feat: add Italian TN/ITN with raw-field tagging

README.md

@@ -45,6 +45,7 @@ ITN converts spoken expressions into written forms, for example `十二` → `12
 | German | `de` | Yes | Yes | `german` |
 | Hindi | `hi` | Yes | Yes | `hindi` |
 | Hindi–English (code-mixed) | `hi_en` | No | Yes | `hindi_english` |
+| Italian | `it` | Yes | Yes | `italian` |
 | Japanese | `ja` | Yes | Yes | `japanese` |
 | Kinyarwanda | `rw` | Yes | Yes | `kinyarwanda` |
 | Korean | `ko` | Yes | Yes | `korean` |

itn/italian/data/electronic/domain.tsv

@@ -0,0 +1,14 @@
+.com	com
+.org	org
+.gov	gov
+.uk	UK
+.fr	FR
+.net	net
+.html	html
+.py	python
+.ru	RU
+.us	US
+.de	DE
+.it	IT
+.jpg	jpeg
+.edu	edu

itn/italian/data/electronic/server_name.tsv

@@ -0,0 +1,11 @@
+gmail
+nvidia
+outlook
+hotmail
+yahoo
+live
+yandex
+orange
+wanadoo
+web
+comcast

itn/italian/data/electronic/symbols.tsv

@@ -0,0 +1,22 @@
+.	punto
+-	trattino
+_	trattino basso
+!	punto esclamativo
+#	cancelletto
+$	dollaro
+%	percento
+&	e commerciale
+'	apostrofo
+*	asterisco
++	più
+/	slash
+=	uguale
+?	punto interrogativo
+^	esponente
+`	accento
+{	apri graffa
+|	barra verticale
+}	chiudi graffa
+~	tilde
+,	virgola
+:	due punti

itn/italian/data/measure/denominators.tsv

@@ -0,0 +1,4 @@
+s	secondo
+h	ora
+min	minuto
+d	giorno

itn/italian/data/measure/units.tsv

@@ -0,0 +1,51 @@
+km	chilometro	chilometri
+m	metro	metri
+cm	centimetro	centimetri
+mm	millimetro	millimetri
+dm	decimetro	decimetri
+μm	micrometro	micrometri
+nm	nanometro	nanometri
+ha	ettaro	ettari
+mi	miglio	miglia
+m²	metro quadrato	metri quadrati
+km²	chilometro quadrato	chilometri quadrati
+m³	metro cubo	metri cubi
+g	grammo	grammi
+mg	milligrammo	milligrammi
+μg	microgrammo	microgrammi
+kg	chilogrammo	chilogrammi
+kg	chilo	chili
+t	tonnellata	tonnellate
+l	litro	litri
+ml	millilitro	millilitri
+cl	centilitro	centilitri
+dl	decilitro	decilitri
+ms	millisecondo	millisecondi
+Hz	hertz	hertz
+kHz	kilohertz	kilohertz
+MHz	megahertz	megahertz
+GHz	gigahertz	gigahertz
+V	volt	volt
+mV	millivolt	millivolt
+kV	kilovolt	kilovolt
+W	watt	watt
+kW	kilowatt	kilowatt
+MW	megawatt	megawatt
+GW	gigawatt	gigawatt
+TW	terawatt	terawatt
+A	ampere	ampere
+mA	milliampere	milliampere
+Ω	ohm	ohm
+dB	decibel	decibel
+kB	kilobyte	kilobyte
+MB	megabyte	megabyte
+GB	gigabyte	gigabyte
+TB	terabyte	terabyte
+kcal	chilocaloria	chilocalorie
+cal	caloria	calorie
+bar	bar	bar
+°C	grado celsius	gradi celsius
+°F	grado fahrenheit	gradi fahrenheit
+K	kelvin	kelvin
+%	percento	percento
+%	per cento	per cento

itn/italian/data/money/currency.tsv

@@ -0,0 +1,11 @@
+€	euro	euro	centesimo	centesimi
+$	dollaro	dollari	centesimo	centesimi
+£	sterlina	sterline	penny	penny
+£	sterlina	sterline	penny	pence
+¥	yen	yen		
+₽	rublo	rubli	copeca	copeche
+₹	rupia	rupie	paisa	paise
+USD	dollaro statunitense	dollari statunitensi	centesimo	centesimi
+CAD	dollaro canadese	dollari canadesi	centesimo	centesimi
+HKD	dollaro di hong kong	dollari di hong kong	centesimo	centesimi
+CHF	franco svizzero	franchi svizzeri	centesimo	centesimi

itn/italian/data/numbers/digit.tsv

@@ -0,0 +1,9 @@
+1	uno
+2	due
+3	tre
+4	quattro
+5	cinque
+6	sei
+7	sette
+8	otto
+9	nove

