diff --git a/docs/metadata-workflow.rst b/docs/metadata-workflow.rst
--- a/docs/metadata-workflow.rst
+++ b/docs/metadata-workflow.rst
@@ -263,7 +263,7 @@
 
     def normalize_license(self, s):
         if isinstance(s, str):
-            return {"@id": "https://spdx.org/licenses/" + s}
+            return rdflib.URIRef("https://spdx.org/licenses/" + s)
 
 This method will automatically get called by ``_translate_dict`` when it
 finds a ``license`` field in ``content_dict``.
diff --git a/mypy.ini b/mypy.ini
--- a/mypy.ini
+++ b/mypy.ini
@@ -26,5 +26,8 @@
 [mypy-pytest.*]
 ignore_missing_imports = True
 
+[mypy-rdflib.*]
+ignore_missing_imports = True
+
 [mypy-xmltodict.*]
 ignore_missing_imports = True
diff --git a/requirements.txt b/requirements.txt
--- a/requirements.txt
+++ b/requirements.txt
@@ -5,6 +5,7 @@
 # cf https://forge.softwareheritage.org/T3815
 frozendict != 2.1.2
 pyld
+rdflib
 sentry-sdk
 typing-extensions
 xmltodict
diff --git a/swh/indexer/codemeta.py b/swh/indexer/codemeta.py
--- a/swh/indexer/codemeta.py
+++ b/swh/indexer/codemeta.py
@@ -1,4 +1,4 @@
-# Copyright (C) 2018  The Software Heritage developers
+# Copyright (C) 2018-2022  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
@@ -12,6 +12,7 @@
 from typing import Any, List
 
 from pyld import jsonld
+import rdflib
 
 import swh.indexer
 from swh.indexer.namespaces import ACTIVITYSTREAMS, CODEMETA, FORGEFED, SCHEMA
@@ -61,7 +62,7 @@
     uri = jsonld.JsonLdProcessor.get_context_value(
         _PROCESSED_CODEMETA_CONTEXT, local_name, "@id"
     )
-    assert uri.startswith(("@", CODEMETA._uri, SCHEMA._uri)), (local_name, uri)
+    assert uri.startswith(("@", CODEMETA, SCHEMA)), (local_name, uri)
     return uri
 
 
@@ -82,7 +83,7 @@
         if not local_name:
             continue
         canonical_name = make_absolute_uri(local_name)
-        if canonical_name in PROPERTY_BLACKLIST:
+        if rdflib.URIRef(canonical_name) in PROPERTY_BLACKLIST:
             continue
         terms.add(canonical_name)
         for (col, value) in zip(header, line):  # For each cell in the row
@@ -92,7 +93,9 @@
                     # For each of the data source's properties that maps
                     # to this canonical name
                     if local_name.strip():
-                        codemeta_translation[col][local_name.strip()] = canonical_name
+                        codemeta_translation[col][local_name.strip()] = rdflib.URIRef(
+                            canonical_name
+                        )
 
     return (terms, codemeta_translation)
 
@@ -112,10 +115,10 @@
             "documentUrl": url,
             "document": CODEMETA_CONTEXT,
         }
-    elif url == CODEMETA._uri:
+    elif url == CODEMETA:
         raise Exception(
             "{} is CodeMeta's URI, use {} as context url".format(
-                CODEMETA._uri, CODEMETA_CONTEXT_URL
+                CODEMETA, CODEMETA_CONTEXT_URL
             )
         )
     else:
@@ -132,7 +135,7 @@
     """
     contexts: List[Any] = [CODEMETA_CONTEXT_URL]
     if forgefed:
-        contexts.append({"as": ACTIVITYSTREAMS._uri, "forge": FORGEFED._uri})
+        contexts.append({"as": str(ACTIVITYSTREAMS), "forge": str(FORGEFED)})
     return jsonld.compact(doc, contexts, options={"documentLoader": _document_loader})
 
 
@@ -141,40 +144,6 @@
     return jsonld.expand(doc, options={"documentLoader": _document_loader})
 
 
-def merge_values(v1, v2):
-    """If v1 and v2 are of the form `{"@list": l1}` and `{"@list": l2}`,
-    returns `{"@list": l1 + l2}`.
-    Otherwise, make them lists (if they are not already) and concatenate
-    them.
-
-    >>> merge_values('a', 'b')
-    ['a', 'b']
-    >>> merge_values(['a', 'b'], 'c')
-    ['a', 'b', 'c']
-    >>> merge_values({'@list': ['a', 'b']}, {'@list': ['c']})
-    {'@list': ['a', 'b', 'c']}
-    """
-    if v1 is None:
-        return v2
-    elif v2 is None:
-        return v1
-    elif isinstance(v1, dict) and set(v1) == {"@list"}:
-        assert isinstance(v1["@list"], list)
-        if isinstance(v2, dict) and set(v2) == {"@list"}:
-            assert isinstance(v2["@list"], list)
-            return {"@list": v1["@list"] + v2["@list"]}
-        else:
-            raise ValueError("Cannot merge %r and %r" % (v1, v2))
-    else:
-        if isinstance(v2, dict) and "@list" in v2:
-            raise ValueError("Cannot merge %r and %r" % (v1, v2))
-        if not isinstance(v1, list):
-            v1 = [v1]
-        if not isinstance(v2, list):
-            v2 = [v2]
-        return v1 + v2
-
-
 def merge_documents(documents):
     """Takes a list of metadata dicts, each generated from a different
     metadata file, and merges them.
diff --git a/swh/indexer/metadata_dictionary/base.py b/swh/indexer/metadata_dictionary/base.py
--- a/swh/indexer/metadata_dictionary/base.py
+++ b/swh/indexer/metadata_dictionary/base.py
@@ -6,14 +6,17 @@
 import json
 import logging
 from typing import Any, Callable, Dict, List, Optional, Tuple, TypeVar
+import uuid
 import xml.parsers.expat
 
+from pyld import jsonld
+import rdflib
 from typing_extensions import TypedDict
 import xmltodict
 import yaml
 
-from swh.indexer.codemeta import compact, merge_values
-from swh.indexer.namespaces import SCHEMA
+from swh.indexer.codemeta import _document_loader, compact
+from swh.indexer.namespaces import RDF, SCHEMA
 from swh.indexer.storage.interface import Sha1
 
 
@@ -25,7 +28,8 @@
 
 
 TTranslateCallable = TypeVar(
-    "TTranslateCallable", bound=Callable[[Any, Dict[str, Any], Any], None]
+    "TTranslateCallable",
+    bound=Callable[[Any, rdflib.Graph, rdflib.term.BNode, Any], None],
 )
 
 
@@ -145,7 +149,7 @@
     def supported_terms(cls):
         # one-to-one mapping from the original key to a CodeMeta term
         simple_terms = {
-            term
+            str(term)
             for (key, term) in cls.mapping.items()
             if key in cls.string_fields
             or hasattr(cls, "normalize_" + cls._normalize_method_name(key))
@@ -153,7 +157,7 @@
 
         # more complex mapping from the original key to JSON-LD
         complex_terms = {
-            term
+            str(term)
             for meth_name in dir(cls)
             if meth_name.startswith("translate_")
             for term in getattr(getattr(cls, meth_name), "produced_terms", [])
@@ -174,7 +178,20 @@
             the indexer
 
         """
-        translated_metadata = {"@type": SCHEMA.SoftwareSourceCode}
+        graph = rdflib.Graph()
+
+        # The main object being described (the SoftwareSourceCode) does not necessarily
+        # may or may not have an id.
+        # Either way, we temporarily use this URI to identify it. Unfortunately,
+        # we cannot use a blank node as we need to use it for JSON-LD framing later,
+        # and blank nodes cannot be used for framing in JSON-LD >= 1.1
+        root_id = (
+            "https://www.softwareheritage.org/schema/2022/indexer/tmp-node/"
+            + str(uuid.uuid4())
+        )
+        root = rdflib.URIRef(root_id)
+        graph.add((root, RDF.type, SCHEMA.SoftwareSourceCode))
+
         for k, v in content_dict.items():
             # First, check if there is a specific translation
             # method for this key
@@ -182,40 +199,66 @@
                 self, "translate_" + self._normalize_method_name(k), None
             )
             if translation_method:
-                translation_method(translated_metadata, v)
+                translation_method(graph, root, v)
             elif k in self.mapping:
                 # if there is no method, but the key is known from the
                 # crosswalk table
                 codemeta_key = self.mapping[k]
 
