diff --git a/source/characterProcessing.py b/source/characterProcessing.py
index b725c79bd8c..9dc904d58c1 100644
--- a/source/characterProcessing.py
+++ b/source/characterProcessing.py
@@ -1,8 +1,8 @@
-#characterProcessing.py
-#A part of NonVisual Desktop Access (NVDA)
-#Copyright (C) 2010-2018 NV Access Limited, World Light Information Limited, Hong Kong Blind Union, Babbage B.V.
-#This file is covered by the GNU General Public License.
-#See the file COPYING for more details.
+# A part of NonVisual Desktop Access (NVDA)
+# Copyright (C) 2010-2020 NV Access Limited, World Light Information Limited,
+# Hong Kong Blind Union, Babbage B.V., Julien Cochuyt
+# This file is covered by the GNU General Public License.
+# See the file COPYING for more details.
 
 import time
 import os
@@ -130,6 +130,7 @@ def getCharacterDescription(locale,character):
 SYMLVL_MOST = 200
 SYMLVL_ALL = 300
 SYMLVL_CHAR = 1000
+SYMLVL_UNCHANGED = -1
 SPEECH_SYMBOL_LEVEL_LABELS = {
 	# Translators: The level at which the given symbol will be spoken.
 	SYMLVL_NONE: pgettext("symbolLevel", "none"),
diff --git a/source/config/configSpec.py b/source/config/configSpec.py
index 9e23fbe4fe7..8a5ce1f7145 100644
--- a/source/config/configSpec.py
+++ b/source/config/configSpec.py
@@ -1,6 +1,6 @@
 # -*- coding: UTF-8 -*-
 # A part of NonVisual Desktop Access (NVDA)
-# Copyright (C) 2006-2020 NV Access Limited, Babbage B.V., Davy Kager, Bill Dengler
+# Copyright (C) 2006-2020 NV Access Limited, Babbage B.V., Davy Kager, Bill Dengler, Julien Cochuyt
 # This file is covered by the GNU General Public License.
 # See the file COPYING for more details.
 
@@ -31,6 +31,8 @@
 	# The synthesizer to use
 	synth = string(default=auto)
 	symbolLevel = integer(default=100)
+	# Symbol Level when reviewing by word (unchanged if -1)
+	symbolLevelWord = integer(default=300)
 	trustVoiceLanguage = boolean(default=true)
 	includeCLDR = boolean(default=True)
 	beepSpeechModePitch = integer(default=10000,min=50,max=11025)
diff --git a/source/gui/settingsDialogs.py b/source/gui/settingsDialogs.py
index d4fc5899d84..6bc5ebdbda0 100644
--- a/source/gui/settingsDialogs.py
+++ b/source/gui/settingsDialogs.py
@@ -2,7 +2,8 @@
 # A part of NonVisual Desktop Access (NVDA)
 # Copyright (C) 2006-2020 NV Access Limited, Peter Vágner, Aleksey Sadovoy,
 # Rui Batista, Joseph Lee, Heiko Folkerts, Zahari Yurukov, Leonard de Ruijter,
-# Derek Riemer, Babbage B.V., Davy Kager, Ethan Holliger, Bill Dengler, Thomas Stivers
+# Derek Riemer, Babbage B.V., Davy Kager, Ethan Holliger, Bill Dengler, Thomas Stivers,
+# Julien Cochuyt
 # This file is covered by the GNU General Public License.
 # See the file COPYING for more details.
 import logging
@@ -1461,6 +1462,21 @@ def makeSettings(self, settingsSizer):
 			characterProcessing.CONFIGURABLE_SPEECH_SYMBOL_LEVELS.index(curLevel)
 		)
 
+		self.symbolLevelWord = settingsSizerHelper.addItem(
+			wx.CheckBox(
+				self,
+				# Translators: The label for a setting in the Speech category
+				label=_("Speak all punctuations and symbols when reviewing by &word"),
+				style=wx.CHK_3STATE
+			)
+		)
+		if config.conf["speech"]["symbolLevelWord"] == characterProcessing.SYMLVL_ALL:
+			self.symbolLevelWord.Set3StateValue(wx.CHK_CHECKED)
+		elif config.conf["speech"]["symbolLevelWord"] == characterProcessing.SYMLVL_UNCHANGED:
+			self.symbolLevelWord.Set3StateValue(wx.CHK_UNCHECKED)
+		else:
+			self.symbolLevelWord.Set3StateValue(wx.CHK_UNDETERMINED)
+
 		# Translators: This is the label for a checkbox in the
 		# voice settings panel (if checked, text will be read using the voice for the language of the text).
 		trustVoiceLanguageText = _("Trust voice's language when processing characters and symbols")
@@ -1548,8 +1564,14 @@ def onSave(self):
 
 		config.conf["speech"]["autoLanguageSwitching"] = self.autoLanguageSwitchingCheckbox.IsChecked()
 		config.conf["speech"]["autoDialectSwitching"] = self.autoDialectSwitchingCheckbox.IsChecked()
-		config.conf["speech"]["symbolLevel"]=characterProcessing.CONFIGURABLE_SPEECH_SYMBOL_LEVELS[self.symbolLevelList.GetSelection()]
-		config.conf["speech"]["trustVoiceLanguage"]=self.trustVoiceLanguageCheckbox.IsChecked()
+		config.conf["speech"]["symbolLevel"] = characterProcessing.CONFIGURABLE_SPEECH_SYMBOL_LEVELS[
+			self.symbolLevelList.GetSelection()
+		]
+		if self.symbolLevelWord.Get3StateValue() == wx.CHK_CHECKED:
+			config.conf["speech"]["symbolLevelWord"] = characterProcessing.SYMLVL_ALL
+		elif self.symbolLevelWord.Get3StateValue() == wx.CHK_UNCHECKED:
+			config.conf["speech"]["symbolLevelWord"] = characterProcessing.SYMLVL_UNCHANGED
+		config.conf["speech"]["trustVoiceLanguage"] = self.trustVoiceLanguageCheckbox.IsChecked()
 		currentIncludeCLDR = config.conf["speech"]["includeCLDR"]
 		config.conf["speech"]["includeCLDR"] = newIncludeCldr = self.includeCLDRCheckbox.IsChecked()
 		if currentIncludeCLDR is not newIncludeCldr:
diff --git a/source/speech/__init__.py b/source/speech/__init__.py
index 614dd850018..e3fc254be47 100755
--- a/source/speech/__init__.py
+++ b/source/speech/__init__.py
@@ -1084,9 +1084,13 @@ def speakTextInfo(
 		return _speakWithoutPauses.speakWithoutPauses(flatSpeechGen)
 
 	speechGen = GeneratorWithReturn(speechGen)
-	# Force symbol level to all when moving by characters or words (#11779)
-	if unit in (textInfos.UNIT_CHARACTER, textInfos.UNIT_WORD):
+	symbolLevel = None
+	if unit == textInfos.UNIT_CHARACTER:
 		symbolLevel = characterProcessing.SYMLVL_ALL
+	elif unit == textInfos.UNIT_WORD:
+		symbolLevelWord = config.conf["speech"]["symbolLevelWord"]
+		if symbolLevelWord != characterProcessing.SYMLVL_UNCHANGED:
+			symbolLevel = symbolLevelWord
 	else:
 		symbolLevel = None
 	for seq in speechGen: