GB2443652A

GB2443652A - Disambiguating text input using probability

Info

Publication number: GB2443652A
Application number: GB0622248A
Authority: GB
Inventors: S Bastien Racani Re; Frederic Claret-Tournier; Paul Hoffman
Original assignee: Samsung Electronics Co Ltd
Current assignee: Samsung Electronics Co Ltd
Priority date: 2006-11-08
Filing date: 2006-11-08
Publication date: 2008-05-14
Anticipated expiration: 2026-11-08
Also published as: GB0622248D0; KR100883334B1; GB2443653A; GB2443652B; KR20080041974A; GB0625447D0

Abstract

A mobile communications device such as a mobile telephone comprises input means for disambiguating text formed from input characters, by determining a word or expression partially or wholly matching the text input and determining the word or expression with the highest probability of matching the text input. Determination of the probability is achieved by a language modelling technique e.g. Prediction by Partial Matching (PPM), or the context of the text input and/or a training text. In this way the text input can be disambiguated more accurately with regard to the context of the instant text input, as well as past and present text input, rather than just the frequency of occurrence of a word or expression in general language. In a modification, the input means may also take account of the situation (e.g. work/leisure, time of day), device profile or a message recipient in determining the probability of a match.

Description

MOBILE COMMUNICATIONS

This invention relates to entry of text to mobile communications devices.

ckground An efficient way of entering text on a telephone keypad is desirable for using text messaging services such as the Short Message Service (SMS) provided by GSM and UMTS networks.

Currently, common ways of text-entry are the so-called multi-press method (also referred to as tap'-method) and predictive text entry methods, such as T9 Text rnput ("T9 ") by Tegic Communications, Inc., iTAPTM Intelligent Keypad Entry System by Motorola, Inc., and eZiText' by Zi Corporation.

On a typical telephone keypad, groups of letters in alphabetical order are associated with number keys. This is illustrated in Fig. IC. For example, "a", "b", and "c", are typically associated with number 2. Thus, any single press of a key is ambiguous, as it may represent any of the associated sets of three or four letters.

The multi-press method uses multiple taps on the same key to resolve this ambiguity. The user taps the key the number of times corresponding to the position of the letter in standard ordering. For example on the 2 key, the user taps once for "a", twice for "b", etc..

The disadvantage of the multi-press system is that more than one keystroke per letter is required.

An improved method is the predictive text entry. It allows the user to enter text by pressing only one key per letter. As the user enters a word letter by letter the system automatically compares all possible letter combinations that can result from the entered sequence of keys with a dictionary of words and thus "guess" the intended letters and words. However, often many dictionary words share the same numerical codes and in these cases the system presents the user with alternatives in a list. The user then selects the intended word from the list.

Some predictive text entry systems select and display the word which is most often used in a particular language as the most probable word and display the other words sharing the same numerical code (in order of decreasing probability of occurrence in that language) in a list. This reduces further the need of interrupting text entry for selecting one of the words in the list if the word selected as the most probable solution is the one desired by the user.

The use of predictive text entry systems makes text entry considerably more efficient.

However, a disadvantage arises as users often, particularly when entering SMS messages, wish to enter non-dictionary words, such as names, slang or abbreviations.

It is therefore an aim of the present invention to alleviate at least some of the disadvantages described above. It is another aim of the present invention to provide an improved method and system for entering text into a mobile communications device.

According to one embodiment of the present invention, there is provided a method of entering text on a mobile communications device, comprising the steps of: i) receiving text entry of one or more key presses of ambiguous alphanumeric characters; ii) determining possible words or expressions matching the text entry of step i); and iii) using language modelling, determining the word or expression with the highest probability of matching the already entered text.

According to another embodiment of the present invention, there is provided a method of entering text on a mobile communications device, comprising the steps of: i) receiving text entry of one or more key presses of ambiguous alphanumeric characters; ii) determining possible words or expressions matching the text entry of step i); and iii) taking into account the context of the already entered text andlor a training text, determining the word or expression with the highest probability of matching.

Embodiments of the present invention will now be described, by example only, with reference to the accompanying figures, whereby: Fig. 1A is a schematic front view of a mobile terminal in which the present invention can be implemented; Fig. lB is a schematic illustration of some of the elements of the mobile terminal of Fig. IA; p Fig IC is a schematic diagram of the numeric keypad of the mobile terminal of Fig. IA; Fig. 2 is a flowchart diagram illustrating the text entry method according to one embodiment of the present invention; and Fig. 3 is a flowchart diagram illustrating an example of one feature of a text entry method according to another embodiment of the present invention.

Figure IA is a schematic illustration of a mobile communication terminal 10. The terminal 10 includes a display 26, microphone 16, speakers 18, a keypad 21, antenna 20 and navigation keys 23.

Referring now to Figure IB, the terminal 10 comprises a processor 22, radio means 24 for communicating with other devices via a mobile communications network, an antenna means 26, a memory 28 and a user interface 30. A subscriber identity module 32 (SIM) for GSM terminals or a universal SIM (USIM) for UMTS mobile terminals can usually be inserted IS into the mobile terminal to enable the provision of services via a mobile telecommunications network. The SIM or USIM includes a memory clip and a microprocessor.

Fig. 1C schematically illustrates the numerical keypad of the mobile terminal of Fig. IA in more detail.

Each number key 40 indicates its number 42, followed by three or four associated letters 44. For example, "A", "B" and "C" are typically associated with number 2, "D", "E", "F" with number 3 etc..

In this way alphanumeric text can be entered using the numerical keypad 21.

Text entry systems are usually software applications running on the terminal's processor 22. The dictionary required for the predictive systems are typically stored in memory 28.

First Embodiment In order to provide an improved text entry system for keypads with a small number of keys like, for example, a mobile communications device with 12 keys, it is desirable to provide an improved method of selecting one of the multiple letters associated with one key is desirable.

In the following an embodiment based on language modelling will be described in more detail.

The text entry system is provided as a software application, stored in the mobile device's memory and running on the processor.

In the following it will now be referred to Figure 2.

The process starts in step 101, wherein the user starts to enter text into the mobile terminal. In step 102, the user presses one or more keys for desired alphanumeric characters. The algorithm now searches for all possible words which could be written with the letters associated with the keys pressed by the user so far (step 103).

In step 104 the algoritimi then checks whether one or more possible words are found. If only one matching word is found, this word is displayed (step 105). If the displayed word is the one desired by the user, the user confirms the selection (step 115) and can continue with step 101 by entering a new word.

On the other hand, if in step 104 more than one matching word is found, the process continues in step 106 by calculating the probability for each word to occur in the text entered by the user.

These probabilities are computed using Prediction by Partial Matching (PPM). PPM counts the number of times that words and sequences of words are encountered in a "training text". This "training text" is a combination of "a prior training text", a "prior use text" and an "usage text", made of: * A "prior training text" is a text stored in the device prior to the text entry system or algorithm being used by the user.

* A "prior user text" is text chosen by the user to train the text entry system. This may include one or more different texts selected by the user.

* A "usage text" is the text recorded as the user input. This may include all the text entered so far using the text entry system, or a selection of these texts.

Any of the above three training texts of any combination of texts can be used. The "usage text" is particularly advantageous as it allows the text entry to dynamically adapt to the user, thus providing for a self-improving text entry system.

The training texts used for the application are stored in the mobile device's memory, in the SIM/USIM memory, or other memory connected to the mobile device, such as a memory card or the like.

in step 107 the algorithm selects the most probable word or expression to the user, and in addition presents the user with the list of other words found in step 104. The words in the list arc ordered, starting with the word with the second highest probability of occurrence and continuing with words of decreasing probability.

If the displayed word is the one the user wished to enter, the user confirms this by pressing a non-alphanumeric key (step 114) and can now continue in step 101 by entering a new word.

If the word desired by the user is a word of the list proposed in step 107, the user can now browse through the list by using the navigation keys of the keypad (step 108) or enter one or more further alphanumeric characters by pressing the associated key in order to select one of words displayed in the list (step 102).

If, however, the word desired by the user is not in the list the user may enter a spelling mode (step 113), as will be described in more detail below.

If in step 103 no word is found which matches the entered keys so far the algorithm searches for possible words whose beginning can be written with the given sequence of keys (step 109). In step 110 it is checked whether more than one word is found.

If only a single word is found in step 110 the process continues in step by displaying this word.

If more than one word is found in step 110 the algorithm calculates the probabilities of occurrence for each word (step Ill). Here the probabilities are calculated by either proposing the full word with the highest probability or the sequence of letters such that the sum of probabilities of words completing this sequence of letters is the highest.

The word or sequence of letters with the highest probability is displayed to the user together with the list of other matching words as described above instepl08.

If in step 109 no matching word can be found (step 112) the process continues by entering spelling mode (step 113).

If the user presses at any time a non-alphanumeric character, the process continues in step 101 and interprets each following key press to relate to entry ofa new word.

The spelling mode starts when the user has pressed a set of keys and either no word which first letters could be given by this set of keys was found, or none of the words suggested correspond to what the user wishes to type.

Different versions of spelling modes may be applied, such as those in the following examples.

The spelling mode starts with a blank screen and lets the user type a word with the multi-press text entering system. Alternatively, the spelling mode uses the set of keys pressed so far to suggest a set of letters that the user can edit and/or complete.

Words entered in the spelling mode are added to the dictionary (i.e. the "training text" ) and if often used, can end up with a higher probability than previously existing words.

In the following a few examples are given to illustrate the above

description.

Example I:

Starting from an empty screen User presses key 2: ABC' "A" is a word, "B" is a not word, "C" is not a word.

= Propose A on the screen.

User presses key 6: MNO' Possible words are "AN", "AM", "CO", "CM". "AN" is the most likely used => Propose AN on the screen.

Utpresses key 3: DEE" Possible words are "AND", "COD", ,,2M),* "AND" is the most likely used = Propose AND on the screen.

Example 2:

Starting from a screen with A' followed by a space User presses key 2: ABC' "A" is a word, "B" is a not word, "C" is not a word.

= Propose A on the screen.

User presses key 6: WINO' Possible words are "AN", "AM", "CO", "CM". This time the word "AN" is very unlikely after "A". "CM" is the most likely word after "A".

Propose CM on the screen.

User presses key 3: DEF' Possible words are "AND", "COD", ,.2ND,, This time the word "AND" is very unlikely after "A". is the most likely word after "A".

Propose 2 on the screen (T9 would have proposed AND).

Example 3:

Starting from a screen with "1 DON'T KNOW" User presses keys 43 Here T9 would propose "HE" but our system would have learnt that "iF" is the most likely word after "KNOW" even though in general the word "HE" is more common than the word "IF".

= Propose IF on the screen

Example 4:

Starting from a screen with "GONE" User presses keys 4663 I' Here T9 would propose "GOOD" but our system would have learnt that "HOME" is the most likely word after "GONE", even though in general the word "GOOD" is more con-mion than the word "HOME".

Propose HOME on the screen In the first embodiment it has been described how, given a sequence of key presses and a context i.e. the preceding words in the text being written, the probability of each word can be determined that can be written with the sequence of letters. Subsequently, the word that has the highest probability is proposed to the user. The probabilities for each word are computed using PPM and are dependent on the number of occurrences of words in one or more training text and on the context (i.e. the preceding words) of these Occurrences.

Second Embodiment In the following a second embodiment is described which is an extended version of the first embodiment. In addition to the features described above the system includes the additional feature that the probability of the word being written not only depends on the context (i.e. the preceding words) but also on the current situation.

The system takes into account situation dependent parameters in order to determine the probability of each word that can be written with the given sequence of keys.

One way of taking into account situation dependent parameters is to use different training texts depending on the situation.

For example, the system can use a particular training text for each recipient of the text being entered or for each group of recipients, such as friends, family, business contacts etc. Alternatively, different training texts can be used depending on the settings or profile currently being used on the mobile device. For example, a first training text may be used for the profile used when being in the office or at a meeting, whereas another training text may be used for the profile used during free time.

A further alternative is to use different training texts depending on the time of the day and/or day of the week.

For example, a first training text is used on week days between 9.00 am and 6.00 pm, whereas a second training text is used on the weekend or after 6.00 pm on week days.

It is understood that the above described situation dependent usage may be combined in any possible way.

Instead of using different training texts for different situations, training texts can be shared between situations by giving different weights to different training texts. For example occurrences of words in the training text specific to the current situation are given a higher weight and thus count more, while occurrences of words in other training texts are counted in a normal way. In this way, situations that happen often would have large training texts and probabilities computed for them would be accurate, while situations with small training texts, for which probabilities could not normally be computed accurately, would be able to use information from other similar situations until their training texts grow.

Referring now to Figure 3, an example is illustrated using a flow chart diagram.

The process starts by the user, who is a trader in this example, pressing numerical keys "4 6 5 3" as a text entry on the keypad of his or her mobile device (step 201). In steps 202 the system determines the words that can be written by this given sequence of keys as "hold", "gold", "hole" and "golf'. If the message containing the above sequence was written on a Monday morning to a business contact (step 203), the system uses a first training text for determining the probabilities (step 204). In step 205 the system determines that the word "gold" has the highest probability of occurrence out of the possible words determined in step 202. Thus the word is suggested to the user.

On the other hand, if the message is written on a Friday evening to a friend and golf partner (step 206), the system uses a second training text (step 207). Using this second training text the system determines that the word "golf' has the highest probability of occurrence (step 208) and proposes this while the user puts in the text.

In the above embodiment it has been described that the situation for calculating situation dependent probabilities include the recipient or group of

S

recipients of the message to be written the profile or settings or the timing of the day/day of the week. However, it is appreciated that other situations may be taken into account when calculating the situation dependent probabilities, such as for example, situations derived from calendar entries.

Whilst in the above described embodiments the text entry system is described to be provided as a software application being stored in the mobile device's memory and running on the mobile device's processor, it is appreciated that the text entry system can be implemented in other ways. For example, the application may be stored on a memory card connected to the mobile device or may be stored on the SIM/USIM card memory and/or running on the SIM/USIM card's processor.

Whilst in the above described embodiments PPM has been described as a technique for determining the probabilities it is appreciated that alternatively other techniques may be used.

It is to be understood that the above describes embodiments are set out by way of example only, and that many variations or modifications are possible within the scope of the appended claims.

Claims

S

CLAIMS: 1. A method of entering text on a mobile communications device, comprising the steps of: i) receiving text entry of one or more key presses of ambiguous alphanumeric characters; ii) determining possible words or expressions matching the text entry of step 1); and iii) using language modelling, determining the word or expression with the highest probability of matching the already entered text.
2. A method according to claim I, wherein the word or expression determined in step iii) is displayed to the user.
3. A method according to claim 1 or 2, wherein the probability of occurrence is determined for the words or expressions determined in step ii).
4. A method according to claim 1, 2 or 3, wherein the words or expressions determined in step ii) are displayed to the user.
5. A method according to claim 4, wherein the words or expressions determined in step ii) are ordered according to probability of occurrence.
6. A method according to claim 4 or 5, wherein the user selects of the words or expressions displayed.
7. A method according to claim 4, 5 or 6, wherein the user selects one of the words or expressions by browsing through a list of words or expressions.
8. A method according to claim 4, 5 or 6, wherein the user selects one of the words or expressions by entering a further alphanumeric character.
9. A method according to any preceding claim, wherein in step ii) words or expressions are determined which match the entered key sequence.
10. A method according to any of claims I to 8, wherein in step ii) words or expressions are determined whose beginning match the entered key sequence
11. A method according to claim 10, wherein, for determining the highest probability in step iii), the full word or expression with the highest probability of occurrence is determined.
12. A method according to claim 10, wherein, for determining the highest probability in step iii), the sequence of letters is determined such that the sum of probabilities of words completing the entered sequence is highest.
13. A method according to any preceding claim, wherein Prediction by Partial Matching is used for determining the probability in step iii).
14. A method according to any preceding claim, wherein the probability of the occurrence of a particular word or expression is determined from one or more training texts.
15. A method according to claim 14, wherein the one or more training texts include one or more of the following: a text prior to entering text on the mobile device, a text selected by the user for training purposes, the text being entered by the user.
16. A method according to claim 14 or 15, wherein the number of words are counted in the one or more training texts to determine the probability in step iii).
17. A method according to any preceding claim, wherein in step iii) additional parameters are taken into account when determining the word or expression with the highest probability of matching.
18. A method according to claim 17, wherein the additional parameters are situation dependent parameters.
19. A method according to claim 17 or 18, wherein the additional parameters include one or more of the following: the recipient or a group of recipients, selected settings or profile of the mobile device, the date and or time of entering text or the day of the week.
20. A method according to claim 17, 18 or 19, wherein different training texts are used in order to take into account the additional parameters.
21. A method according to any of claims 17 to 20, wherein a plurality of training texts are used for determining the probabilities in step iii), and wherein at least one of the training texts is taken into account with a different weight in order to take into account the additional parameters.
22. A method of entering text on a mobile communications device, comprising the steps of: i) receiving text entry of one or more key presses of ambiguous alphanumeric characters; ii) determining possible words or expressions matching the text entry of step i); and iii) taking into account the context of the already entered text and/or a training text, determining the word or expression with the highest probability of matching.

S
23. A program or application adapted to perform the method of any of claims 1 to 22 when running on a processor.
24. A mobile terminal for use in a mobile communications network adapted to perform the method of any of claims I to 22.