-                # if there is a normalization method, use it on the value
+                # if there is a normalization method, use it on the value,
+                # and add its results to the triples
                 normalization_method = getattr(
                     self, "normalize_" + self._normalize_method_name(k), None
                 )
                 if normalization_method:
                     v = normalization_method(v)
+                    if v is None:
+                        pass
+                    elif isinstance(v, list):
+                        for item in reversed(v):
+                            graph.add((root, codemeta_key, item))
+                    else:
+                        graph.add((root, codemeta_key, v))
                 elif k in self.string_fields and isinstance(v, str):
-                    pass
+                    graph.add((root, codemeta_key, rdflib.Literal(v)))
                 elif k in self.string_fields and isinstance(v, list):
-                    v = [x for x in v if isinstance(x, str)]
+                    for item in v:
+                        graph.add((root, codemeta_key, rdflib.Literal(item)))
                 else:
                     continue
 
-                # set the translation metadata with the normalized value
-                if codemeta_key in translated_metadata:
-                    translated_metadata[codemeta_key] = merge_values(
-                        translated_metadata[codemeta_key], v
-                    )
-                else:
-                    translated_metadata[codemeta_key] = v
+        self.extra_translation(graph, root, content_dict)
+
+        # Convert from rdflib's internal graph representation to JSON
+        s = graph.serialize(format="application/ld+json")
+
+        # Load from JSON to a list of Python objects
+        jsonld_graph = json.loads(s)
+
+        # Use JSON-LD framing to turn the graph into a rooted tree
+        # frame = {"@type": str(SCHEMA.SoftwareSourceCode)}
+        translated_metadata = jsonld.frame(
+            jsonld_graph,
+            {"@id": root_id},
+            options={
+                "documentLoader": _document_loader,
+                "processingMode": "json-ld-1.1",
+            },
+        )
 
-        self.extra_translation(translated_metadata, content_dict)
+        # Remove the temporary id we added at the beginning
+        if isinstance(translated_metadata["@id"], list):
+            translated_metadata["@id"].remove(root_id)
+        else:
+            del translated_metadata["@id"]
 
         return self.normalize_translation(translated_metadata)
 
-    def extra_translation(self, translated_metadata: Dict[str, Any], d: Dict[str, Any]):
-        """Called at the end of the translation process, and may add arbitrary keys
-        to ``translated_metadata`` based on the input dictionary (passed as ``d``).
+    def extra_translation(
+        self, graph: rdflib.Graph, root: rdflib.term.Node, d: Dict[str, Any]
+    ):
+        """Called at the end of the translation process, and may add arbitrary triples
+        to ``graph`` based on the input dictionary (passed as ``d``).
         """
         pass
 
diff --git a/swh/indexer/metadata_dictionary/cff.py b/swh/indexer/metadata_dictionary/cff.py
--- a/swh/indexer/metadata_dictionary/cff.py
+++ b/swh/indexer/metadata_dictionary/cff.py
@@ -1,9 +1,21 @@
-from typing import Dict, List, Optional, Union
+# Copyright (C) 2021-2022  The Software Heritage developers
+# See the AUTHORS file at the top-level directory of this distribution
+# License: GNU General Public License version 3, or any later version
+# See top-level LICENSE file for more information
+
+from typing import List
+
+from rdflib import BNode, Graph, Literal, URIRef
+import rdflib.term
 
 from swh.indexer.codemeta import CROSSWALK_TABLE
-from swh.indexer.namespaces import SCHEMA
+from swh.indexer.namespaces import RDF, SCHEMA
 
 from .base import YamlMapping
+from .utils import add_map
+
+DOI = URIRef("https://doi.org/")
+SPDX = URIRef("https://spdx.org/licenses/")
 
 
 class CffMapping(YamlMapping):
@@ -14,41 +26,41 @@
     mapping = CROSSWALK_TABLE["Citation File Format Core (CFF-Core) 1.0.2"]
     string_fields = ["keywords", "license", "abstract", "version", "doi"]
 
-    def normalize_authors(self, d: List[dict]) -> Dict[str, list]:
-        result = []
-        for author in d:
-            author_data: Dict[str, Optional[Union[str, Dict]]] = {
-                "@type": SCHEMA.Person
-            }
-            if "orcid" in author and isinstance(author["orcid"], str):
-                author_data["@id"] = author["orcid"]
-            if "affiliation" in author and isinstance(author["affiliation"], str):
-                author_data[SCHEMA.affiliation] = {
-                    "@type": SCHEMA.Organization,
-                    SCHEMA.name: author["affiliation"],
-                }
-            if "family-names" in author and isinstance(author["family-names"], str):
-                author_data[SCHEMA.familyName] = author["family-names"]
-            if "given-names" in author and isinstance(author["given-names"], str):
-                author_data[SCHEMA.givenName] = author["given-names"]
-
-            result.append(author_data)
-
-        result_final = {"@list": result}
-        return result_final
-
-    def normalize_doi(self, s: str) -> Dict[str, str]:
+    def _translate_author(self, graph: Graph, author: dict) -> rdflib.term.Node:
+        node: rdflib.term.Node
+        if "orcid" in author and isinstance(author["orcid"], str):
+            node = URIRef(author["orcid"])
+        else:
+            node = BNode()
+        graph.add((node, RDF.type, SCHEMA.Person))
+        if "affiliation" in author and isinstance(author["affiliation"], str):
+            affiliation = BNode()
+            graph.add((node, SCHEMA.affiliation, affiliation))
+            graph.add((affiliation, RDF.type, SCHEMA.Organization))
+            graph.add((affiliation, SCHEMA.name, Literal(author["affiliation"])))
+        if "family-names" in author and isinstance(author["family-names"], str):
+            graph.add((node, SCHEMA.familyName, Literal(author["family-names"])))
+        if "given-names" in author and isinstance(author["given-names"], str):
+            graph.add((node, SCHEMA.givenName, Literal(author["given-names"])))
+        return node
+
+    def translate_authors(
+        self, graph: Graph, root: URIRef, authors: List[dict]
+    ) -> None:
+        add_map(graph, root, SCHEMA.author, self._translate_author, authors)
+
+    def normalize_doi(self, s: str) -> URIRef:
         if isinstance(s, str):
-            return {"@id": "https://doi.org/" + s}
+            return DOI + s
 
-    def normalize_license(self, s: str) -> Dict[str, str]:
+    def normalize_license(self, s: str) -> URIRef:
         if isinstance(s, str):
-            return {"@id": "https://spdx.org/licenses/" + s}
+            return SPDX + s
 
-    def normalize_repository_code(self, s: str) -> Dict[str, str]:
+    def normalize_repository_code(self, s: str) -> URIRef:
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
-    def normalize_date_released(self, s: str) -> Dict[str, str]:
+    def normalize_date_released(self, s: str) -> Literal:
         if isinstance(s, str):
-            return {"@value": s, "@type": SCHEMA.Date}
+            return Literal(s, datatype=SCHEMA.Date)
diff --git a/swh/indexer/metadata_dictionary/composer.py b/swh/indexer/metadata_dictionary/composer.py
--- a/swh/indexer/metadata_dictionary/composer.py
+++ b/swh/indexer/metadata_dictionary/composer.py
@@ -4,11 +4,18 @@
 # See top-level LICENSE file for more information
 
 import os.path
+from typing import Optional
+
+from rdflib import BNode, Graph, Literal, URIRef
 
 from swh.indexer.codemeta import _DATA_DIR, _read_crosstable
-from swh.indexer.namespaces import SCHEMA
+from swh.indexer.namespaces import RDF, SCHEMA
 
 from .base import JsonMapping, SingleFileIntrinsicMapping
+from .utils import add_map
+
+SPDX = URIRef("https://spdx.org/licenses/")
+
 
 COMPOSER_TABLE_PATH = os.path.join(_DATA_DIR, "composer.csv")
 
@@ -35,23 +42,24 @@
 
     def normalize_homepage(self, s):
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
     def normalize_license(self, s):
         if isinstance(s, str):
-            return {"@id": "https://spdx.org/licenses/" + s}
+            return SPDX + s
 
-    def normalize_authors(self, author_list):
-        authors = []
-        for author in author_list:
-            author_obj = {"@type": SCHEMA.Person}
+    def _translate_author(self, graph: Graph, author) -> Optional[BNode]:
+        if not isinstance(author, dict):
+            return None
+        node = BNode()
+        graph.add((node, RDF.type, SCHEMA.Person))
 
-            if isinstance(author, dict):
-                if isinstance(author.get("name", None), str):
-                    author_obj[SCHEMA.name] = author.get("name", None)
-                if isinstance(author.get("email", None), str):
-                    author_obj[SCHEMA.email] = author.get("email", None)
+        if isinstance(author.get("name"), str):
+            graph.add((node, SCHEMA.name, Literal(author["name"])))
+        if isinstance(author.get("email"), str):
+            graph.add((node, SCHEMA.email, Literal(author["email"])))
 
-                authors.append(author_obj)
+        return node
 
-        return {"@list": authors}
+    def translate_authors(self, graph: Graph, root: URIRef, authors) -> None:
+        add_map(graph, root, SCHEMA.author, self._translate_author, authors)
diff --git a/swh/indexer/metadata_dictionary/dart.py b/swh/indexer/metadata_dictionary/dart.py
--- a/swh/indexer/metadata_dictionary/dart.py
+++ b/swh/indexer/metadata_dictionary/dart.py
@@ -6,10 +6,15 @@
 import os.path
 import re
 
+from rdflib import RDF, BNode, Graph, Literal, URIRef
+
 from swh.indexer.codemeta import _DATA_DIR, _read_crosstable
 from swh.indexer.namespaces import SCHEMA
 
 from .base import YamlMapping
+from .utils import add_map
+
+SPDX = URIRef("https://spdx.org/licenses/")
 
 PUB_TABLE_PATH = os.path.join(_DATA_DIR, "pubspec.csv")
 
@@ -43,33 +48,32 @@
 
     def normalize_license(self, s):
         if isinstance(s, str):
-            return {"@id": "https://spdx.org/licenses/" + s}
+            return SPDX + s
 
     def normalize_homepage(self, s):
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
-    def normalize_author(self, s):
-        name_email_regex = "(?P<name>.*?)( <(?P<email>.*)>)"
-        author = {"@type": SCHEMA.Person}
+    def _translate_author(self, graph, s):
+        name_email_re = re.compile("(?P<name>.*?)( <(?P<email>.*)>)")
         if isinstance(s, str):
-            match = re.search(name_email_regex, s)
+            author = BNode()
+            graph.add((author, RDF.type, SCHEMA.Person))
+            match = name_email_re.search(s)
             if match:
                 name = match.group("name")
                 email = match.group("email")
-                author[SCHEMA.email] = email
+                graph.add((author, SCHEMA.email, Literal(email)))
             else:
                 name = s
 
-            author[SCHEMA.name] = name
+            graph.add((author, SCHEMA.name, Literal(name)))
 
-            return {"@list": [author]}
+            return author
 
-    def normalize_authors(self, authors_list):
-        authors = {"@list": []}
+    def translate_author(self, graph: Graph, root, s) -> None:
+        add_map(graph, root, SCHEMA.author, self._translate_author, [s])
 
-        if isinstance(authors_list, list):
-            for s in authors_list:
-                author = self.normalize_author(s)["@list"]
-                authors["@list"] += author
-            return authors
+    def translate_authors(self, graph: Graph, root, authors) -> None:
+        if isinstance(authors, list):
+            add_map(graph, root, SCHEMA.author, self._translate_author, authors)
diff --git a/swh/indexer/metadata_dictionary/github.py b/swh/indexer/metadata_dictionary/github.py
--- a/swh/indexer/metadata_dictionary/github.py
+++ b/swh/indexer/metadata_dictionary/github.py
@@ -3,17 +3,17 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-import json
-from typing import Any, Dict, Tuple
+from typing import Any, Tuple
+
+from rdflib import RDF, BNode, Graph, Literal, URIRef
 
 from swh.indexer.codemeta import CROSSWALK_TABLE
-from swh.indexer.namespaces import ACTIVITYSTREAMS, FORGEFED
+from swh.indexer.namespaces import ACTIVITYSTREAMS, FORGEFED, SCHEMA
 
 from .base import BaseExtrinsicMapping, JsonMapping, produce_terms
+from .utils import prettyprint_graph  # noqa
 
-
-def _prettyprint(d):
-    print(json.dumps(d, indent=4))
+SPDX = URIRef("https://spdx.org/licenses/")
 
 
 class GitHubMapping(BaseExtrinsicMapping, JsonMapping):
@@ -33,94 +33,81 @@
     def extrinsic_metadata_formats(cls) -> Tuple[str, ...]:
         return ("application/vnd.github.v3+json",)
 
-    def _translate_dict(self, content_dict: Dict[str, Any], **kwargs) -> Dict[str, Any]:
-        d = super()._translate_dict(content_dict, **kwargs)
-        d["type"] = FORGEFED.Repository
-        return d
+    def extra_translation(self, graph, root, content_dict):
+        graph.remove((root, RDF.type, SCHEMA.SoftwareSourceCode))
+        graph.add((root, RDF.type, FORGEFED.Repository))
 
     @produce_terms(FORGEFED.forks, ACTIVITYSTREAMS.totalItems)
-    def translate_forks_count(
-        self, translated_metadata: Dict[str, Any], v: Any
-    ) -> None:
+    def translate_forks_count(self, graph: Graph, root: BNode, v: Any) -> None:
         """
 
-        >>> translated_metadata = {}
-        >>> GitHubMapping().translate_forks_count(translated_metadata, 42)
-        >>> _prettyprint(translated_metadata)
+        >>> graph = Graph()
+        >>> root = URIRef("http://example.org/test-software")
+        >>> GitHubMapping().translate_forks_count(graph, root, 42)
+        >>> prettyprint_graph(graph, root)
         {
-            "https://forgefed.org/ns#forks": [
-                {
-                    "@type": "https://www.w3.org/ns/activitystreams#OrderedCollection",
-                    "https://www.w3.org/ns/activitystreams#totalItems": 42
-                }
-            ]
+            "@id": ...,
+            "https://forgefed.org/ns#forks": {
+                "@type": "https://www.w3.org/ns/activitystreams#OrderedCollection",
+                "https://www.w3.org/ns/activitystreams#totalItems": 42
+            }
         }
         """
         if isinstance(v, int):
-            translated_metadata.setdefault(FORGEFED.forks, []).append(
-                {
-                    "@type": ACTIVITYSTREAMS.OrderedCollection,
-                    ACTIVITYSTREAMS.totalItems: v,
-                }
-            )
+            collection = BNode()
+            graph.add((root, FORGEFED.forks, collection))
+            graph.add((collection, RDF.type, ACTIVITYSTREAMS.OrderedCollection))
+            graph.add((collection, ACTIVITYSTREAMS.totalItems, Literal(v)))
 
     @produce_terms(ACTIVITYSTREAMS.likes, ACTIVITYSTREAMS.totalItems)
-    def translate_stargazers_count(
-        self, translated_metadata: Dict[str, Any], v: Any
-    ) -> None:
+    def translate_stargazers_count(self, graph: Graph, root: BNode, v: Any) -> None:
         """
 
-        >>> translated_metadata = {}
-        >>> GitHubMapping().translate_stargazers_count(translated_metadata, 42)
-        >>> _prettyprint(translated_metadata)
+        >>> graph = Graph()
+        >>> root = URIRef("http://example.org/test-software")
+        >>> GitHubMapping().translate_stargazers_count(graph, root, 42)
+        >>> prettyprint_graph(graph, root)
         {
-            "https://www.w3.org/ns/activitystreams#likes": [
-                {
-                    "@type": "https://www.w3.org/ns/activitystreams#Collection",
-                    "https://www.w3.org/ns/activitystreams#totalItems": 42
-                }
-            ]
+            "@id": ...,
+            "https://www.w3.org/ns/activitystreams#likes": {
+                "@type": "https://www.w3.org/ns/activitystreams#Collection",
+                "https://www.w3.org/ns/activitystreams#totalItems": 42
+            }
         }
         """
         if isinstance(v, int):
-            translated_metadata.setdefault(ACTIVITYSTREAMS.likes, []).append(
-                {
-                    "@type": ACTIVITYSTREAMS.Collection,
-                    ACTIVITYSTREAMS.totalItems: v,
-                }
-            )
+            collection = BNode()
+            graph.add((root, ACTIVITYSTREAMS.likes, collection))
+            graph.add((collection, RDF.type, ACTIVITYSTREAMS.Collection))
+            graph.add((collection, ACTIVITYSTREAMS.totalItems, Literal(v)))
 
     @produce_terms(ACTIVITYSTREAMS.followers, ACTIVITYSTREAMS.totalItems)
-    def translate_watchers_count(
-        self, translated_metadata: Dict[str, Any], v: Any
-    ) -> None:
+    def translate_watchers_count(self, graph: Graph, root: BNode, v: Any) -> None:
         """
 
