reliable = entities[i].reliable();
-        matches->emplace_back(id, handle, count, form, reliable);
-      }
-      break;
-    }
-    phrase = phrase->next();
+void PhraseTable::GetMatches(const Phrase *phrase, Handles *matches) const {
+  if (phrase == nullptr) {
+    matches->clear();
+    return;
+  }
+  const EntityPhrase *entities = phrase->entities();
+  matches->resize(phrase->num_entities());
+  for (int i = 0; i < phrase->num_entities(); ++i) {
+    int index = entities[i].index;
+    (*matches)[i] = GetEntityHandle(index);
   }
 }
 
+void PhraseTable::GetMatches(const Phrase *phrase, MatchList *matches) const {
+  if (phrase == nullptr) {
+    matches->clear();
+    return;
+  }
+  const EntityPhrase *entities = phrase->entities();
+  matches->resize(phrase->num_entities());
+  for (int i = 0; i < phrase->num_entities(); ++i) {
+    int index = entities[i].index;
+    Match &match = (*matches)[i];
+    match.id = entity_index_.GetEntityId(index);
+    match.item = GetEntityHandle(index);
+    auto &entity = entities[i];
+    match.count = entity.count();
+    match.form = entity.form();
+    match.reliable = entity.reliable();
+  }
+}
+
+void PhraseTable::Lookup(uint64 fp, Handles *matches) const {
+  GetMatches(Find(fp), matches);
+}
+
+void PhraseTable::Lookup(uint64 fp, MatchList *matches) const {
+  GetMatches(Find(fp), matches);
+}
+
 }  // namespace nlp
 }  // namespace sling
diff --git a/sling/nlp/wiki/phrase-table.h b/sling/nlp/kb/phrase-table.h
similarity index 88%
rename from sling/nlp/wiki/phrase-table.h
rename to sling/nlp/kb/phrase-table.h
index 43e02973..e9dab4c9 100644
--- a/sling/nlp/wiki/phrase-table.h
+++ b/sling/nlp/kb/phrase-table.h
@@ -32,8 +32,6 @@ namespace nlp {
 class PhraseTable {
  public:
   struct Match {
-    Match(Text id, Handle item, int count, int form, bool reliable)
-      : id(id), item(item), count(count), form(form), reliable(reliable) {}
     Text id;        // entity id of matching item
     Handle item;    // matching item
     int count;      // frequency of matching item
@@ -49,17 +47,17 @@ class PhraseTable {
   void Load(Store *store, const string &filename);
 
   // Find all entities matching a phrase fingerprint.
-  void Lookup(uint64 fp, Handles *matches);
+  void Lookup(uint64 fp, Handles *matches) const;
 
   // Find all entities matching a phrase fingerprint and return list of matches.
-  void Lookup(uint64 fp, MatchList *matches);
+  void Lookup(uint64 fp, MatchList *matches) const;
 
   // Text normalization flags.
   const string &normalization() const { return normalization_; }
 
  private:
   // Get handle for entity.
-  Handle GetEntityHandle(int index);
+  Handle GetEntityHandle(int index) const;
 
   // Entity phrase with entity index and frequency. The count_and_flags field
   // contains the count in the lower 29 bit. Bit 29 and 30 contain the case
@@ -150,6 +148,20 @@ class PhraseTable {
     }
   };
 
+ public:
+  // Opaque public type for phrase items in the phrase table.
+  typedef PhraseItem Phrase;
+
+  // Find matching phrase in phrase table. Return null if phrase is not found.
+  const Phrase *Find(uint64 fp) const;
+
+  // Get matching handles for phrase.
+  void GetMatches(const Phrase *phrase, Handles *matches) const;
+
+  // Get matching items for phrase.
+  void GetMatches(const Phrase *phrase, MatchList *matches) const;
+
+ private:
   // Repository with name table.
   Repository repository_;
 
@@ -166,7 +178,7 @@ class PhraseTable {
   Handles *entity_table_ = nullptr;
 
   // Text normalization flags.
-  string normalization_ = "lcp";
+  string normalization_ = "lcn";
 };
 
 }  // namespace nlp
diff --git a/sling/nlp/wiki/BUILD b/sling/nlp/wiki/BUILD
index 9d01c262..1766edce 100644
--- a/sling/nlp/wiki/BUILD
+++ b/sling/nlp/wiki/BUILD
@@ -66,7 +66,7 @@ cc_library(
   deps = [
     ":wiki-extractor",
     "//sling/base:registry",
-    "//sling/frame:object",
+    "//sling/frame",
     "//sling/nlp/document",
     "//sling/string:numbers",
     "//sling/string:strcat",
@@ -151,13 +151,13 @@ cc_library(
 )
 
 cc_library(
-  name = "profile-aliases",
-  srcs = ["profile-aliases.cc"],
+  name = "aliases",
+  srcs = ["aliases.cc"],
   deps = [
     ":wiki",
     "//sling/base",
     "//sling/file:textmap",
-    "//sling/frame:object",
+    "//sling/frame",
     "//sling/nlp/document:phrase-tokenizer",
     "//sling/task",
     "//sling/task:frames",
@@ -167,62 +167,6 @@ cc_library(
   alwayslink = 1,
 )
 
