Language Supplement, UK English

Nuance Recognizer 10.0.0

October 25, 2010

This guide is for the application developer who wishes to create speech recognition applications. It supplements the Nuance product documentation and details the grammars and pronunciation guidelines for UK English. The Language Pack Guide covers general details about grammars.

For more information on language-related application development topics, see the documentation included with your set of licensed Nuance speech products.

Be sure to review the release notes distributed with this product for the latest information, as well as restrictions and known problems.

This document includes these sections:

This document uses the font Arial Unicode MS to display IPA characters and any other special characters (in the Vocabulary items and pronunciations section). Arial Unicode MS is normally distributed with Microsoft Office. If the UK English characters don't appear when you view this document, make sure you choose View-->Encoding-->Unicode (UTF-8) in your browser.

Grammar documents

Built-in grammars for UK English (en-GB):

Creating grammars

The following subsections describe key issues for working with grammar documents in the UK English language. For detailed information on creating grammars, see your product documentation.

Character encoding

UK English grammars must specify Latin-1 character encoding, also known as ISO-8859-1. (Incorrect encoding will result in grammar compilation errors.) The first line of each grammar should be:

<?xml version="1.0" encoding="ISO-8859-1" ?>

Testing grammars with the ParseTool program

The ParseTool program lets you type sentences into a grammar and returns the grammar's results. The tool is stored in the bin directory of the installation baseline. See your product documentation for information.

When entering UK English text for the built-in grammars, note the following:

Valid characters in grammars

In order to define which characters can be used with this language pack please read the sections "Valid characters in grammars" and "Checking pronunciations with dicttest" in the Grammar Developer guide (accessible through the Product Documentation program shortcut).

alphanum_lc built-in grammar

The alphanum_lc built-in grammar recognizes a connected string of up to 20 digits and lowercase alphabetic characters, such as "a8f9h23". For example, this grammar could be used to recognize a product code or user id.  The “lc” in the name of this built-in means lowercase.  The possible characters are the lowercase letters a-z and the digits 0-9. The application layer can adjust the case of the returned letters as needed for further processing.

Note: This grammar replaces the alphanum built-in grammar.

alphanum built-in grammar

(NOTE: for backward-compatibility only. Otherwise, use alphanum_lc builtin)
This grammar has been replaced by the alphanum_lc grammar, but is still available. The alphanum builtin-grammar has been retained for backward-compatibility. For new implementations, please use the alphanum_lc builtin grammar. 

The alphanum built-in grammar recognizes a connected string of up to 20 digits and uppercase or lowercase alphabetic characters, such as “A8f9h23”. For example, this grammar could be used to recognize a product code or order number. The possible characters are the uppercase letters A-Z, lowercase letters a-z, and digits 0-9. Uppercase and lowercase letters are homonyms (e.g., “B” and “b”), so the inclusion of both is redundant for the purposes of speech recognition of case insensitive items such as product codes. Thus, the alphanum built-in grammar has been replaced by the alphanum_lc grammar.

boolean built-in grammar

The boolean grammar collects an affirmative or negative response.

Properties

The y and n parameters let you associate any two touchtone buttons as synonyms for yes and no.

Parameter

Description

y

Desired DTMF digit to be equivalent to "yes" (default = 1)

n

Desired DTMF digit to be equivalent to "no" (default = 2)

Examples

Caller says...

MEANING key

yes

correct

true

no

false

ccexpdate built-in grammar

The ccexpdate grammar understands the expiration date on a credit card. Expiration dates are usually a month and a year, and are often embossed on a credit card in the form "mm/yy." The grammar recognizes variations on the date, for example, "December 2007," "twelve oh seven," "twelve of two thousand and seven," "twelve slash zero seven," etc.

creditcard built-in grammar

The creditcard grammar understands a caller saying a credit card number, optionally preceding the number with the credit card name, or the words "account number" or "account." For example, a caller can say, "visa account number four oh one seven...," "mastercard five zero zero two...," or "three seven three five...."

currency built-in grammar

The currency grammar collects currency using pounds and pence (or `p' or "penny"). The grammar also accepts Euro and Cent.

Return keys/values

MEANING

Contains a string in the following form:

currencymain_unit_amount.subunit amount

If the caller explicitly says "euros" or "cents", then a currency value of EUR is added as a prefix.

If the caller explicitly says "pounds," "pence," "penny," or "p" then a currency value of GBP is added as a prefix.

If the caller does not explicitly indicate the currency type, then no prefix is added.

If the caller omits the main unit or subunit amount, then that field is zero. The string contains a leading zero if the subunit amount is collected without the main unit.