-        >>> translated_metadata = {}
-        >>> GitHubMapping().translate_watchers_count(translated_metadata, 42)
-        >>> _prettyprint(translated_metadata)
+        >>> graph = Graph()
+        >>> root = URIRef("http://example.org/test-software")
+        >>> GitHubMapping().translate_watchers_count(graph, root, 42)
+        >>> prettyprint_graph(graph, root)
         {
-            "https://www.w3.org/ns/activitystreams#followers": [
-                {
-                    "@type": "https://www.w3.org/ns/activitystreams#Collection",
-                    "https://www.w3.org/ns/activitystreams#totalItems": 42
-                }
-            ]
+            "@id": ...,
+            "https://www.w3.org/ns/activitystreams#followers": {
+                "@type": "https://www.w3.org/ns/activitystreams#Collection",
+                "https://www.w3.org/ns/activitystreams#totalItems": 42
+            }
         }
         """
         if isinstance(v, int):
-            translated_metadata.setdefault(ACTIVITYSTREAMS.followers, []).append(
-                {
-                    "@type": ACTIVITYSTREAMS.Collection,
-                    ACTIVITYSTREAMS.totalItems: v,
-                }
-            )
+            collection = BNode()
+            graph.add((root, ACTIVITYSTREAMS.followers, collection))
+            graph.add((collection, RDF.type, ACTIVITYSTREAMS.Collection))
+            graph.add((collection, ACTIVITYSTREAMS.totalItems, Literal(v)))
 
     def normalize_license(self, d):
         """
 
         >>> GitHubMapping().normalize_license({'spdx_id': 'MIT'})
-        {'@id': 'https://spdx.org/licenses/MIT'}
+        rdflib.term.URIRef('https://spdx.org/licenses/MIT')
         """
         if isinstance(d, dict) and isinstance(d.get("spdx_id"), str):
-            return {"@id": "https://spdx.org/licenses/" + d["spdx_id"]}
+            return SPDX + d["spdx_id"]
diff --git a/swh/indexer/metadata_dictionary/maven.py b/swh/indexer/metadata_dictionary/maven.py
--- a/swh/indexer/metadata_dictionary/maven.py
+++ b/swh/indexer/metadata_dictionary/maven.py
@@ -1,4 +1,4 @@
-# Copyright (C) 2018-2021  The Software Heritage developers
+# Copyright (C) 2018-2022  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
@@ -6,10 +6,13 @@
 import os
 from typing import Any, Dict
 
+from rdflib import Graph, Literal, URIRef
+
 from swh.indexer.codemeta import CROSSWALK_TABLE
 from swh.indexer.namespaces import SCHEMA
 
 from .base import SingleFileIntrinsicMapping, XmlMapping
+from .utils import prettyprint_graph  # noqa
 
 
 class MavenMapping(XmlMapping, SingleFileIntrinsicMapping):
@@ -27,14 +30,13 @@
     def _translate_dict(self, d: Dict[str, Any]) -> Dict[str, Any]:
         return super()._translate_dict(d.get("project") or {})
 
-    def extra_translation(self, translated_metadata, d):
-        repositories = self.parse_repositories(d)
-        if repositories:
-            translated_metadata[SCHEMA.codeRepository] = repositories
+    def extra_translation(self, graph: Graph, root, d):
+        self.parse_repositories(graph, root, d)
 
-    def parse_repositories(self, d):
+    def parse_repositories(self, graph: Graph, root, d):
         """https://maven.apache.org/pom.html#Repositories
 
+        >>> import rdflib
         >>> import xmltodict
         >>> from pprint import pprint
         >>> d = xmltodict.parse('''
@@ -47,21 +49,19 @@
         ...   </repository>
         ... </repositories>
         ... ''')
-        >>> MavenMapping().parse_repositories(d)
+        >>> MavenMapping().parse_repositories(rdflib.Graph(), rdflib.BNode(), d)
         """
         repositories = d.get("repositories")
         if not repositories:
-            results = [self.parse_repository(d, self._default_repository)]
+            self.parse_repository(graph, root, d, self._default_repository)
         elif isinstance(repositories, dict):
             repositories = repositories.get("repository") or []
             if not isinstance(repositories, list):
                 repositories = [repositories]
-            results = [self.parse_repository(d, repo) for repo in repositories]
-        else:
-            results = []
-        return [res for res in results if res] or None
+            for repo in repositories:
+                self.parse_repository(graph, root, d, repo)
 
-    def parse_repository(self, d, repo):
+    def parse_repository(self, graph: Graph, root, d, repo):
         if not isinstance(repo, dict):
             return
         if repo.get("layout", "default") != "default":
@@ -75,23 +75,18 @@
             and isinstance(artifact_id, str)
         ):
             repo = os.path.join(url, *group_id.split("."), artifact_id)
-            return {"@id": repo}
+            graph.add((root, SCHEMA.codeRepository, URIRef(repo)))
 
     def normalize_groupId(self, id_):
         """https://maven.apache.org/pom.html#Maven_Coordinates
 
         >>> MavenMapping().normalize_groupId('org.example')
-        {'@id': 'org.example'}
+        rdflib.term.Literal('org.example')
         """
         if isinstance(id_, str):
-            return {"@id": id_}
-
-    def translate_licenses(self, translated_metadata, d):
-        licenses = self.parse_licenses(d)
-        if licenses:
-            translated_metadata[SCHEMA.license] = licenses
+            return Literal(id_)
 
-    def parse_licenses(self, licenses):
+    def translate_licenses(self, graph, root, licenses):
         """https://maven.apache.org/pom.html#Licenses
 
         >>> import xmltodict
@@ -113,8 +108,16 @@
                 }
             }
         }
-        >>> MavenMapping().parse_licenses(d["licenses"])
-        [{'@id': 'https://www.apache.org/licenses/LICENSE-2.0.txt'}]
+        >>> graph = Graph()
+        >>> root = URIRef("http://example.org/test-software")
+        >>> MavenMapping().translate_licenses(graph, root, d["licenses"])
+        >>> prettyprint_graph(graph, root)
+        {
+            "@id": ...,
+            "http://schema.org/license": {
+                "@id": "https://www.apache.org/licenses/LICENSE-2.0.txt"
+            }
+        }
 
         or, if there are more than one license:
 
@@ -132,9 +135,16 @@
         ...   </license>
         ... </licenses>
         ... ''')
-        >>> pprint(MavenMapping().parse_licenses(d["licenses"]))
-        [{'@id': 'https://www.apache.org/licenses/LICENSE-2.0.txt'},
-         {'@id': 'https://opensource.org/licenses/MIT'}]
+        >>> graph = Graph()
+        >>> root = URIRef("http://example.org/test-software")
+        >>> MavenMapping().translate_licenses(graph, root, d["licenses"])
+        >>> pprint(set(graph.triples((root, URIRef("http://schema.org/license"), None))))
+        {(rdflib.term.URIRef('http://example.org/test-software'),
+          rdflib.term.URIRef('http://schema.org/license'),
+          rdflib.term.URIRef('https://opensource.org/licenses/MIT')),
+         (rdflib.term.URIRef('http://example.org/test-software'),
+          rdflib.term.URIRef('http://schema.org/license'),
+          rdflib.term.URIRef('https://www.apache.org/licenses/LICENSE-2.0.txt'))}
         """
 
         if not isinstance(licenses, dict):
@@ -144,8 +154,6 @@
             licenses = [licenses]
         elif not isinstance(licenses, list):
             return
-        return [
-            {"@id": license["url"]}
-            for license in licenses
-            if isinstance(license, dict) and isinstance(license.get("url"), str)
-        ] or None
+        for license in licenses:
+            if isinstance(license, dict) and isinstance(license.get("url"), str):
+                graph.add((root, SCHEMA.license, URIRef(license["url"])))
diff --git a/swh/indexer/metadata_dictionary/npm.py b/swh/indexer/metadata_dictionary/npm.py
--- a/swh/indexer/metadata_dictionary/npm.py
+++ b/swh/indexer/metadata_dictionary/npm.py
@@ -6,10 +6,15 @@
 import re
 import urllib.parse
 
+from rdflib import RDF, BNode, Graph, Literal, URIRef
+
 from swh.indexer.codemeta import CROSSWALK_TABLE
 from swh.indexer.namespaces import SCHEMA
 
 from .base import JsonMapping, SingleFileIntrinsicMapping
+from .utils import add_list, prettyprint_graph  # noqa
+
+SPDX = URIRef("https://spdx.org/licenses/")
 
 
 class NpmMapping(JsonMapping, SingleFileIntrinsicMapping):
@@ -38,13 +43,13 @@
         ...     'type': 'git',
         ...     'url': 'https://example.org/foo.git'
         ... })
-        {'@id': 'git+https://example.org/foo.git'}
+        rdflib.term.URIRef('git+https://example.org/foo.git')
         >>> NpmMapping().normalize_repository(
         ...     'gitlab:foo/bar')
-        {'@id': 'git+https://gitlab.com/foo/bar.git'}
+        rdflib.term.URIRef('git+https://gitlab.com/foo/bar.git')
         >>> NpmMapping().normalize_repository(
         ...     'foo/bar')
-        {'@id': 'git+https://github.com/foo/bar.git'}
+        rdflib.term.URIRef('git+https://github.com/foo/bar.git')
         """
         if (
             isinstance(d, dict)
@@ -67,7 +72,7 @@
         else:
             return None
 