-cc_library(
-  name = "name-table-builder",
-  srcs = ["name-table-builder.cc"],
-  deps = [
-    "//sling/base",
-    "//sling/file:repository",
-    "//sling/frame:object",
-    "//sling/task",
-    "//sling/task:frames",
-    "//sling/util:mutex",
-    "//sling/util:unicode",
-  ],
-  alwayslink = 1,
-)
-
-cc_library(
-  name = "name-table",
-  srcs = ["name-table.cc"],
-  hdrs = ["name-table.h"],
-  deps = [
-    "//sling/base",
-    "//sling/file:repository",
-    "//sling/string:text",
-    "//sling/util:unicode",
-  ],
-)
-
-cc_library(
-  name = "phrase-table-builder",
-  srcs = ["phrase-table-builder.cc"],
-  deps = [
-    ":wiki",
-    "//sling/base",
-    "//sling/file:repository",
-    "//sling/frame:object",
-    "//sling/nlp/document:phrase-tokenizer",
-    "//sling/task",
-    "//sling/task:frames",
-    "//sling/util:mutex",
-  ],
-  alwayslink = 1,
-)
-
-cc_library(
-  name = "phrase-table",
-  srcs = ["phrase-table.cc"],
-  hdrs = ["phrase-table.h"],
-  deps = [
-    "//sling/base",
-    "//sling/file:repository",
-    "//sling/frame:store",
-    "//sling/frame:object",
-    "//sling/string:text",
-  ],
-)
-
 cc_binary(
   name = "parse-wiki-text",
   srcs = ["parse-wiki-text.cc"],
diff --git a/sling/nlp/wiki/profile-aliases.cc b/sling/nlp/wiki/aliases.cc
similarity index 85%
rename from sling/nlp/wiki/profile-aliases.cc
rename to sling/nlp/wiki/aliases.cc
index 38abc665..afd642d6 100644
--- a/sling/nlp/wiki/profile-aliases.cc
+++ b/sling/nlp/wiki/aliases.cc
@@ -29,8 +29,8 @@
 namespace sling {
 namespace nlp {
 
-// Extract aliases from profiles.
-class ProfileAliasExtractor : public task::FrameProcessor {
+// Extract aliases for items.
+class AliasExtractor : public task::FrameProcessor {
  public:
   void Startup(task::Task *task) override {
     string lang = task->Get("language", "en");
@@ -72,12 +72,27 @@ class ProfileAliasExtractor : public task::FrameProcessor {
       } else if (s.name == n_demonym_) {
         // Output demonyms as demonym aliases.
         AddAlias(&a, store->Resolve(s.value), SRC_WIKIDATA_DEMONYM);
+      } else if (s.name == n_iso3166_country_code_2_ ||
+                 s.name == n_iso3166_country_code_3_) {
+        // Output country codes as alternative names.
+        AddAlias(&a, store->Resolve(s.value), SRC_WIKIDATA_NAME);
+      } else if (s.name == n_short_name_) {
+        // Output short names as alternative or foreign names.
+        Handle lang = Handle::nil();
+        Frame f(store, s.value);
+        if (f.valid()) lang = f.GetHandle(n_lang_);
+        if (lang.IsNil() || lang == language_) {
+          AddAlias(&a, store->Resolve(s.value), SRC_WIKIDATA_NAME);
+        } else {
+          AddAlias(&a, store->Resolve(s.value), SRC_WIKIDATA_FOREIGN);
+        }
       } else if (s.name == n_instance_of_) {
         // Discard categories, disambiguations, info boxes and templates.
-        if (wikitypes_.IsCategory(s.value) ||
-            wikitypes_.IsDisambiguation(s.value) ||
-            wikitypes_.IsInfobox(s.value) ||
-            wikitypes_.IsTemplate(s.value)) {
+        Handle type = store->Resolve(s.value);
+        if (wikitypes_.IsCategory(type) ||
+            wikitypes_.IsDisambiguation(type) ||
+            wikitypes_.IsInfobox(type) ||
+            wikitypes_.IsTemplate(type)) {
           return;
         }
       }
@@ -124,12 +139,16 @@ class ProfileAliasExtractor : public task::FrameProcessor {
   Name n_native_name_{names_, "P1559"};
   Name n_native_label_{names_, "P1705"};
   Name n_demonym_{names_, "P1549"};
+  Name n_short_name_{names_, "P1813"};
+  Name n_iso3166_country_code_2_{names_, "P297"};
+  Name n_iso3166_country_code_3_{names_, "P298"};
+
   Name n_instance_of_{names_, "P31"};
 };
 
-REGISTER_TASK_PROCESSOR("profile-alias-extractor", ProfileAliasExtractor);
+REGISTER_TASK_PROCESSOR("alias-extractor", AliasExtractor);
 
-class ProfileAliasReducer : public task::Reducer {
+class AliasReducer : public task::Reducer {
  public:
   struct Alias {
     std::unordered_map<string, int> variants;
@@ -165,11 +184,11 @@ class ProfileAliasReducer : public task::Reducer {
     Store store(&commons_);
     std::unordered_map<uint64, Alias *> aliases;
     for (task::Message *message : input.messages()) {
-      // Get next alias profile.
-      Frame profile = DecodeMessage(&store, message);
+      // Get next set of aliases for item.
+      Frame batch = DecodeMessage(&store, message);
 
-      // Get all aliases from profile.
-      for (const Slot &slot : profile) {
+      // Get all aliases for item.
+      for (const Slot &slot : batch) {
         if (slot.name != n_alias_) continue;
         Frame alias(&store, slot.value);
         string name = alias.GetString(n_name_);
@@ -238,7 +257,7 @@ class ProfileAliasReducer : public task::Reducer {
       merged.Add(n_alias_, a.Create());
     }
 
-    // Output alias profile.
+    // Output selected aliases.
     Output(input.shard(), task::CreateMessage(qid, merged.Create()));
 
     // Delete alias table.
@@ -247,7 +266,7 @@ class ProfileAliasReducer : public task::Reducer {
 
   // Check if alias should be selected.
   bool SelectAlias(Alias *alias, bool toxic) {
-    // Keep aliases from trusted sources.
+    // Keep aliases from "trusted" sources.
     if (alias->sources & (WIKIDATA_LABEL |
                           WIKIPEDIA_TITLE |
                           WIKIPEDIA_REDIRECT)) {
@@ -255,7 +274,9 @@ class ProfileAliasReducer : public task::Reducer {
     }
 
     // Only keep Wikidata alias if it is not toxic.
-    if ((alias->sources & WIKIDATA_ALIAS) && !toxic) return true;
+    if (alias->sources & (WIKIDATA_ALIAS | WIKIDATA_NAME)) {
+      return !toxic;
+    }
 
     // Keep foreign, native and demonym aliases supported by Wikipedia aliases.
     if (alias->sources & (WIKIDATA_FOREIGN |
@@ -295,6 +316,7 @@ class ProfileAliasReducer : public task::Reducer {
     WIKIDATA_NATIVE = 1 << SRC_WIKIDATA_NATIVE,
     WIKIDATA_DEMONYM = 1 << SRC_WIKIDATA_DEMONYM,
     WIKIPEDIA_LINK = 1 << SRC_WIKIPEDIA_LINK,
+    WIKIDATA_NAME = 1 << SRC_WIKIDATA_NAME,
   };
 
   // Commons store.
@@ -326,7 +348,7 @@ class ProfileAliasReducer : public task::Reducer {
   std::set<uint64> toxic_aliases_;
 };
 
-REGISTER_TASK_PROCESSOR("profile-alias-reducer", ProfileAliasReducer);
+REGISTER_TASK_PROCESSOR("alias-reducer", AliasReducer);
 