SWI_literal

contains the exact text that was recognized.

Examples

Caller says

MEANING

five pounds

GBP5.00

five euros

EUR5.00

five pence

five p

GBP0.05

five cents

EUR0.05

five pounds and five pence

GBP5.05

five euros and five cents

EUR5.05

five pounds and twenty-five

five pounds twenty-five

GBP5.25

five twenty-five

5.25

six hundred twenty-five thousand four hundred sixty-four pounds

GBP625464.00

one pound zero pence

GBP1.00

one twenty two

1.22

date built-in grammar

The date grammar accepts a date spoken in any of several formats.

Recognized phrases include "4 June," "4 June 2006," ""4, 6, 2006," "the 4th," "4th June," and "Monday, the 4th of June."

The grammar also accepts "yesterday" "today," and "tomorrow" which return values of -1, 0, and +1 respectively into the MEANING key.

Examples

Caller says

MEANING key

the 5th of January, 2000

20000105

Yesterday

-1

Today

0

Tomorrow

+1

the fourth

??????04

Wednesday

(Phrase not recognized)

Wednesday the 12th

??????12

June 4

June 4th

????0604

June 4, 1997

19970604

June 4, 97

??970604

Wednesday, June 4, 1997

19970604

the 6th

??????06

4, 6

????0604

10, 12

????1210

10, 12, 97

??971210

digits built-in grammar

Valid characters are the digits 0-9. The digit `0' can be pronounced as either "oh" or "zero."

national insurance built-in grammar

The national_insurance grammar understands NI numbers. The valid format is aannnnnnb (where a is alphabetic, n is a digit, and the optional b is the alphabetic A, B, C, or D). Illegal numbers are rejected.

The advantage of using this grammar rather than an alphanum grammar is that NI numbers have constraints that reduce that set of possible recognition hypotheses (and thus increase recognition accuracy).

Return keys/values

Upon return, the key MEANING is assigned to the recognized number.

number built-in grammar

The number grammar recognizes whole numeric numbers (the caller must not speak the individual digits).

Examples

Numbers from -999,999,999.99 to 999,999,999.99 are recognized, but by default the minallowed parameter is set to zero, which limits recognition to positive values.

Caller says

MEANING key

twenty five

25

twelve thousand three hundred forty five

12345

twelve hundred

1200

minus two

-2

negative two

(Phrase not recognized; the word "negative" is not allowed)

fourteen point five six

14.56

fourteen dot fifty six

(Phrase not recognized; the words "dot" and "fifty six" are not allowed)

phone built-in grammar

The phone built-in grammar accepts telephone numbers (landline and cellular) using the following conventions:

Optional leading 1 before a number.

Numbers reserved for emergency services: 999 (emergency) and 100 (operator)

The grammar does not allow natural number phrases such as "three two four five double two ." Callers can generally speak natural numbers telephone extension numbers (for example, saying "extension fifty two" instead of "extension five two").

Return keys/values

Upon return, the MEANING key is assigned to a variable length character result representing the recognized phone number.

Parameter properties

Additionally, as stipulated in the VoiceXML specification, the caller may specify an extension, for example, "five four two three five six seven extension two thousand." By default, extensions of one to four digits long are supported.

Property

Description

minextension

Minimum numeric value allowed for an extension (default is 1).

maxextension

Maximum numeric value allowed for an extension. Set this to 0 to disallow extensions. (Default is 9999.)

postcode built-in grammar

The postcode grammar recognizes valid alphanumeric postcodes in the UK. The following table shows valid formats ("A" indicates an alphabetic character and "N" indicates a digit):

Format

Example

AN NAA

B1 3AW

ANN NAA

M15 4PT

AAN NAA

CH4 9GB

AANN NAA

SW19 3PZ

ANA NAA

W1A 4RR

AANA NAA

SW1A 1AA

All the formats consist of two parts, separated by a space. The first part of the postcode only accepts values found in the Royal Mail database.

Return keys/values

Upon return, the key MEANING is assigned to the recognized postal code. The string is alphanumeric, all lowercase, and contains no spaces. For example, "SW1A 1AA" is returned as "sw1a1aa".

time built-in grammar

The time grammar recognizes a time of day.

Examples

For each entry, the values returned in the MEANING and QUALIFIER keys are shown. (Not shown are the values of the HOUR, MINUTE, and AMPM keys.)

Caller says

MEANING

QUALIFIER

now, immediately...

