Grammar update system and method

BACKGROUND OF THE INVENTION

[0001] 1. Field of the Invention

[0002] The present invention relates to a grammar update system and method for updating grammar data used for enhancing speech recognition precision in an interactive voice response (IVR) system.

[0003] 2. Description of the Related Art

[0004] In recent years, due to the rapid advancement of a computer environment, such as an enhanced speed of a CPU, an IVR system using an ASR (Auto Speech Recognition) such as a voice portal server, which is conventionally considered to have an excessively heavy operation processing load, is being actively adopted in general applications.

[0005] In an IVR system adopted in a conventional application, in order to enhance speech recognition precision, speech recognition processing is generally performed often by using grammar in accordance with the state of interaction. More specifically, grammar is one of the factors that influence the enhancement of precision of speech recognition, the pursuit of ease to use for a user, enlargement of a degree of freedom in utterance, and the like.

[0006] For example, in the case of using a so-called dictation technique so as to enhance a degree of freedom in user's utterance, it is required to consider phrases that are hardly uttered. This increases an operation processing amount involved in speech recognition and decreases a speech recognition ratio as a whole, which lowers a level of user's satisfaction and aggravates a time for achieving a task and an achievement percentage.

[0007] Thus, there is a strong demand for obtaining finite-state automaton grammar having a scale and complexity sufficient for recognizing user's utterance assumed in accordance with the situation of interaction and applying the grammar to an IVR system.

[0008] However, it is practically very difficult to create sufficient grammar data. Therefore, actually, various phrases are assumed and organized in a format of grammar data.

[0009] Furthermore, in order to collect a large amount of speech data, the following is often performed: a number of subjects are allowed to use the present system or a pseudo system, whereby a number of phrases are extracted and organized as grammar data.

[0010] The following is also performed: speech data and the like in an existing speech interaction system are stored as log data, and a system manager creates new grammar with respect to a portion where misrecognition is caused due to the grammar while extracting the misrecognized portion.

[0011] In the case of using the above-mentioned methods, a considerable number of processes are required for creating grammar data. In some cases, it is even difficult to obtain grammar data capable of realizing recognition precision consistent with the number of processes.

[0012] For example, regarding dialect, it is required to collect a number of subjects that speak dialect in each area or in a particular area, or to obtain a technical person who sufficiently understands dialect. However, this is practically difficult.

[0013] Alternatively, there is a remarkable difference between male and female colloquial expressions. Therefore, a male technical person may not appropriately create grammar data in accordance with a female.

SUMMARY OF THE INVENTION

[0014] Therefore, with the foregoing in mind, it is an object of the present invention to provide a grammar update system and method capable of updating grammar data to be used at a predetermined timing while detecting an update portion of required grammar data by utilizing the result of speech data recognition without using grammar data.

[0015] In order to achieve the above-mentioned object, a grammar update system of the present invention includes: an update original grammar recording part for storing grammar data for speech interaction used for recognizing speech data; a speech newly-recognizing part for newly recognizing the speech data without using the grammar data; a new-recognition result determining part for determining whether or not a new-recognition result in the speech newly-recognizing part is acceptable using the grammar data; and a grammar adding and updating part for, in a case where it is determined that the new-recognition result is unacceptable using the grammar data in the new-recognition result determining part, specifying a portion to be added and updated from the grammar data stored in the update original grammar recording part, and adding the portion to be added and updated to the grammar data so as to update the grammar data.

[0016] According to the above configuration, an update portion of grammar data to be required can be detected while speech recognition processing using grammar data and speech recognition processing without using the grammar data are performed in parallel. Therefore, the grammar data to be used can be updated neither excessively nor insufficiently, and recognition precision can be enhanced.

[0017] Next, in order to achieve the above-mentioned object, a grammar update system of the present invention includes: a log data recording part for storing at least speech data and a speech recognition result as log data; an update original grammar recording part for storing grammar data for speech interaction used for recognizing the speech data; a log obtaining part for obtaining the speech data stored in the log data recording part; a speech newly-recognizing part for newly recognizing the speech data obtained in the log obtaining part without using the grammar data; a new-recognition result determining part for determining whether or not a new-recognition result in the speech newly-recognizing part is acceptable using the grammar data; and a grammar adding and updating part for, in a case where it is determined that the new-recognition result is unacceptable using the grammar data in the new-recognition result determining part, specifying a portion to be added and updated from the grammar data stored in the update original grammar recording part, and adding the portion to be added and updated to the grammar data so as to update the grammar data.

[0018] According to the above configuration, an update portion of grammar to be required can be detected while using log data. Therefore, grammar data can be updated neither excessively nor insufficiently, and recognition precision can be enhanced.