itn/italian/data/numbers/hundreds.tsv

@@ -0,0 +1,9 @@
+1	cento
+2	duecento
+3	trecento
+4	quattrocento
+5	cinquecento
+6	seicento
+7	settecento
+8	ottocento
+9	novecento

itn/italian/data/numbers/quantities.tsv

@@ -0,0 +1,4 @@
+milione
+milioni
+miliardo
+miliardi

itn/italian/data/numbers/teen.tsv

@@ -0,0 +1,10 @@
+10	dieci
+11	undici
+12	dodici
+13	tredici
+14	quattordici
+15	quindici
+16	sedici
+17	diciassette
+18	diciotto
+19	diciannove

itn/italian/data/numbers/tens.tsv

@@ -0,0 +1,8 @@
+2	venti
+3	trenta
+4	quaranta
+5	cinquanta
+6	sessanta
+7	settanta
+8	ottanta
+9	novanta

itn/italian/data/whitelist.tsv

@@ -0,0 +1,19 @@
+dott.	dottor
+dott.ssa	dottoressa
+prof.	professore
+prof.ssa	professoressa
+sig.ra	signora
+sig.na	signorina
+avv.	avvocato
+arch.	architetto
+ing.	ingegnere
+rag.	ragioniere
+geom.	geometra
+on.	onorevole
+rev.	reverendo
+sen.	senatore
+ecc.	eccetera
+v.le	viale
+p.za	piazza
+dip.	dipartimento
+c.v.d.	come volevasi dimostrare

itn/italian/inverse_normalizer.py

@@ -0,0 +1,77 @@
+# Copyright (c) 2026 Zhendong Peng ([email protected])
+#
+# 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.
+
+import pynini
+from pynini.lib.pynutil import add_weight
+
+from itn.italian.rules import money
+from itn.italian.rules.cardinal import Cardinal
+from itn.italian.rules.decimal import Decimal
+from itn.italian.rules.electronic import Electronic
+from itn.italian.rules.measure import Measure
+from itn.italian.rules.money import Money
+from itn.italian.rules.punctuation import Punctuation
+from itn.italian.rules.time import Time
+from itn.italian.rules.whitelist import WhiteList
+from itn.italian.rules.word import Word
+from tn.processor import Processor, RuleSpec
+
+TOKEN_ORDERS = dict(money.TOKEN_ORDER)
+
+# Written punctuation that attaches to the previous / next token.
+CLOSING = ".,;:!?)}"
+OPENING = "({"
+
+
+class InverseNormalizer(Processor):
+
+    def __init__(self, cache_dir=None, overwrite_cache=False):
+        super().__init__(name="it_inverse_normalizer", ordertype="itn", token_orders=TOKEN_ORDERS)
+        self.build_fst("it_itn", cache_dir, overwrite_cache, {})
+
+    def build_tagger_and_verbalizer(self):
+        cardinal = Cardinal()
+        decimal = Decimal(cardinal)
+        punctuation = Punctuation()
+        word = Word()
+        rules = (
+            RuleSpec(WhiteList(), 1.01),
+            RuleSpec(Time(), 1.05),
+            RuleSpec(Money(cardinal, decimal), 1.07),
+            RuleSpec(decimal, 1.08),
+            RuleSpec(Measure(cardinal, decimal), 1.08),
+            RuleSpec(cardinal, 1.1),
+            RuleSpec(Electronic(), 1.1),
+            RuleSpec(word, 100),
+            RuleSpec(punctuation, 1.1, verbalize=False),
+        )
+        classify = self.tagger_union(rules)
+        # Punctuation splits off only when attached to another token.
+        punct = add_weight(punctuation.tagger, 1.1)
+        token = punct.star + classify + punct.star
+        self.tagger = self.DELETE_SPACE + token + (self.DELETE_EXTRA_SPACE + token).star + self.DELETE_SPACE
+        self.tagger |= word.add_tokens(word.tag_field("value", self.SPACE.plus))
+
+        def marks(chars):
+            graph = pynini.union(*[pynini.escape(char) for char in chars])
+            return punctuation.delete_tokens(punctuation.verbalize_field("value", graph))
+
+        other_marks = [mark for mark in "!#$%&'()*+,-./:;<=>?@^_`{|}~" if mark not in CLOSING + OPENING]
+        closing = marks(CLOSING)
+        opening = marks(OPENING)
+        other = self.verbalizer_union(rules) | marks(other_marks)
+        # Written Italian attaches closing marks to the previous token and
+        # opening marks to the next one: ``( cinque ) .`` -> ``(5).``.
+        item = opening.star + other | add_weight(opening.plus, 1.0)
+        self.verbalizer = (item | closing) + (closing | self.INSERT_SPACE + item).star

itn/italian/rules/cardinal.py

@@ -0,0 +1,100 @@
+# Copyright (c) 2026 Zhendong Peng ([email protected])
+#
+# 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.
+
+import pynini
+from pynini.lib.pynutil import delete, insert
+
+from itn.italian.rules.utils import load_rows, string_map, table_map
+from tn.processor import Processor
+
+
+def two_digit_pairs():
+    """Returns ``(written, spoken)`` pairs for 1..99 with two-digit written forms.
+
+    Tens elide their final vowel before ``uno``/``otto`` (``ventuno``,
+    ``ventotto``); compounds of ``tre`` accept both ``ventitre`` and
+    ``ventitré``.
+    """
+
+    units = {int(written): spoken for written, spoken in load_rows("italian/data/numbers/digit.tsv")}
+    pairs = [("0{}".format(value), spoken) for value, spoken in units.items()]
+    pairs += [(written, spoken) for written, spoken in load_rows("italian/data/numbers/teen.tsv")]
+    for written, ten in load_rows("italian/data/numbers/tens.tsv"):
+        pairs.append((written + "0", ten))
+        for value, unit in units.items():
+            stem = ten[:-1] if unit[0] in "uo" else ten
+            pairs.append((written + str(value), stem + unit))
+            if unit == "tre":
+                pairs.append((written + str(value), stem + "tré"))
+    return pairs
+
+
+class Cardinal(Processor):
+    """Cardinals, e.g. ``tremila cinquecento quarantaquattro`` -> ``3544``.
+
+    The tagger keeps the spoken words; the verbalizer writes the digits.
+    Components may be joined (``centocinquantacinque``) or spaced (``cento
+    cinquantacinque``). ``miliardo`` is 10^9; NeMo's TN style of counting
+    thousands of millions (``tremila ... milioni``) is accepted too.
+    """
+
+    def __init__(self):
+        super().__init__("cardinal", token_orders={})
+        self.build_graphs()
+        self.sign = pynini.cross("meno", "-")
+        sign_tag = (self.tag_field("negative", self.sign) + pynini.accep(" ")).ques
+        sign_verb = (self.verbalize_field("negative", self.sign) + delete(" ")).ques
+        self.tagger = self.add_tokens(sign_tag + self.tag_field("integer", self.graph))
+        self.verbalizer = self.delete_tokens(sign_verb + self.verbalize_field("integer", self.graph))
+
+    def build_graphs(self):
+        digit = self.DIGIT
+        join = delete(" ").ques
+        space = delete(" ")
+        two = string_map(two_digit_pairs())
+        hundreds = table_map("italian/data/numbers/hundreds.tsv")
+        # ``centotto``/``centottanta`` drop the final ``o`` of the hundred.
+        elided = string_map([(written, spoken[:-1]) for written, spoken in load_rows("italian/data/numbers/hundreds.tsv")])
+        two_ott = ("ott" + self.VSIGMA) @ two
+        three = (
+            insert("0") + two
+            | hundreds + insert("00")
+            | hundreds + join + two
+            | elided + two_ott
+        ).optimize()
+        three_no_one = (three @ (pynini.closure(digit, 3, 3) - "001")).optimize()
+        rest3 = insert("000") | join + three
+        six = (
+            insert("000") + three
+            | pynini.cross("mille", "001") + rest3
+            | three_no_one + join + delete("mila") + rest3
+        ).optimize()
+        rest6 = insert("000000") | space + six
+        millions = pynini.cross("un milione", "001") | three_no_one + delete(" milioni")
+        billions = pynini.cross("un miliardo", "001") | three_no_one + delete(" miliardi")
+        thousand_millions = six @ ((pynini.closure(digit, 3, 3) - "000") + pynini.closure(digit, 3, 3))
+        padded = (
+            billions + (space + millions | insert("000")) + rest6
+            | insert("000") + millions + rest6
+            | insert("000000") + six
+            | thousand_millions + delete(" milioni") + rest6
+        )
+        strip = delete("0").star + (digit - "0") + digit.star
+        self.number = (pynini.cross("zero", "0") | padded @ strip).optimize()
+        self.graph = self.number
+        self.two = two
+        # Unpadded readings below one thousand, e.g. for decimal digit groups.
+        self.below_thousand = (pynini.cross("zero", "0") | three @ strip).optimize()
+        # Counting contexts (measures, money) also say ``un``/``una`` for 1.
+        self.counted = (self.number | pynini.cross(pynini.union("un", "una"), "1")).optimize()