(Phrase not recognized)

--

in a half hour

(Phrase not recognized)

--

at noon

1200p

exact

at midnight

1200a

exact

before noon

1200p

before

after thirteen thirty

1330h

after

twenty twenty

2020h

exact

eight twenty in the morning

0820a

exact

half past eight

0830?

exact

half eight

(Phrase not recognized)

--

seven fifteen pm

quarter past seven in the evening

0715p

exact

twenty four hundred hours

twenty four hundred

0000h

exact

Vocabulary items and pronunciations

This chapter describes considerations for vocabularies and their pronunciations in UK English (en-GB).

Specially tuned pronunciations

The following table shows common words that are fine-tuned by Nuance. Each of these words contains "word-specific phonemes;" that is, phonemes and associated models created especially for the words.

Words with tuned pronunciations (do not modify):

UK English pronunciations

This section provides detailed reference information to help create pronunciation dictionaries. It is intended for people who have sufficient knowledge of the English language as spoken in the United Kingdom. It provides information about transcription and pronunciation.

As reference pronunciation dictionary we use:

Wells, John C.: Longman Pronunciation Dictionary.
Burnt Mill: Longman 1990. (ISBN 0-582-96411-3)

In this dictionary you will find the UK English as well as the American English pronunciation.

If you are not sure how a certain word is pronounced you can refer to the IPA transcriptions given there and then convert them into the SAMPA symbols, given in The UK English symbol set in alphabetical order.

The UK English phoneme system

The UK English phoneme system can be divided into two groups:

Furthermore, it is possible to define six different types of consonants:

Within the vowel group, further distinctions can be made between front, central and back vowels and diphthongs.

UK English spelling does have a certain complexity, since the orthography of most of its constituent words does not necessarily reflect their pronunciation. This lack of rigid structure means that the relationship between spelling (grapheme) and sound (phoneme) is difficult to define. Generally speaking, the phonetic transcription of a word is influenced by:

UK English symbol set grouped by phoneme classes

The following table shows all phonemes used in UK English transcriptions. These are listed according to their phoneme classes with their SAMPA and IPA representations.

Phoneme class

SAMPA

IPA

Examples of usage

Consonants

Plosives

b

b

bin

/bIn/

p

p

pin

/pIn/

g

g

give

/gIv/

k

k

skin

/skIn/

d

d

dummy

/dVmi:/

t

t

tin

/tIn/

Fricatives

v

v

saving

/seYvIN/

f

f

coffee

/kQfi:/

D

ð

this

/DIs/

T

θ

thin

/TIn/

z

z

crazy

/kreYzi:/

s

s

sin

/sIn/

S

ʃ

ship

/SIp/

Z

ʒ

vision

/vIZ@n/

h

h

hit

/hIt/

Affricates

tS

ʧ

chat

/tS{t/

dZ

ʤ

ginger

/dZIndZ@/

Nasals

m

m

mock

/mQk/

n

n

knock

/nQk/

N

ŋ

thing

/TIN/

Laterals

l

l

long

/lQN/

Vowels

Semivowels

r

r

run

/rVn/

j

j

yes

/jes/

w

w

wet

/wet/

Single vowels

I

ɪ

pit

/pIt/

i:

i:

ease

/i:z/

e

e

pet

/pet/

u:

u:

lose

/lu:z/

@

ə

away

/@weY/

{

æ

bad

/b{d/

A:

ɑ:

stars

/stA:z/

Q

ɒ

pot

/pQt/

O:

ɔ:

north

/nO:T/

V

ʌ

cut

/kVt/

3:

ɜ:

furs

/f3:z/

U

ʊ

put

/pUt/

Diphthongs

eY

eɪ

raise

/reYz/

aY

aɪ

rise

/raYz/

QY

ɔɪ

noise

/nQYz/

@W

əʊ

nose

/n@Wz/

aW

au̬ / aʊ̬

rouse

/raWz/

eR

eə

stairs

/steRz/

IR

ɪə

appear

/@pIR/

UR

ʊə

tourist

/tURrIst/

UK English consonants

English consonants typically consist of

Plosives

There are three voiced and three voiceless plosives in UK English, which can be arranged in pairs as shown here:

Voiced

Voiceless

/b/

bit

rabid

cab

/bIt/

/r{bId/

/k{b/

/p/

pit

rapid

cap

/pIt/

/r{pId/

/k{p/

/g/

gold

degree

bag

/g@Wld/

/dIgri:/

/b{g/

/k/

cold

decree

back

/k@Wld/

/dIkri:/

/b{k/

/d/

down

medal

sad

/daWn/

/med@l/

/s{d/

/t/

town

metal

sat

/taWn/

/met@l/

/s{t/

Fricatives

There are nine fricatives in the UK English SAMPA symbol set, five voiceless and four voiced:

Voiced

Voiceless

/v/

vine

even

prove

/vaYn/

/i:v@n/

/pru:v/

/f/

fine

rougher

proof

/faYn/

/rVf@/

/pru:f/

/D/

this

worthy

with

/DIs/

/w3:Di:/

/wID/

/T/

thin

earthy

oath

/TIn/

/3:Ti:/

/@WT/

/z/

zone

razor

plays

/z@Wn/

/reYz@/

/pleYz/

/s/

sign

racer

place

/saYn/

/reYs@/

/pleYs/

/Z/

gendarme

vision

/ZQndA:m/

/vIZ@n/

/S/

shine

mission

dish

/SaYn/

/mIS@n/

/dIS/

 

 

 

/h/

house

behind

/haWs/

/bIhaYnd/

In UK English the voiceless fricative /h/ does not appear in the final position.

Affricates

In UK English there are two affricates: /dZ/ and /tS/.

Note, that in SAMPA affricates are always represented by two single phonemes.

Voiced

Voiceless

/dZ/

gin

ridges

large

/dZIn/

/rIdZIz/

/lA:dZ/

/tS/

chin

riches

much

/tSIn/

/rItS@z/

/mVtS/

Nasals

There are three nasals in UK English, /m/, /n/, and /N/. The velar nasal /N/ (back of the tongue touches the soft palate) never appears in the initial position.

/m/

man

hammer

ham

/m{n/

/h{m@/

/h{m/

/n/

net

enter

run

/net/

/ent@/

/rVn/

/N/

sing

finger

/sIN/

/fINg@/

Pronunciation note: The grapheme <n> before <c>, <g>, <k>, <q>, and <x> is pronounced as /N/.

Syllabic /m/ and /n/ are represented as /@m/ and /@n/ respectively, for example: garden /gA:d@n/.

Laterals

There is one lateral in UK English: /l/.

/l/

long

falling

roll

/lQN/

/fO:lIN/

/r@Wl/

Syllabic /l/ is represented as /@l/, for example: level /lev@l/.

Semivowels

A semivowel is articulated by allowing air to escape over the center of the tongue through a stricture (in the case of /w/ two strictures) that is not so narrow as to cause audible friction. Semivowels are articulated like vowels, but function as consonants since they are not syllabic. They can also be referred to as approximants.

There are three semivowels in UK English, /r/, /j/, and /w/, shown below:

/r/

rich

blurring

/rItS/

/bl3:rIN/

/j/

young

view

/jVN/

/vju:/

/w/

win

away

/wIn/

/@weY/

In UK English final <r> is usually not pronounced, unless it appears in combined words as a linking-r, for example: faraway /fA:r@weY/.

UK English vowels

Front, central and back vowels

UK English single vowels (monophthongs) can be divided into three groups according to their place of articulation: front, central or back. Within each group vowels differ in their degree of mouth opening. Length is of minor importance in the UK English vowel system, and the length of a particular vowel in a given word may change considerably in connected speech. Thus the colon, which appears in some phonetic symbols to denote length, is used in the transcription of UK English to denote a different vowel quality rather than quantity (length).

The three vowel groups are shown in the following table, ranging in each group from closed (top) to open (bottom) mouth:

Front

Central

Back

/i:/

ease

believe

free

/i:z/

/bIi:v/

/fri:/

 

 

 

/u:/

ooze

goose

two

/u:z/

/gu:z/

/tu:/

 

 

 

 

 

 

/U/

umlaut

put

/UmlaWt/

/pUt/

/I/

itch

pit

/ItS/

/pIt/

 

 

 

 

 

 

/e/

pet

ever

/pet/

/ev@/

/@/

about

success

liver

/@baWt/

/s@kses/

/lIv@/

 

 

 

 

 

 

/3:/

urban

nurse

fur

/3:b@n/

/n3:s/

/f3:/

/V/

utter

cut

/Vt@/

/kVt/

 

 

 

 

 

 

/O:/

awe

north

cause

/O:/

/nO:T/

/kO:z/

/{/

apt

sad

/{pt/

/s{d/

 

 

 

 

 

 

 

 

 

 

 

 

/A:/

start

father

bar

/stA:t/

/fA:D@/

=/bA:/

 

 

 

 

 

 

/Q/

optimistic

pot

/QptImIstIk/

/pQt/

Pronunciation note: The short-o sound, is regularly transcribed as /Q/, like in moral /mQr@l/.