[0019] Furthermore, in the above-mentioned grammar update system of the present invention, it is preferable that the grammar adding and updating part is capable of setting restriction matter regarding the update of the grammar data. Because of this, special dialect, the difference between male and female colloquial expressions, and the like can be reflected.

[0020] Furthermore, it is preferable that the above-mentioned grammar update system of the present invention includes an update information presenting part for presenting the added and updated grammar data or an update portion in the grammar data; and an update instruction input part for confirming whether or not update processing is performed on a basis of the presented update portion of the grammar data. The reasons for this is that the recognition precision can be enhanced by providing a chance of performing tuning processing by a system manager regarding a recognition result.

[0021] Furthermore, the present invention is characterized by a recording medium storing software for executing the function of the above-mentioned grammar update system as the processes of a computer. More specifically, the present invention is characterized by a recording medium storing computer-executable software for realizing a grammar update method and processes thereof. The method includes the operations of: storing grammar data for speech interaction used for recognizing speech data; newly recognizing the speech data without using the grammar data; determining whether or not a new-recognition result in the newly-recognizing operation is acceptable using the grammar data; and in a case where it is determined that the new-recognition result is unacceptable using the grammar data in the determining operation, specifying a portion to be added and updated from the stored grammar data, and adding the portion to be added and updated to the grammar data so as to update the grammar data.

[0022] According to the above configuration, by loading the program onto a computer for execution, a grammar update system can be realized, in which an update portion of grammar data to be required can be detected while speech recognition processing using grammar data and speech recognition processing without using the grammar data are performed in parallel; therefore, the grammar data to be used can be updated neither excessively nor insufficiently, and recognition precision can be enhanced.

[0023] Furthermore, the present invention is characterized by a recording medium storing software for executing the function of the above-mentioned grammar update system as the processes of a computer. More specifically, the present invention is characterized by a recording medium storing computer-executable software for realizing a grammar update method and processes thereof. The method includes the operations of: storing at least speech data and speech recognition result as log data, storing grammar data for speech interaction used for recognizing the speech data, and obtaining the stored speech data; newly recognizing the obtained speech data without using the grammar data; determining whether or not a new-recognition result in the newly-recognizing process is acceptable using the grammar data; and in a case where it is determined that the new-recognition result is unacceptable using the grammar data in the determining process, specifying a portion to be added and updated from the stored grammar data, and adding the portion to be added and updated to the grammar data so as to update the grammar data.

[0024] According to the above configuration, by loading the program onto a computer for execution, a grammar update system can be realized, in which an update portion of grammar data to be required can be detected while using log data; therefore, the grammar data to be used can be updated neither excessively nor insufficiently, and recognition precision can be enhanced.

[0025] These and other advantages of the present invention will become apparent to those skilled in the art upon reading and understanding the following detailed description with reference to the accompanying figures.

BRIEF DESCRIPTION OF THE DRAWINGS

[0026]FIG. 1 is a diagram showing a configuration of a grammar update system of an embodiment according to the present invention.

[0027]FIG. 2 is a diagram illustrating a configuration of a grammar update system of an example according to the present invention.

[0028]FIG. 3 is a diagram showing a configuration of a grammar update system of the embodiment according to the present invention.

[0029]FIG. 4 is a diagram illustrating a configuration of a grammar update system of another example according to the present invention.

[0030]FIG. 5 is a diagram illustrating grammar data in the grammar update system of the embodiment according to the present invention.

[0031]FIG. 6 is a diagram illustrating a sentence structure corresponding to the grammar data in the grammar update system of the embodiment according to the present invention.

[0032]FIG. 7 is a diagram illustrating a configuration of a grammar update system of still another example according to the preset invention.

[0033]FIG. 8 is a flow diagram illustrating the processing in the grammar update system of the embodiment according to the present invention.

[0034]FIG. 9 is a view illustrating a computer environment.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0035] Hereinafter, a grammar update system of an embodiment according to the present invention will be described with reference to the drawings.

[0036]FIG. 1 is a diagram showing a configuration of the grammar update system of the embodiment according to the present invention. In the present embodiment, an update portion of grammar data is detected while recognition processing of speech data using grammar data and recognition processing of speech data without using grammar data are performed in parallel.

[0037] In FIG. 1, a generally used IVR system <b>1</b> is composed of a speech input/output part <b>11</b> for basically inputting speech data and outputting response speech data, a speech recognizing part <b>12</b> for recognizing speech data, an interaction managing part <b>13</b> for determining a response based on the recognition result, and a speech synthesizing part <b>14</b> for synthesizing response speech data to be output. The speech recognizing part <b>12</b> uses grammar data stored in an update original grammar recording part <b>15</b> so as to enhance recognition precision.

