diff --git a/src/gramform/tagops.py b/src/gramform/tagops.py
new file mode 100644
index 0000000..a5220ad
--- /dev/null
+++ b/src/gramform/tagops.py
@@ -0,0 +1,229 @@
+# -*- coding: utf-8 -*-
+# emacs: -*- mode: python; py-indent-offset: 4; indent-tabs-mode: nil -*-
+# vi: set ft=python sts=4 ts=4 sw=4 et:
+"""
+Data tags
+~~~~~~~~~
+Transformations and grammar for operations on data tags.
+"""
+from dataclasses import dataclass, field
+from functools import reduce
+from typing import (
+    Any,
+    Callable,
+    Dict,
+    Literal,
+    Mapping,
+    Optional,
+    Sequence,
+    Tuple,
+)
+
+from .grammar import (
+    Grammar,
+    Grouping,
+    GroupingPool,
+    LeafInterpreter,
+    Literalisation,
+    TransformPool,
+    TransformPrimitive,
+)
+
+
+@dataclass(frozen=True)
+class DataTagGrammar(Grammar):
+    groupings: GroupingPool = GroupingPool(
+        Grouping(open='(', close=')'),
+    )
+    transforms: TransformPool = field(
+        default_factory=lambda: TransformPool(
+            UnionNode(),
+            IntersectionNode(),
+            ComplementNode(),
+            ExclusiveOrNode(),
+        )
+    )
+    whitespace: bool = False
+    default_interpreter: Optional[LeafInterpreter] = field(
+        default_factory=lambda: TagSelectInterpreter()
+    )
+    default_root_transform: Optional[TransformPrimitive] = field(
+        default_factory=lambda: ReturnSelected()
+    )
+
+
+@dataclass(frozen=True)
+class TagSelectInterpreter(LeafInterpreter):
+    def __call__(self, leaf: str) -> Callable:
+        def select_by_tag(
+            tags: Mapping[str, Sequence[str]],
+            keys: Sequence[str],
+        ) -> Tuple[Mapping[str, Any], Sequence[str]]:
+            return tags[leaf], keys
+
+        return select_by_tag
+
+
+@dataclass(frozen=True)
+class ReturnSelected(TransformPrimitive):
+    min_arity: int = 1
+    max_arity: int = 1
+    priority: int = float('inf')
+    associative: bool = False
+    commutative: bool = False
+    literals: Sequence[Literalisation] = ()
+
+    def __call__(self, *pparams, **params) -> Callable:
+        f = pparams[0]
+
+        def return_selected(
+            arg: Any,
+            **datatypes,
+        ) -> Mapping[str, Any]:
+            keys = set(datatypes.keys())
+            return {k: datatypes[k] for k in f(arg, keys)[0]}
+
+        return return_selected
+
+
+# ---------------------------------- Union --------------------------------- #
+
+
+@dataclass(frozen=True)
+class UnionInfixLiteralisation(Literalisation):
+    affix: Literal['prefix', 'suffix', 'infix', 'leaf'] = 'infix'
+    regex: str = r'\|'
+
+    def parse_params(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        return params
+
+
+@dataclass(frozen=True)
+class UnionNode(TransformPrimitive):
+    min_arity: int = 2
+    max_arity: int = float('inf')
+    priority: int = 4
+    associative: bool = True
+    commutative: bool = True
+    literals: Sequence[Literalisation] = (UnionInfixLiteralisation(),)
+
+    def ascend(self, *pparams, **params) -> Callable:
+        def union(
+            arg: Any,
+            keys: Sequence[str],
+        ) -> Tuple[Mapping[str, Any], Sequence[str]]:
+            arg = tuple(set(f(arg, keys)[0]) for f in pparams)
+            arg = reduce((lambda x, y: x | y), arg)
+            return arg, keys
+
+        return union
+
+
+# ------------------------------ Intersection ------------------------------ #
+
+
+@dataclass(frozen=True)
+class IntersectionInfixLiteralisation(Literalisation):
+    affix: Literal['prefix', 'suffix', 'infix', 'leaf'] = 'infix'
+    regex: str = r'\&'
+
+    def parse_params(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        return params
+
+
+@dataclass(frozen=True)
+class IntersectionNode(TransformPrimitive):
+    min_arity: int = 2
+    max_arity: int = float('inf')
+    priority: int = 2
+    associative: bool = True
+    commutative: bool = True
+    literals: Sequence[Literalisation] = (IntersectionInfixLiteralisation(),)
+
+    def ascend(self, *pparams, **params) -> Callable:
+        def intersection(
+            arg: Any,
+            keys: Sequence[str],
+        ) -> Tuple[Mapping[str, Any], Sequence[str]]:
+            arg = tuple(set(f(arg, keys)[0]) for f in pparams)
+            print(arg)
+            arg = reduce((lambda x, y: x & y), arg)
+            return arg, keys
+
+        return intersection
+
+
+# ------------------------------- Complement ------------------------------- #
+
+
+@dataclass(frozen=True)
+class ComplementPrefixExclLiteralisation(Literalisation):
+    affix: Literal['prefix', 'suffix', 'infix', 'leaf'] = 'prefix'
+    regex: str = r'\!'