Diphthongs

There are eight diphthongs in the UK English phoneme inventory:

/eY/

aim

face

hay

/eYm/

/feYs/

/heY/

/aY/

ice

price

high

/aYs/

/praYs/

/haY/

/QY/

oyster

toys

boy

/QYst@/

/tQYz/

/bQY/

/@W/

omen

home

blow

/@Wm@n/

/h@Wm/

/bl@W/

/aW/

our

house

now

/aW@/

/haWs/

/naW/

/IR/

ear

near

/IR/

/nIR/

/eR/

air

area

square

/eR/

/eRrIR/

/skweR/

/UR/

cure

/kjUR/

Diphthongs can artificially emerge in a transcription when the individual phonemes that usually form a diphthong are placed adjacent in a word. For example, autoimmune. However, instances of such words are rare and can be ignored.

Specific pronunciation transcription methods

Linking-r

In UK English, a word pronounced in isolation never ends in /r/. However, in connected speech the final <r> is pronounced as if it is followed by a vocal, as in the combined words:

faraway

/fA:r@weY/

The inserted-r sound is known as `linking-r' and should be transcribed to avoid liaison problems.

Syllabic consonants

The consonants <l>, <m>, and <n> can sometimes form a syllable on their own. In these cases they are transcribed as /@l/, /@m/, and /@n/ respectively.

Pronunciation of foreign words

To transcribe foreign words, you must use the UK English SAMPA symbols.

If you use a different symbol set your system will be incapable of understanding the input.

Every language has a different phoneme inventory, so you may have problems in covering each and every sound. For the most common cases we offer the following transcription examples.

French nasals

Try to apply a pronunciation that has been adapted to UK English, for example

bonbon

/bQnbQn/

The original transcription /bo~bo~/ cannot be realized because the French phoneme /o~/ is not part of the UK English SAMPA symbol set.

Vowel 'y' in German and French

The vowel 'y', found in some German or French words can be represented by /u:/ or /jU/, such as:

Duchenne

/du:Sen/

Dubonnet

/djUbQneY/

Conveniently this reflects the pronunciation commonly used by UK English speakers who are not fully conversant within the particular language.

German fricative 'x'

Palatal and velar fricatives that occur in, for example, German, can be transcribed as /k/, instead of 'x'. As in:

Reich

/raYk/

Multiple pronunciations (variants)

Since it is possible to have more than one pronunciation for a word by using pronunciation variants, it may be difficult to determine how many pronunciation variants should be created. The general rule is: Variants should only be created if the pronunciation differs in more than one phoneme. Minor systematic variations can usually be reflected in the training material for the phonemes, and need not be covered by pronunciation variants. If such a word causes recognition errors, the creation of pronunciation variants may help to solve the problem.

The UK English symbol set in alphabetical order

The following table shows the UK English symbol set in alphabetical order:

SAMPA

IPA

Examples of usage

@

ə

away

/@weY/

@W

əʊ

nose

/n@Wz/

{

æ

bad

/b{d/

3:

ɜ:

furs

/f3:z/

A:

ɑ:

stars

/stA:z/

aW

au̬/ aʊ̬

rouse

/raWz/

aY

aɪ

rise

/raYz/

b

b

bin

/bIn/

d

d

dummy

/dVmi:/

D

ð

this

/DIs/

dZ

ʤ

ginger

/dZIndZ@/

e

e

pet

/pet/

eR

eə

stairs

/steRz/

eY

eɪ

raise

/reYz/

f

f

coffee

/kQfi:/

g

g

give

/gIv/

h

h

hit

/hIt/

I

ɪ

pit

/pIt/

i:

i:

ease

/i:z/

IR

ɪə

appear

/@pIR/

j

j

yes

/jes/

k

k

skin

/skIn/

l

l

long

/lQN/

m

m

mock

/mQk/

n

n

knock

/nQk/

N

ŋ

thing

/TIN/

O:

ɔ:

north

/nO:T/

p

p

pin

/pIn/

Q

ɒ

pot

/pQt/

QY

ɔɪ

noise

/nQYz/

r

r

run

/rVn/

s

s

sin

/sIn/

S

ʃ

ship

/SIp/

t

t

tin

/tIn/

T

θ

thin

/TIn/

tS

ʧ

chat

/tS{t/

U

ʊ

put

/pUt/

u:

u:

lose

/lu:z/

UR

ʊə

tourist

/tURrIst/

v

v

saving

/seYvIN/

V

ʌ

cut

/kVt/

w

w

wet

/wet/

z

z

crazy

/kreYzi:/

Z

ʒ

vision

/vIZ@n/

Online documentation and technical support

This language pack includes online documentation in HTML format. To access additional product documentation, look for the Product Documentation program shortcut. Send comments on Nuance documentation to techdoc@nuance.com.

Technical support is provided online at Nuance Network.