-        return {"@id": url}
+        return URIRef(url)
 
     def normalize_bugs(self, d):
         """https://docs.npmjs.com/files/package.json#bugs
@@ -76,15 +81,15 @@
         ...     'url': 'https://example.org/bugs/',
         ...     'email': 'bugs@example.org'
         ... })
-        {'@id': 'https://example.org/bugs/'}
+        rdflib.term.URIRef('https://example.org/bugs/')
         >>> NpmMapping().normalize_bugs(
         ...     'https://example.org/bugs/')
-        {'@id': 'https://example.org/bugs/'}
+        rdflib.term.URIRef('https://example.org/bugs/')
         """
         if isinstance(d, dict) and isinstance(d.get("url"), str):
-            return {"@id": d["url"]}
+            return URIRef(d["url"])
         elif isinstance(d, str):
-            return {"@id": d}
+            return URIRef(d)
         else:
             return None
 
@@ -92,36 +97,75 @@
         r"^ *" r"(?P<name>.*?)" r"( +<(?P<email>.*)>)?" r"( +\((?P<url>.*)\))?" r" *$"
     )
 
-    def normalize_author(self, d):
+    def translate_author(self, graph: Graph, root, d):
         r"""https://docs.npmjs.com/files/package.json#people-fields-author-contributors'
 
         >>> from pprint import pprint
-        >>> pprint(NpmMapping().normalize_author({
+        >>> root = URIRef("http://example.org/test-software")
+        >>> graph = Graph()
+        >>> NpmMapping().translate_author(graph, root, {
         ...     'name': 'John Doe',
         ...     'email': 'john.doe@example.org',
         ...     'url': 'https://example.org/~john.doe',
-        ... }))
-        {'@list': [{'@type': 'http://schema.org/Person',
-                    'http://schema.org/email': 'john.doe@example.org',
-                    'http://schema.org/name': 'John Doe',
-                    'http://schema.org/url': {'@id': 'https://example.org/~john.doe'}}]}
-        >>> pprint(NpmMapping().normalize_author(
+        ... })
+        >>> prettyprint_graph(graph, root)
+        {
+            "@id": ...,
+            "http://schema.org/author": {
+                "@list": [
+                    {
+                        "@type": "http://schema.org/Person",
+                        "http://schema.org/email": "john.doe@example.org",
+                        "http://schema.org/name": "John Doe",
+                        "http://schema.org/url": {
+                            "@id": "https://example.org/~john.doe"
+                        }
+                    }
+                ]
+            }
+        }
+        >>> graph = Graph()
+        >>> NpmMapping().translate_author(graph, root,
         ...     'John Doe <john.doe@example.org> (https://example.org/~john.doe)'
-        ... ))
-        {'@list': [{'@type': 'http://schema.org/Person',
-                    'http://schema.org/email': 'john.doe@example.org',
-                    'http://schema.org/name': 'John Doe',
-                    'http://schema.org/url': {'@id': 'https://example.org/~john.doe'}}]}
-        >>> pprint(NpmMapping().normalize_author({
+        ... )
+        >>> prettyprint_graph(graph, root)
+        {
+            "@id": ...,
+            "http://schema.org/author": {
+                "@list": [
+                    {
+                        "@type": "http://schema.org/Person",
+                        "http://schema.org/email": "john.doe@example.org",
+                        "http://schema.org/name": "John Doe",
+                        "http://schema.org/url": {
+                            "@id": "https://example.org/~john.doe"
+                        }
+                    }
+                ]
+            }
+        }
+        >>> graph = Graph()
+        >>> NpmMapping().translate_author(graph, root, {
         ...     'name': 'John Doe',
         ...     'email': 'john.doe@example.org',
         ...     'url': 'https:\\\\example.invalid/~john.doe',
-        ... }))
-        {'@list': [{'@type': 'http://schema.org/Person',
-                    'http://schema.org/email': 'john.doe@example.org',
-                    'http://schema.org/name': 'John Doe'}]}
+        ... })
+        >>> prettyprint_graph(graph, root)
+        {
+            "@id": ...,
+            "http://schema.org/author": {
+                "@list": [
+                    {
+                        "@type": "http://schema.org/Person",
+                        "http://schema.org/email": "john.doe@example.org",
+                        "http://schema.org/name": "John Doe"
+                    }
+                ]
+            }
+        }
         """  # noqa
-        author = {"@type": SCHEMA.Person}
+        author = BNode()
+        graph.add((author, RDF.type, SCHEMA.Person))
         if isinstance(d, dict):
             name = d.get("name", None)
             email = d.get("email", None)
@@ -137,32 +181,32 @@
             return None
 
         if name and isinstance(name, str):
-            author[SCHEMA.name] = name
+            graph.add((author, SCHEMA.name, Literal(name)))
         if email and isinstance(email, str):
-            author[SCHEMA.email] = email
+            graph.add((author, SCHEMA.email, Literal(email)))
         if url and isinstance(url, str):
             # Workaround for https://github.com/digitalbazaar/pyld/issues/91 : drop
             # URLs that are blatantly invalid early, so PyLD does not crash.
             parsed_url = urllib.parse.urlparse(url)
             if parsed_url.netloc:
-                author[SCHEMA.url] = {"@id": url}
+                graph.add((author, SCHEMA.url, URIRef(url)))
 
-        return {"@list": [author]}
+        add_list(graph, root, SCHEMA.author, [author])
 
     def normalize_description(self, description):
         r"""Try to re-decode ``description`` as UTF-16, as this is a somewhat common
         mistake that causes issues in the database because of null bytes in JSON.
 
         >>> NpmMapping().normalize_description("foo bar")
-        'foo bar'
+        rdflib.term.Literal('foo bar')
         >>> NpmMapping().normalize_description(
         ...     "\ufffd\ufffd#\x00 \x00f\x00o\x00o\x00 \x00b\x00a\x00r\x00\r\x00 \x00"
         ... )
-        'foo bar'
+        rdflib.term.Literal('foo bar')
         >>> NpmMapping().normalize_description(
         ...     "\ufffd\ufffd\x00#\x00 \x00f\x00o\x00o\x00 \x00b\x00a\x00r\x00\r\x00 "
         ... )
-        'foo bar'
+        rdflib.term.Literal('foo bar')
         >>> NpmMapping().normalize_description(
         ...     # invalid UTF-16 and meaningless UTF-8:
         ...     "\ufffd\ufffd\x00#\x00\x00\x00 \x00\x00\x00\x00f\x00\x00\x00\x00"
@@ -213,32 +257,34 @@
             if description:
                 if description.startswith("# "):
                     description = description[2:]
-                return description.rstrip()
-        return description
+                return Literal(description.rstrip())
+            else:
+                return None
+        return Literal(description)
 
     def normalize_license(self, s):
         """https://docs.npmjs.com/files/package.json#license
 
         >>> NpmMapping().normalize_license('MIT')
-        {'@id': 'https://spdx.org/licenses/MIT'}
+        rdflib.term.URIRef('https://spdx.org/licenses/MIT')
         """
         if isinstance(s, str):
-            return {"@id": "https://spdx.org/licenses/" + s}
+            return SPDX + s
 
     def normalize_homepage(self, s):
         """https://docs.npmjs.com/files/package.json#homepage
 
         >>> NpmMapping().normalize_homepage('https://example.org/~john.doe')
-        {'@id': 'https://example.org/~john.doe'}
+        rdflib.term.URIRef('https://example.org/~john.doe')
         """
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
     def normalize_keywords(self, lst):
         """https://docs.npmjs.com/files/package.json#homepage
 
         >>> NpmMapping().normalize_keywords(['foo', 'bar'])
-        ['foo', 'bar']
+        [rdflib.term.Literal('foo'), rdflib.term.Literal('bar')]
         """
         if isinstance(lst, list):
-            return [x for x in lst if isinstance(x, str)]
+            return [Literal(x) for x in lst if isinstance(x, str)]
diff --git a/swh/indexer/metadata_dictionary/nuget.py b/swh/indexer/metadata_dictionary/nuget.py
--- a/swh/indexer/metadata_dictionary/nuget.py
+++ b/swh/indexer/metadata_dictionary/nuget.py
@@ -7,17 +7,22 @@
 import re
 from typing import Any, Dict, List
 