 }  // namespace nlp
 }  // namespace sling
diff --git a/sling/nlp/wiki/phrase-table-builder.cc b/sling/nlp/wiki/phrase-table-builder.cc
deleted file mode 100644
index d19ad345..00000000
--- a/sling/nlp/wiki/phrase-table-builder.cc
+++ /dev/null
@@ -1,249 +0,0 @@
-// Copyright 2017 Google Inc.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#include <algorithm>
-#include <string>
-#include <unordered_map>
-
-#include "sling/base/logging.h"
-#include "sling/base/types.h"
-#include "sling/file/repository.h"
-#include "sling/nlp/document/phrase-tokenizer.h"
-#include "sling/nlp/wiki/wiki.h"
-#include "sling/task/frames.h"
-#include "sling/task/task.h"
-#include "sling/util/mutex.h"
-
-namespace sling {
-namespace nlp {
-
-// Build phrase table repository from aliases.
-class PhraseTableBuilder : public task::FrameProcessor {
- public:
-  void Startup(task::Task *task) override {
-    // Get language for names.
-    string lang = task->Get("language", "en");
-    language_ = commons_->Lookup("/lang/" + lang);
-    task->Fetch("noisy_alias_sources", &noisy_alias_sources_);
-
-    // Set phrase normalization.
-    tokenizer_.set_normalization(
-        ParseNormalization(task->Get("normalization", "lcp")));
-
-    // Statistics.
-    num_aliases_ = task->GetCounter("aliases");
-    num_phrases_ = task->GetCounter("phrases");
-    num_entities_ = task->GetCounter("entities");
-    num_instances_ = task->GetCounter("instances");
-  }
-
-  void Process(Slice key, const Frame &frame) override {
-    MutexLock lock(&mu_);
-
-    // Get index for entity.
-    int index;
-    string id(key.data(), key.size());
-    auto fe = entity_mapping_.find(id);
-    if (fe == entity_mapping_.end()) {
-      index = entity_table_.size();
-      entity_table_.emplace_back(id);
-      num_entities_->Increment();
-      entity_mapping_[id] = index;
-    } else {
-      index = fe->second;
-    }
-
-    // Add aliases.
-    for (const Slot &s : frame) {
-      if (s.name == n_alias_) {
-        // Check language.
-        Frame alias(frame.store(), s.value);
-        if (alias.GetHandle(n_lang_) != language_) continue;
-        num_aliases_->Increment();
-
-        // Compute phrase fingerprint.
-        Text name = alias.GetText(n_name_);
-        int count = alias.GetInt(n_count_, 1);
-        int sources = alias.GetInt(n_sources_, 0);
-        int form = alias.GetInt(n_form_, 0);
-        uint64 fp = tokenizer_.Fingerprint(name);
-        if (fp == 1) continue;
-
-        // Look up or add phrase for entity to phrase table.
-        Phrase *&phrase = phrase_table_[fp];
-        if (phrase == nullptr) {
-          phrase = new Phrase(fp);
-          num_phrases_->Increment();
-        }
-
-        // Add entity to phrase.
-        bool reliable = (sources & ~noisy_alias_sources_);
-        uint32 count_and_flags = count | (form << 29);
-        if (reliable) count_and_flags |= (1 << 31);
-        phrase->entities.emplace_back(index, count_and_flags);
-
-        // Add alias count to entity frequency.
-        entity_table_[index].count += count;
-        num_instances_->Increment(count);
-      }
-    }
-  }
-
-  void Flush(task::Task *task) override {
-    // Build phrase repository.
-    Repository repository;
-
-    // Add normalization flags to repository.
-    string norm = NormalizationString(tokenizer_.normalization());
-    repository.AddBlock("normalization", norm.data(), norm.size());
-
-    // Write entity map.
-    LOG(INFO) << "Build entity map";
-    File *entity_index_block = repository.AddBlock("EntityIndex");
-    File *entity_item_block = repository.AddBlock("EntityItems");
-    uint32 offset = 0;
-    for (Entity &entity : entity_table_) {
-      // Write entity index entry.
-      entity_index_block->WriteOrDie(&offset, sizeof(uint32));
-
-      // Write count and id to entity entry.
-      CHECK_LT(entity.id.size(), 256);
-      uint8 idlen = entity.id.size();
-      entity_item_block->WriteOrDie(&entity.count, sizeof(uint32));
-      entity_item_block->WriteOrDie(&idlen, sizeof(uint8));
-      entity_item_block->WriteOrDie(entity.id.data(), idlen);
-
-      // Compute offset of next entry.
-      offset += sizeof(uint32) + sizeof(uint8) + idlen;
-    }
-
-    // Write phrase map.
-    LOG(INFO) << "Build phrase map";
-    int num_phrases = phrase_table_.size();
-    int num_buckets = (num_phrases + 32) / 32;
-    std::vector<RepositoryMapItem *> items;
-    items.reserve(num_phrases);
-    for (auto &it : phrase_table_) {
-      Phrase *phrase = it.second;
-      items.push_back(phrase);
-
-      // Sort entities in decreasing order.
-      std::sort(phrase->entities.begin(), phrase->entities.end(),
-          [](const EntityPhrase &a, const EntityPhrase &b) {
-            return a.count() > b.count();
-          });
-    }
-    repository.WriteMap("Phrase", &items, num_buckets);
-
-    // Write repository to file.
-    const string &filename = task->GetOutput("repository")->resource()->name();
-    CHECK(!filename.empty());
-    LOG(INFO) << "Write phrase repository to " << filename;
-    repository.Write(filename);
-    LOG(INFO) << "Repository done";
-
-    // Clear collected data.
-    for (auto &it : phrase_table_) delete it.second;
-    phrase_table_.clear();
-    entity_table_.clear();
-    entity_mapping_.clear();
-  }
-
- private:
-  // Entity with id and frequency.
-  struct Entity {
-    Entity(const string &id) : id(id) {}
-    string id;
-    uint32 count = 0;
-  };
-
-  // Entity phrase with index and frequency. The count_and_flags field contains
-  // the count in the lower 29 bit. Bit 29 and 30 contain the case form, and
-  // bit 31 contains the reliable source flag.
-  struct EntityPhrase {
-    EntityPhrase(int index, uint32 count_and_flags)
-        : index(index), count_and_flags(count_and_flags) {}
-    uint32 index;
-    uint32 count_and_flags;
-    int count() const { return count_and_flags & ((1 << 29) - 1); }
-  };
-
-  // Phrase with fingerprint and entity distribution.
-  struct Phrase : public RepositoryMapItem {
-    // Initialize new phrase.
-    Phrase(uint64 fingerprint) : fingerprint(fingerprint) {}
-
-    // Write phrase to repository.
-    int Write(File *file) const override {
-      file->WriteOrDie(&fingerprint, sizeof(uint64));
-      uint32 count = entities.size();
-      file->WriteOrDie(&count, sizeof(uint32));
-      for (const EntityPhrase &ep : entities) {
-        file->WriteOrDie(&ep, sizeof(EntityPhrase));
-      }
-      return sizeof(uint64) + sizeof(uint32) + count * sizeof(EntityPhrase);
-    }
-
-    // Use phrase fingerprint as the hash code.
-    uint64 Hash() const override { return fingerprint; }
-
-    uint64 fingerprint;                  // phrase fingerprint
-    std::vector<EntityPhrase> entities;  // list of entities for name phrase
-  };
-
-  // Symbols.
-  Name n_lang_{names_, "lang"};
-  Name n_name_{names_, "name"};
-  Name n_alias_{names_, "alias"};
-  Name n_count_{names_, "count"};
-  Name n_form_{names_, "form"};
-  Name n_sources_{names_, "sources"};
-
-  // Language for aliases.
-  Handle language_;
-
-  // Noisy alias sources. Aliases that are only backed by noisy alias sources
-  // are not marked as reliable.
-  int noisy_alias_sources_ =
-    (1 << SRC_WIKIPEDIA_ANCHOR) |
-    (1 << SRC_WIKIPEDIA_LINK) |
-    (1 << SRC_WIKIPEDIA_DISAMBIGUATION);
-
-  // Phrase tokenizer.
-  PhraseTokenizer tokenizer_;
-
-  // Sorted name table mapping phrase fingerprints to entities.
-  std::unordered_map<uint64, Phrase *> phrase_table_;
-
-  // Entity table with id and frequency count.
-  std::vector<Entity> entity_table_;
-
-  // Mapping of entity id to entity index in entity table.
-  std::unordered_map<string, int> entity_mapping_;
-
-  // Statistics.
-  task::Counter *num_phrases_ = nullptr;
-  task::Counter *num_entities_ = nullptr;
-  task::Counter *num_aliases_ = nullptr;
-  task::Counter *num_instances_ = nullptr;
-
-  // Mutex for serializing access to repository.
-  Mutex mu_;
-};
-
-REGISTER_TASK_PROCESSOR("phrase-table-builder", PhraseTableBuilder);
-
-}  // namespace nlp
-}  // namespace sling
-
diff --git a/sling/nlp/wiki/wiki-annotator.cc b/sling/nlp/wiki/wiki-annotator.cc
index ad97f5c8..6f623da3 100644
--- a/sling/nlp/wiki/wiki-annotator.cc
+++ b/sling/nlp/wiki/wiki-annotator.cc
@@ -209,25 +209,23 @@ void WikiAnnotator::Link(const Node &node,
                          bool unanchored) {
   // Resolve link.
   Text link = resolver_->ResolveLink(node.name());
-  if (link.empty()) {
-    if (!unanchored) extractor->ExtractChildren(node);
-    return;
-  }
 
   if (unanchored) {
-    // Extract anchor as plain text.
-    WikiPlainTextSink plain;
-    extractor->Enter(&plain);
-    extractor->ExtractChildren(node);
-    extractor->Leave(&plain);
-
-    // Add thematic frame for link.
-    if (!plain.text().empty()) {
-      Builder theme(store_);
-      theme.AddIsA(n_link_);
-      theme.Add(n_name_, plain.text());
-      theme.AddIs(store_->Lookup(link));
-      AddTheme(theme.Create().handle());
+    if (!link.empty()) {
+      // Extract anchor as plain text.
+      WikiPlainTextSink plain;
+      extractor->Enter(&plain);
+      extractor->ExtractChildren(node);
+      extractor->Leave(&plain);
+
+      // Add thematic frame for link.
+      if (!plain.text().empty()) {
+        Builder theme(store_);
+        theme.AddIsA(n_link_);
+        theme.Add(n_name_, plain.text());
+        theme.AddIs(store_->Lookup(link));
+        AddTheme(theme.Create().handle());
+      }
     }
   } else {
     // Output anchor text.
@@ -237,7 +235,8 @@ void WikiAnnotator::Link(const Node &node,
 
     // Evoke frame for link.
     if (begin != end) {
-      AddMention(begin, end, store_->Lookup(link));
+      Handle evoke = link.empty() ? Handle::nil() : store_->Lookup(link);
+      AddMention(begin, end, evoke);
     }
   }
 }
@@ -277,7 +276,9 @@ void WikiAnnotator::AddToDocument(Document *document) {
     int begin = document->Locate(a.begin.AsInt());
     int end = document->Locate(a.end.AsInt());
     Span *span = document->AddSpan(begin, end);
-    span->Evoke(a.evoked);
+    if (!a.evoked.IsNil()) {
+      span->Evoke(a.evoked);
+    }
   }
 
   // Add thematic frames.
diff --git a/sling/nlp/wiki/wiki-annotator.h b/sling/nlp/wiki/wiki-annotator.h
index 1a9d3eab..31b2cb39 100644
--- a/sling/nlp/wiki/wiki-annotator.h
+++ b/sling/nlp/wiki/wiki-annotator.h
@@ -166,8 +166,8 @@ class WikiAnnotator : public WikiTextSink {
   // the store and links will be resolved using the resolver.
   WikiAnnotator(Store *store, WikiLinkResolver *resolver);
 
-  // Initialize sub-annotator based on another annotator. Plase notice that this
-  // is not a copy constructor.
+  // Initialize sub-annotator based on another annotator. Please notice that
+  // this is not a copy constructor.
   explicit WikiAnnotator(WikiAnnotator *other);
 
   // Wiki sink interface receiving the annotations from the extractor.
diff --git a/sling/nlp/wiki/wiki-macros.cc b/sling/nlp/wiki/wiki-macros.cc
index d5d50982..61083625 100644
--- a/sling/nlp/wiki/wiki-macros.cc
+++ b/sling/nlp/wiki/wiki-macros.cc
@@ -513,8 +513,11 @@ REGISTER_WIKI_MACRO("measure", MeasureTemplate);
 // Template macro for info boxes.
 class InfoboxTemplate : public WikiMacro {
  public:
+  ~InfoboxTemplate() { if (docnames_) docnames_->Release(); }
+
   void Init(const Frame &config) override {
     Store *store = config.store();
+    docnames_ = new DocumentNames(store);
     Handle n_class = store->Lookup("class");
     Handle n_fields = store->Lookup("fields");
     Handle n_group = store->Lookup("group");
@@ -566,7 +569,7 @@ class InfoboxTemplate : public WikiMacro {
         field = &f->second;
         if (!field->group.IsNil()) index = 0;
       } else {
-        // Try to remove number suffix for repreated field.
+        // Try to remove number suffix for repeated field.
         int i = name.size() - 1;
         int power = 1;
         index = 0;
@@ -596,7 +599,7 @@ class InfoboxTemplate : public WikiMacro {
       templ.extractor()->Leave(&value);
 