+
+    def parse_params(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        return params
+
+
+@dataclass(frozen=True)
+class ComplementPrefixTildeLiteralisation(Literalisation):
+    affix: Literal['prefix', 'suffix', 'infix', 'leaf'] = 'prefix'
+    regex: str = r'\~'
+
+    def parse_params(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        return params
+
+
+@dataclass(frozen=True)
+class ComplementNode(TransformPrimitive):
+    min_arity: int = 1
+    max_arity: int = 1
+    priority: int = 1
+    literals: Sequence[Literalisation] = (
+        ComplementPrefixExclLiteralisation(),
+        ComplementPrefixTildeLiteralisation(),
+    )
+
+    def ascend(self, *pparams, **params) -> Callable:
+        f = pparams[0]
+
+        def complement(
+            arg: Any,
+            keys: Sequence[str],
+        ) -> Tuple[Mapping[str, Any], Sequence[str]]:
+            return set(keys) - set(f(arg, keys)[0]), keys
+
+        return complement
+
+
+# ------------------------------ Exclusive Or ------------------------------ #
+
+
+@dataclass(frozen=True)
+class ExclusiveOrInfixLiteralisation(Literalisation):
+    affix: Literal['prefix', 'suffix', 'infix', 'leaf'] = 'infix'
+    regex: str = r'\^'
+
+    def parse_params(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        return params
+
+
+@dataclass(frozen=True)
+class ExclusiveOrNode(TransformPrimitive):
+    min_arity: int = 2
+    max_arity: int = float('inf')
+    priority: int = 3
+    associative: bool = True
+    commutative: bool = True
+    literals: Sequence[Literalisation] = (ExclusiveOrInfixLiteralisation(),)
+
+    def ascend(self, *pparams, **params) -> Callable:
+        def xor(
+            arg: Any,
+            keys: Sequence[str],
+        ) -> Tuple[Mapping[str, Any], Sequence[str]]:
+            arg = tuple(set(f(arg, keys)[0]) for f in pparams)
+            arg = reduce((lambda x, y: x ^ y), arg)
+            return arg, keys
+
+        return xor
diff --git a/tests/test_tagops.py b/tests/test_tagops.py
new file mode 100644
index 0000000..7a3a8e4
--- /dev/null
+++ b/tests/test_tagops.py
@@ -0,0 +1,59 @@
+# -*- coding: utf-8 -*-
+# emacs: -*- mode: python; py-indent-offset: 4; indent-tabs-mode: nil -*-
+# vi: set ft=python sts=4 ts=4 sw=4 et:
+"""
+Unit tests for data tag grammar
+"""
+from pkg_resources import resource_filename
+from gramform.tagops import DataTagGrammar
+
+
+def dataset():
+    return {
+        'a': 'A',
+        'b': 'B',
+        'c': 'C',
+        'd': 'D',
+        'e': 'E',
+    }
+
+
+def tags():
+    return {
+        'a': 'a',
+        'b': 'b',
+        'c': 'c',
+        'd': 'd',
+        'e': 'e',
+        'ab': {'a', 'b'},
+        'bc': {'b', 'c'},
+        'cd': {'c', 'd'},
+        'de': {'d', 'e'},
+        'abc': {'a', 'b', 'c'},
+        'bcd': {'b', 'c', 'd'},
+        'cde': {'c', 'd', 'e'},
+        'abcd': {'a', 'b', 'c', 'd'},
+        'bcde': {'b', 'c', 'd', 'e'},
+        'abcde': {'a', 'b', 'c', 'd', 'e'},
+    }
+
+
+def test_tags():
+    grammar = DataTagGrammar()
+    f = grammar.compile('~a&bcd')
+    assert(set(f(tags(), **dataset()).keys()) == {'b', 'c', 'd'})
+
+    f = grammar.compile('~bcd|!a')
+    assert(set(f(tags(), **dataset()).keys()) == {'a', 'b', 'c', 'd', 'e'})
+
+    f = grammar.compile('~bcd&~a')
+    assert(set(f(tags(), **dataset()).keys()) == {'e'})
+
+    g = grammar.compile('~(bcd|a)')
+    assert(
+        set(f(tags(), **dataset()).keys()) ==
+        set(g(tags(), **dataset()).keys())
+    )
+
+    f = grammar.compile('~(abc^bcde)^bcd')
+    assert(set(f(tags(), **dataset()).keys()) == {'d'})