+from rdflib import RDF, BNode, Graph, Literal, URIRef
+
 from swh.indexer.codemeta import _DATA_DIR, _read_crosstable
 from swh.indexer.namespaces import SCHEMA
 from swh.indexer.storage.interface import Sha1
 
 from .base import BaseIntrinsicMapping, DirectoryLsEntry, XmlMapping
+from .utils import add_list
 
 NUGET_TABLE_PATH = os.path.join(_DATA_DIR, "nuget.csv")
 
 with open(NUGET_TABLE_PATH) as fd:
     (CODEMETA_TERMS, NUGET_TABLE) = _read_crosstable(fd)
 
+SPDX = URIRef("https://spdx.org/licenses/")
+
 
 class NuGetMapping(XmlMapping, BaseIntrinsicMapping):
     """
@@ -26,8 +31,8 @@
 
     name = "nuget"
     mapping = NUGET_TABLE["NuGet"]
-    mapping["copyright"] = "http://schema.org/copyrightNotice"
-    mapping["language"] = "http://schema.org/inLanguage"
+    mapping["copyright"] = URIRef("http://schema.org/copyrightNotice")
+    mapping["language"] = URIRef("http://schema.org/inLanguage")
     string_fields = [
         "description",
         "version",
@@ -53,12 +58,12 @@
 
     def normalize_projectUrl(self, s):
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
-    def translate_repository(self, translated_metadata, v):
+    def translate_repository(self, graph, root, v):
         if isinstance(v, dict) and isinstance(v["@url"], str):
-            codemeta_key = self.mapping["repository.url"]
-            translated_metadata[codemeta_key] = {"@id": v["@url"]}
+            codemeta_key = URIRef(self.mapping["repository.url"])
+            graph.add((root, codemeta_key, URIRef(v["@url"])))
 
     def normalize_license(self, v):
         if isinstance(v, dict) and v["@type"] == "expression":
@@ -67,7 +72,7 @@
                 re.search(r" with |\(|\)| and ", license_string, re.IGNORECASE)
             ):
                 return [
-                    {"@id": "https://spdx.org/licenses/" + license_type.strip()}
+                    SPDX + license_type.strip()
                     for license_type in re.split(
                         r" or ", license_string, flags=re.IGNORECASE
                     )
@@ -77,22 +82,23 @@
 
     def normalize_licenseUrl(self, s):
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
-    def normalize_authors(self, s):
+    def translate_authors(self, graph: Graph, root, s):
         if isinstance(s, str):
-            author_names = [a.strip() for a in s.split(",")]
-            authors = [
-                {"@type": SCHEMA.Person, SCHEMA.name: name} for name in author_names
-            ]
-            return {"@list": authors}
-
-    def translate_releaseNotes(self, translated_metadata, s):
+            authors = []
+            for author_name in s.split(","):
+                author_name = author_name.strip()
+                author = BNode()
+                graph.add((author, RDF.type, SCHEMA.Person))
+                graph.add((author, SCHEMA.name, Literal(author_name)))
+                authors.append(author)
+            add_list(graph, root, SCHEMA.author, authors)
+
+    def translate_releaseNotes(self, graph: Graph, root, s):
         if isinstance(s, str):
-            translated_metadata.setdefault("http://schema.org/releaseNotes", []).append(
-                s
-            )
+            graph.add((root, SCHEMA.releaseNotes, Literal(s)))
 
     def normalize_tags(self, s):
         if isinstance(s, str):
-            return s.split(" ")
+            return [Literal(tag) for tag in s.split(" ")]
diff --git a/swh/indexer/metadata_dictionary/python.py b/swh/indexer/metadata_dictionary/python.py
--- a/swh/indexer/metadata_dictionary/python.py
+++ b/swh/indexer/metadata_dictionary/python.py
@@ -1,16 +1,18 @@
-# Copyright (C) 2018-2019  The Software Heritage developers
+# Copyright (C) 2018-2021  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import email.parser
 import email.policy
-import itertools
+
+from rdflib import BNode, Literal, URIRef
 
 from swh.indexer.codemeta import CROSSWALK_TABLE
-from swh.indexer.namespaces import SCHEMA
+from swh.indexer.namespaces import RDF, SCHEMA
 
 from .base import DictMapping, SingleFileIntrinsicMapping
+from .utils import add_list
 
 _normalize_pkginfo_key = str.lower
 
@@ -54,25 +56,25 @@
                 d.setdefault(key, []).append(value)
         return self._translate_dict(d)
 
-    def extra_translation(self, translated_metadata, d):
-        author_name = translated_metadata.pop(SCHEMA.author, None)
-        author_email = translated_metadata.pop(SCHEMA.email, None)
-        if author_name or author_email:
-            translated_metadata[SCHEMA.author] = {
-                "@list": [
-                    {
-                        "@type": SCHEMA.Person,
-                        SCHEMA.name: author_name,
-                        SCHEMA.email: author_email,
-                    }
-                ]
-            }
+    def extra_translation(self, graph, root, d):
+        author_names = list(graph.triples((root, SCHEMA.author, None)))
+        author_emails = list(graph.triples((root, SCHEMA.email, None)))
+        graph.remove((root, SCHEMA.author, None))
+        graph.remove((root, SCHEMA.email, None))
+        if author_names or author_emails:
+            author = BNode()
+            graph.add((author, RDF.type, SCHEMA.Person))
+            for (_, _, author_name) in author_names:
+                graph.add((author, SCHEMA.name, author_name))
+            for (_, _, author_email) in author_emails:
+                graph.add((author, SCHEMA.email, author_email))
+            add_list(graph, root, SCHEMA.author, [author])
 
     def normalize_home_page(self, urls):
-        return [{"@id": url} for url in urls]
+        return [URIRef(url) for url in urls]
 
     def normalize_keywords(self, keywords):
-        return list(itertools.chain.from_iterable(s.split(" ") for s in keywords))
+        return [Literal(keyword) for s in keywords for keyword in s.split(" ")]
 
     def normalize_license(self, licenses):
-        return [{"@id": license} for license in licenses]
+        return [URIRef("https://spdx.org/licenses/" + license) for license in licenses]
diff --git a/swh/indexer/metadata_dictionary/ruby.py b/swh/indexer/metadata_dictionary/ruby.py
--- a/swh/indexer/metadata_dictionary/ruby.py
+++ b/swh/indexer/metadata_dictionary/ruby.py
@@ -8,19 +8,26 @@
 import re
 from typing import List
 
+from rdflib import RDF, BNode, Graph, Literal, URIRef
+
 from swh.indexer.codemeta import CROSSWALK_TABLE
 from swh.indexer.metadata_dictionary.base import DirectoryLsEntry
 from swh.indexer.namespaces import SCHEMA
 from swh.indexer.storage.interface import Sha1
 
 from .base import BaseIntrinsicMapping, DictMapping
+from .utils import add_map
+
+SPDX = URIRef("https://spdx.org/licenses/")
 
 
-def name_to_person(name):
-    return {
-        "@type": SCHEMA.Person,
-        SCHEMA.name: name,
-    }
+def name_to_person(graph: Graph, name):
+    if not isinstance(name, str):
+        return None
+    author = BNode()
+    graph.add((author, RDF.type, SCHEMA.Person))
+    graph.add((author, SCHEMA.name, Literal(name)))
+    return author
 
 
 class GemspecMapping(BaseIntrinsicMapping, DictMapping):
@@ -107,30 +114,20 @@
 
     def normalize_homepage(self, s):
         if isinstance(s, str):
-            return {"@id": s}
+            return URIRef(s)
 
     def normalize_license(self, s):
         if isinstance(s, str):
-            return [{"@id": "https://spdx.org/licenses/" + s}]
+            return SPDX + s
 
     def normalize_licenses(self, licenses):
         if isinstance(licenses, list):
-            return [
-                {"@id": "https://spdx.org/licenses/" + license}
-                for license in licenses
-                if isinstance(license, str)
-            ]
+            return [SPDX + license for license in licenses if isinstance(license, str)]
 
-    def normalize_author(self, author):
+    def translate_author(self, graph: Graph, root, author):
         if isinstance(author, str):
-            return {"@list": [name_to_person(author)]}
+            add_map(graph, root, SCHEMA.author, name_to_person, [author])
 