       // Convert field value to LEX format.
-      Document document(store);
+      Document document(store, docnames_);
       document.SetText(value.text());
       GetTokenizer()->Tokenize(&document);
       value.AddToDocument(&document);
@@ -609,7 +612,7 @@ class InfoboxTemplate : public WikiMacro {
       } else {
         auto &group = groups[field->group];
         if (group.size() < index + 1) group.resize(index + 1);
-        auto &element = group[index];
+        Builder *&element = group[index];
         if (element == nullptr) element = new Builder(store);
         element->Add(field->key, lex);
       }
@@ -650,6 +653,9 @@ class InfoboxTemplate : public WikiMacro {
 
   // Infobox fields keyed by field name.
   std::unordered_map<string, Field> fields_;
+
+  // Document names.
+  DocumentNames *docnames_ = nullptr;
 };
 
 REGISTER_WIKI_MACRO("infobox", InfoboxTemplate);
diff --git a/sling/nlp/wiki/wiki.cc b/sling/nlp/wiki/wiki.cc
index 8826a33f..4729b25d 100644
--- a/sling/nlp/wiki/wiki.cc
+++ b/sling/nlp/wiki/wiki.cc
@@ -42,6 +42,7 @@ const char *kAliasSourceName[kNumAliasSources] = {
   "wikidata_native",
   "wikidata_demonym",
   "wikipedia_link",
+  "wikidata_name",
 };
 
 void Wiki::SplitTitle(const string &title,
diff --git a/sling/nlp/wiki/wiki.h b/sling/nlp/wiki/wiki.h
index 1aa39ab6..9dba523e 100644
--- a/sling/nlp/wiki/wiki.h
+++ b/sling/nlp/wiki/wiki.h
@@ -56,9 +56,10 @@ enum AliasSource {
   SRC_WIKIDATA_NATIVE          =  8,  //  256  0x0100
   SRC_WIKIDATA_DEMONYM         =  9,  //  512  0x0200
   SRC_WIKIPEDIA_LINK           = 10,  // 1024  0x0400
+  SRC_WIKIDATA_NAME            = 11,  // 2048  0x0800
 };
 
-static const int kNumAliasSources = 11;
+static const int kNumAliasSources = 12;
 
 extern const char *kAliasSourceName[kNumAliasSources];
 
diff --git a/sling/nlp/wiki/wikidata-converter.cc b/sling/nlp/wiki/wikidata-converter.cc
index bbc7eab1..7ab4b704 100644
--- a/sling/nlp/wiki/wikidata-converter.cc
+++ b/sling/nlp/wiki/wikidata-converter.cc
@@ -86,6 +86,7 @@ WikidataConverter::WikidataConverter(Store *commons, const string &language) {
     lang++;
   }
   language_map_["mul"] = n_lang_mul_.handle();
+  language_map_["zxx"] = n_lang_none_.handle();
 }
 
 Frame WikidataConverter::Convert(const Frame &item) {
@@ -141,7 +142,12 @@ Frame WikidataConverter::Convert(const Frame &item) {
       }
     }
     if (!label.IsNil()) builder.Add(n_name_, label);
-    if (!label_language.IsNil()) builder.Add(n_lang_, label_language);
+    if (!label_language.IsNil()) {
+      auto f = languages_.find(label_language);
+      if (f != languages_.end()) {
+        builder.Add(n_lang_, f->second.language);
+      }
+    }
   }
 
   // Pick description matching label language.
@@ -363,7 +369,9 @@ Handle WikidataConverter::ConvertText(const Frame &value) {
   string langid = value.GetString(s_language_);
   auto f = language_map_.find(langid);
   if (f == language_map_.end()) return Handle::nil();
-  if (f->second == n_lang_mul_) return text.handle();
+  if (f->second == n_lang_mul_ || f->second == n_lang_none_) {
+    return text.handle();
+  }
 
   // Convert text to string qualified by language.
   Builder monoling(store);
diff --git a/sling/nlp/wiki/wikidata-converter.h b/sling/nlp/wiki/wikidata-converter.h
index 08711ad9..bbef0bd4 100644
--- a/sling/nlp/wiki/wikidata-converter.h
+++ b/sling/nlp/wiki/wikidata-converter.h
@@ -138,6 +138,7 @@ class WikidataConverter {
   Name n_lng_{names_, "/w/lng"};
   Name n_globe_{names_, "/w/globe"};
   Name n_lang_mul_{names_, "/lang/mul"};
+  Name n_lang_none_{names_, "/lang/zxx"};
 
   Name n_alias_{names_, "alias"};
   Name n_sources_{names_, "sources"};
diff --git a/sling/nlp/wiki/wikipedia-documents.cc b/sling/nlp/wiki/wikipedia-documents.cc
index 55660f68..0db8aa72 100644
--- a/sling/nlp/wiki/wikipedia-documents.cc
+++ b/sling/nlp/wiki/wikipedia-documents.cc
@@ -213,12 +213,6 @@ class WikipediaDocumentBuilder : public task::FrameProcessor,
     annotator.AddToDocument(&document);
     num_article_tokens_->Increment(document.num_tokens());
     document.Update();
-
-    // Output intro text as alternative title alias.
-    WikiPlainTextSink intro;
-    if (extractor.ExtractIntro(&intro) && !intro.text().empty()) {
-      OutputAlias(qid, intro.text(), SRC_WIKIPEDIA_TITLE);
-    }
   }
 
   // Output alias for article title.
diff --git a/sling/pyapi/BUILD b/sling/pyapi/BUILD
index 4677936b..9e8ab870 100644
--- a/sling/pyapi/BUILD
+++ b/sling/pyapi/BUILD
@@ -42,12 +42,12 @@ cc_library(
     "//sling/myelin:compiler",
     "//sling/myelin:compute",
     "//sling/nlp/document",
-    "//sling/nlp/wiki:phrase-table",
     "//sling/nlp/document:document-tokenizer",
     "//sling/nlp/document:lex",
     "//sling/nlp/document:phrase-tokenizer",
     "//sling/nlp/kb:calendar",
     "//sling/nlp/kb:facts",
+    "//sling/nlp/kb:phrase-table",
     "//sling/nlp/parser",
     "//sling/nlp/parser/trainer:frame-evaluation",
     "//sling/nlp/wiki:wikidata-converter",
@@ -69,12 +69,13 @@ cc_library(
 cc_library(
   name = "tasks",
   deps = [
-    "//sling/nlp/wiki:profile-aliases",
+    "//sling/nlp/wiki:aliases",
     "//sling/nlp/wiki:wikidata-importer",
     "//sling/nlp/wiki:wikipedia-importer",
     "//sling/nlp/wiki:wikipedia-documents",
-    "//sling/nlp/wiki:name-table-builder",
-    "//sling/nlp/wiki:phrase-table-builder",
+
+    "//sling/nlp/kb:name-table-builder",
+    "//sling/nlp/kb:phrase-table-builder",
 
     "//sling/nlp/embedding:fact-embeddings",
     "//sling/nlp/embedding:word-embeddings",
diff --git a/sling/pyapi/pyphrase.h b/sling/pyapi/pyphrase.h
index dd4180bb..b249f2ea 100644
--- a/sling/pyapi/pyphrase.h
+++ b/sling/pyapi/pyphrase.h
@@ -16,7 +16,7 @@
 #define SLING_PYAPI_PYPHRASE_H_
 
 #include "sling/nlp/document/phrase-tokenizer.h"
-#include "sling/nlp/wiki/phrase-table.h"
+#include "sling/nlp/kb/phrase-table.h"
 #include "sling/pyapi/pybase.h"
 #include "sling/pyapi/pystore.h"
 
diff --git a/sling/pyapi/pytask.cc b/sling/pyapi/pytask.cc
index 8b887838..c1c346d4 100644
--- a/sling/pyapi/pytask.cc
+++ b/sling/pyapi/pytask.cc
@@ -294,7 +294,7 @@ PyMemberDef PyResource::members[] = {
 };
 
 void PyResource::Define(PyObject *module) {
-  InitType(&type, "sling.Resource", sizeof(PyResource), false);
+  InitType(&type, "sling.api.Resource", sizeof(PyResource), false);
   type.tp_init = method_cast<initproc>(&PyResource::Init);
   type.tp_dealloc = method_cast<destructor>(&PyResource::Dealloc);
   type.tp_members = members;
@@ -512,7 +512,7 @@ PyObject *PyStartTaskMonitor(PyObject *self, PyObject *args) {
   // Start HTTP server.
   bool start_http_server = false;
   if (http == nullptr) {
-    LOG(INFO) << "Start HTTP server in port " << port;
+    LOG(INFO) << "Start HTTP server on port " << port;
     HTTPServerOptions options;
     http = new HTTPServer(options, port);
     start_http_server = true;
diff --git a/sling/task/frames.cc b/sling/task/frames.cc
index 7b98d253..d22da977 100644
--- a/sling/task/frames.cc
+++ b/sling/task/frames.cc
@@ -19,6 +19,7 @@
 #include "sling/frame/decoder.h"
 #include "sling/frame/object.h"
 #include "sling/frame/reader.h"
+#include "sling/frame/serialization.h"
 #include "sling/frame/store.h"
 #include "sling/frame/wire.h"
 #include "sling/stream/file.h"
@@ -34,7 +35,7 @@ void FrameProcessor::Start(Task *task) {
 
   // Load commons store from file.
   for (Binding *binding : task->GetInputs("commons")) {
-    LoadStore(commons_, binding->resource());
+    LoadStore(binding->resource()->name(), commons_);
   }
 
   // Get output channel (optional).
@@ -151,23 +152,6 @@ Frame DecodeMessage(Store *store, Message *message) {
   }
 }
 