-    def normalize_authors(self, authors):
+    def translate_authors(self, graph: Graph, root, authors):
         if isinstance(authors, list):
-            return {
-                "@list": [
-                    name_to_person(author)
-                    for author in authors
-                    if isinstance(author, str)
-                ]
-            }
+            add_map(graph, root, SCHEMA.author, name_to_person, authors)
diff --git a/swh/indexer/metadata_dictionary/utils.py b/swh/indexer/metadata_dictionary/utils.py
new file mode 100644
--- /dev/null
+++ b/swh/indexer/metadata_dictionary/utils.py
@@ -0,0 +1,72 @@
+# Copyright (C) 2022  The Software Heritage developers
+# See the AUTHORS file at the top-level directory of this distribution
+# License: GNU General Public License version 3, or any later version
+# See top-level LICENSE file for more information
+
+
+import json
+from typing import Callable, Iterable, Optional, Sequence, TypeVar
+
+from pyld import jsonld
+from rdflib import RDF, Graph, URIRef
+import rdflib.term
+
+from swh.indexer.codemeta import _document_loader
+
+
+def prettyprint_graph(graph: Graph, root: URIRef):
+    s = graph.serialize(format="application/ld+json")
+    jsonld_graph = json.loads(s)
+    translated_metadata = jsonld.frame(
+        jsonld_graph,
+        {"@id": str(root)},
+        options={
+            "documentLoader": _document_loader,
+            "processingMode": "json-ld-1.1",
+        },
+    )
+    print(json.dumps(translated_metadata, indent=4))
+
+
+def add_list(
+    graph: Graph,
+    subject: rdflib.term.Node,
+    predicate: rdflib.term.Identifier,
+    objects: Sequence[rdflib.term.Node],
+) -> None:
+    """Adds triples to the ``graph`` so that they are equivalent to this
+    JSON-LD object::
+
+        {
+            "@id": subject,
+            predicate: {"@list": objects}
+        }
+
+    This is a naive implementation of
+    https://json-ld.org/spec/latest/json-ld-api/#list-to-rdf-conversion
+    """
+    # JSON-LD's @list is syntactic sugar for a linked list / chain in the RDF graph,
+    # which is what we are going to construct, starting from the end:
+    last_link: rdflib.term.Node
+    last_link = RDF.nil
+    for item in reversed(objects):
+        link = rdflib.BNode()
+        graph.add((link, RDF.first, item))
+        graph.add((link, RDF.rest, last_link))
+        last_link = link
+    graph.add((subject, predicate, last_link))
+
+
+TValue = TypeVar("TValue")
+
+
+def add_map(
+    graph: Graph,
+    subject: rdflib.term.Node,
+    predicate: rdflib.term.Identifier,
+    f: Callable[[Graph, TValue], Optional[rdflib.term.Node]],
+    values: Iterable[TValue],
+) -> None:
+    """Helper for :func:`add_list` that takes a mapper function ``f``."""
+    nodes = [f(graph, value) for value in values]
+    add_list(graph, subject, predicate, [node for node in nodes if node])
diff --git a/swh/indexer/namespaces.py b/swh/indexer/namespaces.py
--- a/swh/indexer/namespaces.py
+++ b/swh/indexer/namespaces.py
@@ -3,24 +3,8 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-
-class _Namespace:
-    """Handy class to get terms within a namespace by accessing them as attributes.
-
-    This is similar to `rdflib's namespaces
-    <https://rdflib.readthedocs.io/en/stable/namespaces_and_bindings.html>`__
-    """
-
-    def __init__(self, uri: str):
-        if not uri.endswith(("#", "/")):
-            # Sanity check, to make sure it doesn't end with an alphanumerical
-            # character, which is very likely to be invalid.
-            raise ValueError(f"Invalid trailing character for namespace URI: {uri}")
-        self._uri = uri
-
-    def __getattr__(self, term: str) -> str:
-        return self._uri + term
-
+from rdflib import Namespace as _Namespace
+from rdflib import RDF  # noqa
 
 SCHEMA = _Namespace("http://schema.org/")
 CODEMETA = _Namespace("https://codemeta.github.io/terms/")
diff --git a/swh/indexer/tests/metadata_dictionary/test_cff.py b/swh/indexer/tests/metadata_dictionary/test_cff.py
--- a/swh/indexer/tests/metadata_dictionary/test_cff.py
+++ b/swh/indexer/tests/metadata_dictionary/test_cff.py
@@ -1,4 +1,4 @@
-# Copyright (C) 2017-2022  The Software Heritage developers
+# Copyright (C) 2021-2022  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
@@ -44,6 +44,13 @@
         "utf-8"
     )
 
+    result = MAPPINGS["CffMapping"]().translate(content)
+    assert set(result.pop("keywords")) == {
+        "citation",
+        "bibliography",
+        "cff",
+        "CITATION.cff",
+    }
     expected = {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
@@ -76,12 +83,10 @@
 Citation File Format to various other formats such as BibTeX, EndNote, \
 RIS, schema.org, CodeMeta, and .zenodo.json.""",
         "identifier": "https://doi.org/10.5281/zenodo.1162057",
-        "keywords": ["citation", "bibliography", "cff", "CITATION.cff"],
         "license": "https://spdx.org/licenses/Apache-2.0",
         "version": "1.4.0-alpha0",
     }
 
-    result = MAPPINGS["CffMapping"]().translate(content)
     assert expected == result
 
 
diff --git a/swh/indexer/tests/metadata_dictionary/test_composer.py b/swh/indexer/tests/metadata_dictionary/test_composer.py
--- a/swh/indexer/tests/metadata_dictionary/test_composer.py
+++ b/swh/indexer/tests/metadata_dictionary/test_composer.py
@@ -60,11 +60,16 @@
 
     result = MAPPINGS["ComposerMapping"]().translate(raw_content)
 
+    assert set(result.pop("keywords")) == {
+        "polyfill",
+        "shim",
+        "compatibility",
+        "portable",
+    }, result
     expected = {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "symfony/polyfill-mbstring",
-        "keywords": ["polyfill", "shim", "compatibility", "portable"],
         "description": "Symfony polyfill for the Mbstring extension",
         "url": "https://symfony.com",
         "license": "https://spdx.org/licenses/MIT",
diff --git a/swh/indexer/tests/metadata_dictionary/test_dart.py b/swh/indexer/tests/metadata_dictionary/test_dart.py
--- a/swh/indexer/tests/metadata_dictionary/test_dart.py
+++ b/swh/indexer/tests/metadata_dictionary/test_dart.py
@@ -3,6 +3,8 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
+import pytest
+
 from swh.indexer.metadata_dictionary import MAPPINGS
 
 
@@ -41,17 +43,17 @@
 
     result = MAPPINGS["PubMapping"]().translate(raw_content)
 
+    assert set(result.pop("keywords")) == {
+        "polyfill",
+        "shim",
+        "compatibility",
+        "portable",
+        "mbstring",
+    }, result
     expected = {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "newtify",
-        "keywords": [
-            "polyfill",
-            "shim",
-            "compatibility",
-            "portable",
-            "mbstring",
-        ],
         "description": """Have you been turned into a newt?  Would you like to be? \
 This package can help. It has all of the \
 newt-transmogrification functionality you have been looking \
@@ -109,6 +111,7 @@
     assert result == expected
 
 
+@pytest.mark.xfail(reason="https://github.com/w3c/json-ld-api/issues/547")
 def test_normalize_author_authors_pubspec():
     raw_content = """
     authors:
diff --git a/swh/indexer/tests/metadata_dictionary/test_github.py b/swh/indexer/tests/metadata_dictionary/test_github.py
--- a/swh/indexer/tests/metadata_dictionary/test_github.py
+++ b/swh/indexer/tests/metadata_dictionary/test_github.py
@@ -120,7 +120,7 @@
     result = MAPPINGS["GitHubMapping"]().translate(content)
     assert result == {
         "@context": CONTEXT,
-        "type": "https://forgefed.org/ns#Repository",
+        "type": "forge:Repository",
         "forge:forks": {
             "as:totalItems": 1,
             "type": "as:OrderedCollection",
diff --git a/swh/indexer/tests/metadata_dictionary/test_maven.py b/swh/indexer/tests/metadata_dictionary/test_maven.py
--- a/swh/indexer/tests/metadata_dictionary/test_maven.py
+++ b/swh/indexer/tests/metadata_dictionary/test_maven.py
@@ -45,7 +45,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "license": "https://www.apache.org/licenses/LICENSE-2.0.txt",
         "codeRepository": ("http://repo1.maven.org/maven2/com/mycompany/app/my-app"),
@@ -167,7 +167,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
@@ -191,7 +191,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
@@ -211,7 +211,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
         ),
@@ -229,7 +229,7 @@
     assert result == {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
@@ -251,7 +251,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
@@ -288,7 +288,7 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
         "codeRepository": (
             "https://repo.maven.apache.org/maven2/com/mycompany/app/my-app"
@@ -336,20 +336,20 @@
       </licenses>
     </project>"""
     result = MAPPINGS["MavenMapping"]().translate(raw_content)
+    assert set(result.pop("license")) == {
+        "https://www.apache.org/licenses/LICENSE-2.0.txt",
+        "https://opensource.org/licenses/MIT",
+    }, result
+    assert set(result.pop("codeRepository")) == {
+        "http://repo1.maven.org/maven2/com/mycompany/app/my-app",
+        "http://example.org/maven2/com/mycompany/app/my-app",
+    }, result
     assert result == {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "Maven Default Project",
-        "identifier": "com.mycompany.app",
+        "schema:identifier": "com.mycompany.app",
         "version": "1.2.3",
-        "license": [
-            "https://www.apache.org/licenses/LICENSE-2.0.txt",
-            "https://opensource.org/licenses/MIT",
-        ],
-        "codeRepository": [
-            "http://repo1.maven.org/maven2/com/mycompany/app/my-app",
-            "http://example.org/maven2/com/mycompany/app/my-app",
-        ],
     }
 