-void LoadStore(Store *store, Resource *file) {
-  store->LockGC();
-  FileInputStream stream(file->name());
-  Input input(&stream);
-  if (input.Peek() == WIRE_BINARY_MARKER) {
-    Decoder decoder(store, &input);
-    decoder.DecodeAll();
-  } else {
-    Reader reader(store, &input);
-    while (!reader.done()) {
-      reader.Read();
-      CHECK(!reader.error()) << reader.GetErrorMessage(file->name());
-    }
-  }
-  store->UnlockGC();
-}
-
 }  // namespace task
 }  // namespace sling
 
diff --git a/sling/task/frames.h b/sling/task/frames.h
index 8d8f62a5..32337451 100644
--- a/sling/task/frames.h
+++ b/sling/task/frames.h
@@ -84,9 +84,6 @@ Message *CreateMessage(const Frame &frame, bool shallow = false);
 // Decode message as frame.
 Frame DecodeMessage(Store *store, Message *message);
 
-// Load repository into store from input file.
-void LoadStore(Store *store, Resource *file);
-
 }  // namespace task
 }  // namespace sling
 
diff --git a/sling/util/unicode.cc b/sling/util/unicode.cc
index af737144..8d1af38d 100644
--- a/sling/util/unicode.cc
+++ b/sling/util/unicode.cc
@@ -49,6 +49,7 @@ Normalization ParseNormalization(const string &spec) {
       case 'd': flags |= NORMALIZE_DIGITS; break;
       case 'p': flags |= NORMALIZE_PUNCTUATION; break;
       case 'w': flags |= NORMALIZE_WHITESPACE; break;
+      case 'n': flags |= NORMALIZE_NAME; break;
       default:
         LOG(FATAL) << "Unknown normalization specifier: " << spec;
     }
@@ -63,6 +64,7 @@ string NormalizationString(Normalization normalization) {
   if (normalization & NORMALIZE_DIGITS) str.push_back('d');
   if (normalization & NORMALIZE_PUNCTUATION) str.push_back('p');
   if (normalization & NORMALIZE_WHITESPACE) str.push_back('w');
+  if (normalization & NORMALIZE_NAME) str.push_back('n');
   return str;
 }
 
@@ -127,6 +129,10 @@ bool Unicode::IsPunctuation(int c) {
   return Is(c, CATMASK_PUNCTUATION);
 }
 
+bool Unicode::IsNamePunctuation(int c) {
+  return Is(c, CATMASK_NAME_PUNCTUATION) || c == '.';
+}
+
 int Unicode::ToLower(int c) {
   if (c & unicode_tab_mask) return c;
   return unicode_lower_tab[c];
@@ -151,6 +157,9 @@ int Unicode::Normalize(int c, int flags) {
   if (flags & NORMALIZE_PUNCTUATION) {
     if (IsPunctuation(c)) c = 0;
   }
+  if (flags & NORMALIZE_NAME) {
+    if (IsNamePunctuation(c)) c = 0;
+  }
   if (flags & NORMALIZE_WHITESPACE) {
     if (IsWhitespace(c)) c = 0;
   }
diff --git a/sling/util/unicode.h b/sling/util/unicode.h
index 60956798..6987d8bc 100644
--- a/sling/util/unicode.h
+++ b/sling/util/unicode.h
@@ -95,6 +95,10 @@ enum UnicodeCategoryMask {
       (1 << CHARCAT_LINE_SEPARATOR) |
       (1 << CHARCAT_PARAGRAPH_SEPARATOR),
 
+  // Name punctuation.
+  CATMASK_NAME_PUNCTUATION =
+    (1 << CHARCAT_DASH_PUNCTUATION),
+
   // Punctuation.
   CATMASK_PUNCTUATION =
     (1 << CHARCAT_DASH_PUNCTUATION) |
@@ -118,9 +122,10 @@ enum Normalization {
   NORMALIZE_DIGITS      = 0x04,  // replace all digits with 9
   NORMALIZE_PUNCTUATION = 0x08,  // remove punctuation
   NORMALIZE_WHITESPACE  = 0x10,  // remove whitespace
+  NORMALIZE_NAME   =      0x20,  // remove name punctuation (periods and dashes)
 
   // Default normalization.
-  NORMALIZE_DEFAULT = NORMALIZE_CASE | NORMALIZE_LETTERS | NORMALIZE_PUNCTUATION
+  NORMALIZE_DEFAULT = NORMALIZE_CASE | NORMALIZE_LETTERS | NORMALIZE_NAME
 };
 
 // Parse a list of normalization specifiers to a normalization bit mask.
@@ -130,6 +135,7 @@ enum Normalization {
 //   d: NORMALIZE_DIGITS
 //   p: NORMALIZE_PUNCTUATION
 //   w: NORMALIZE_WHITESPACE
+//   n: NORMALIZE_NAME
 Normalization ParseNormalization(const string &spec);
 
 // Return string with normalization specifiers for flags.
@@ -174,6 +180,9 @@ class Unicode {
    // Check if code point is punctuation.
    static bool IsPunctuation(int c);
 
+   // Check if code point is name punctuation.
+   static bool IsNamePunctuation(int c);
+
    // Convert code point to lower case.
    static int ToLower(int c);