 
diff --git a/swh/indexer/tests/metadata_dictionary/test_npm.py b/swh/indexer/tests/metadata_dictionary/test_npm.py
--- a/swh/indexer/tests/metadata_dictionary/test_npm.py
+++ b/swh/indexer/tests/metadata_dictionary/test_npm.py
@@ -147,12 +147,6 @@
                 "license": "https://spdx.org/licenses/Artistic-2.0",
                 "version": "5.0.3",
                 "name": "npm",
-                "keywords": [
-                    "install",
-                    "modules",
-                    "package manager",
-                    "package.json",
-                ],
                 "url": "https://docs.npmjs.com/",
             },
         ),
@@ -160,6 +154,7 @@
 
     for result in results:
         del result.tool["id"]
+        result.metadata.pop("keywords", None)
 
     # The assertion below returns False sometimes because of nested lists
     assert expected_results == results
diff --git a/swh/indexer/tests/metadata_dictionary/test_nuget.py b/swh/indexer/tests/metadata_dictionary/test_nuget.py
--- a/swh/indexer/tests/metadata_dictionary/test_nuget.py
+++ b/swh/indexer/tests/metadata_dictionary/test_nuget.py
@@ -35,7 +35,26 @@
             <file src="bin\\Debug\\*.dll" target="lib" />
         </files>
     </package>"""
+
     result = MAPPINGS["NuGetMapping"]().translate(raw_content)
+
+    assert set(result.pop("keywords")) == {
+        "python3",
+        "java",
+        "cpp",
+        "search-tag",
+    }, result
+
+    assert set(result.pop("license")) == {
+        "https://spdx.org/licenses/MIT",
+        "https://raw.github.com/timrwood/moment/master/LICENSE",
+    }, result
+
+    assert set(result.pop("description")) == {
+        "Sample exists only to show a sample .nuspec file.",
+        "Summary is being deprecated. Use description instead.",
+    }, result
+
     expected = {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
@@ -44,25 +63,11 @@
             {"type": "Person", "name": "Franck Halmaert"},
         ],
         "codeRepository": "https://github.com/NuGet/NuGet.Client.git",
-        "description": [
-            "Sample exists only to show a sample .nuspec file.",
-            "Summary is being deprecated. Use description instead.",
-        ],
-        "license": [
-            "https://spdx.org/licenses/MIT",
-            "https://raw.github.com/timrwood/moment/master/LICENSE",
-        ],
         "url": "http://example.org/",
         "version": "1.2.3",
         "schema:releaseNotes": (
             "See the [changelog](https://github.com/httpie/httpie/releases/tag/3.2.0)."
         ),
-        "keywords": [
-            "python3",
-            "java",
-            "cpp",
-            "search-tag",
-        ],
     }
 
     assert result == expected
@@ -114,13 +119,13 @@
         </files>
     </package>"""
     result = MAPPINGS["NuGetMapping"]().translate(raw_content)
+    assert set(result.pop("license")) == {
+        "https://spdx.org/licenses/BitTorrent-1.0",
+        "https://spdx.org/licenses/GPL-3.0-with-GCC-exception",
+    }
     expected = {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
-        "license": [
-            "https://spdx.org/licenses/BitTorrent-1.0",
-            "https://spdx.org/licenses/GPL-3.0-with-GCC-exception",
-        ],
     }
 
     assert result == expected
diff --git a/swh/indexer/tests/metadata_dictionary/test_python.py b/swh/indexer/tests/metadata_dictionary/test_python.py
--- a/swh/indexer/tests/metadata_dictionary/test_python.py
+++ b/swh/indexer/tests/metadata_dictionary/test_python.py
@@ -38,7 +38,7 @@
 Provides-Extra: testing
 """  # noqa
     result = MAPPINGS["PythonPkginfoMapping"]().translate(raw_content)
-    assert result["description"] == [
+    assert set(result.pop("description")) == {
         "Software Heritage core utilities",  # note the comma here
         "swh-core\n"
         "========\n"
@@ -49,8 +49,7 @@
         "- serialization\n"
         "- logging mechanism\n"
         "",
-    ], result
-    del result["description"]
+    }, result
     assert result == {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
@@ -91,11 +90,11 @@
 Keywords: foo bar baz
 """  # noqa
     result = MAPPINGS["PythonPkginfoMapping"]().translate(raw_content)
+    assert set(result.pop("keywords")) == {"foo", "bar", "baz"}, result
     assert result == {
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "foo",
-        "keywords": ["foo", "bar", "baz"],
     }
 
 
@@ -110,5 +109,5 @@
         "@context": "https://doi.org/10.5063/schema/codemeta-2.0",
         "type": "SoftwareSourceCode",
         "name": "foo",
-        "license": "MIT",
+        "license": "https://spdx.org/licenses/MIT",
     }
diff --git a/swh/indexer/tests/metadata_dictionary/test_ruby.py b/swh/indexer/tests/metadata_dictionary/test_ruby.py
--- a/swh/indexer/tests/metadata_dictionary/test_ruby.py
+++ b/swh/indexer/tests/metadata_dictionary/test_ruby.py
@@ -4,6 +4,7 @@
 # See top-level LICENSE file for more information
 
 from hypothesis import HealthCheck, given, settings, strategies
+import pytest
 
 from swh.indexer.metadata_dictionary import MAPPINGS
 
@@ -39,6 +40,7 @@
     }
 
 
+@pytest.mark.xfail(reason="https://github.com/w3c/json-ld-api/issues/547")
 def test_gemspec_two_author_fields():
     raw_content = b"""
 Gem::Specification.new do |s|
diff --git a/swh/indexer/tests/test_codemeta.py b/swh/indexer/tests/test_codemeta.py
--- a/swh/indexer/tests/test_codemeta.py
+++ b/swh/indexer/tests/test_codemeta.py
@@ -3,13 +3,11 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-import pytest
-
-from swh.indexer.codemeta import CROSSWALK_TABLE, merge_documents, merge_values
+from swh.indexer.codemeta import CROSSWALK_TABLE, merge_documents
 
 
 def test_crosstable():
-    assert CROSSWALK_TABLE["NodeJS"] == {
+    assert {k: str(v) for (k, v) in CROSSWALK_TABLE["NodeJS"].items()} == {
         "repository": "http://schema.org/codeRepository",
         "os": "http://schema.org/operatingSystem",
         "cpu": "http://schema.org/processorRequirements",
@@ -28,32 +26,6 @@
     }
 
 
-def test_merge_values():
-    assert merge_values("a", "b") == ["a", "b"]
-    assert merge_values(["a", "b"], "c") == ["a", "b", "c"]
-    assert merge_values("a", ["b", "c"]) == ["a", "b", "c"]
-
-    assert merge_values({"@list": ["a"]}, {"@list": ["b"]}) == {"@list": ["a", "b"]}
-    assert merge_values({"@list": ["a", "b"]}, {"@list": ["c"]}) == {
-        "@list": ["a", "b", "c"]
-    }
-
-    with pytest.raises(ValueError):
-        merge_values({"@list": ["a"]}, "b")
-    with pytest.raises(ValueError):
-        merge_values("a", {"@list": ["b"]})
-    with pytest.raises(ValueError):
-        merge_values({"@list": ["a"]}, ["b"])
-    with pytest.raises(ValueError):
-        merge_values(["a"], {"@list": ["b"]})
-
-    assert merge_values("a", None) == "a"
-    assert merge_values(["a", "b"], None) == ["a", "b"]
-    assert merge_values(None, ["b", "c"]) == ["b", "c"]
-    assert merge_values({"@list": ["a"]}, None) == {"@list": ["a"]}
-    assert merge_values(None, {"@list": ["a"]}) == {"@list": ["a"]}
-
-
 def test_merge_documents():
     """
     Test the creation of a coherent minimal metadata set