[0038] A grammar update system <b>2</b> of the present embodiment includes a speech newly-recognizing part <b>21</b> for obtaining speech data and the like in the IVR system <b>1</b>, and recognizing the speech data without using the grammar data stored in the update original grammar recording part <b>15</b>, a new-recognition result determining part <b>22</b> for determining whether or not the new-recognition result is acceptable using the grammar data stored in the update original grammar recording part <b>15</b>, and a grammar adding and updating part <b>23</b> for specifying an update portion and adding it to the grammar data so as to update the data.

[0039] Herein, the present embodiment is predicated on the fact that the recognition precision of the speech recognition in the speech newly-recognizing part <b>21</b> is higher than that in the speech recognizing part <b>12</b>. More specifically, the speech newly-recognizing part <b>21</b> adopts a recognition method that places weight on the performance rather than the recognition time, using, for example, a large vocabulary dictation system or the like having a sufficient number of acceptable sentences and sufficient vocabulary. The recognition method in the speech newly-recognizing part <b>21</b> is not particularly limited thereto, and any method with sufficient recognition precision may be used.

[0040] Furthermore, the new-recognition result determining part <b>22</b> determines whether or not the new-recognition result in the speech newly-recognizing part <b>21</b> can be accepted based on the grammar data stored in the update original grammar recording part <b>15</b>.

[0041] In the case where it is determined that the new-recognition result cannot be accepted based on the grammar data, the grammar adding and updating part <b>23</b> specifies such an unaccepted portion as an update portion, and adds the update portion to the, grammar data stored in the update original grammar recording part <b>15</b> so as to update the data, and outputs the updated grammar data to an updated grammar recording part <b>31</b>.

[0042] More specifically, the following application is considered. FIG. 2 is a diagram illustrating the configuration of a grammar update system of an example according to the present invention. In FIG. 2, the case will be described where a grammar being used is updated in real time using a speech log, an interaction state log, and a recognition result log in the IVR system <b>1</b>.

[0043] In FIG. 2, the WR system <b>1</b> includes a speech input/output part <b>11</b> for inputting/outputting speech data to be subjected to speech recognition, and a grammar to be updated recording part <b>16</b> for storing grammar to be updated.

[0044] First, a speech data obtaining part <b>24</b> obtains speech data and gives it to the speech newly-recognizing part <b>21</b> via a speech analyzing part <b>25</b>. The speech analyzing part <b>25</b> converts the obtained speech data to a feature value suitable for speech recognition by acoustic analysis processing such as FFT, and gives the feature value to the speech newly-recognizing part <b>21</b>.

[0045] The speech newly-recognizing part <b>21</b> recognizes the obtained speech data without using the grammar data, and gives the new-recognition result to the new-recognition result determining part <b>22</b>.

[0046] The new-recognition result determining part <b>22</b> obtains the grammar data recorded in the grammar to be updated recording part <b>16</b> in a grammar obtaining part <b>26</b>, and determines whether or not the new-recognition result in the speech newly-recognizing part <b>21</b> can be accepted, by using the grammar data. Then, in the case where the new-recognition result can be accepted, it is determined that the grammar data of the new-recognition result has already been contained in the grammar data recorded in the grammar to be updated recording part <b>16</b>. Therefore, an instruction signal indicating that it is unnecessary to update the grammar data is sent to the grammar adding and updating part <b>23</b>.

[0047] In the case where the new-recognition result cannot be accepted, the new-recognition result in the speech newly-recognizing part <b>21</b> and an instruction signal indicating that the grammar data should be updated are sent to the grammar adding and updating part <b>23</b>.

[0048] Only upon receiving the instruction signal for updating the grammar data, the grammar adding and updating part <b>23</b> matches the grammar data obtained in the grammar obtaining part <b>26</b> with the new-recognition result in the speech newly-recognizing part <b>21</b>, and determines an unmatched portion as an update portion of the grammar data. Then, the grammar adding and updating part <b>23</b> complements the grammar data with the update portion so as to update the grammar data, thereby outputting the updated grammar data to the updated grammar recording part <b>31</b>.

[0049] In the present embodiment, the following is also considered: a grammar portion to be updated is extracted based on log data containing speech data stored in a log data recording part of the generally used IVR system <b>1</b>, so that tuning of the grammar data can be executed at any time.

[0050] In this case, as shown in FIG. 3, a log data recording part <b>17</b> for recording an interactive log is provided in the generally used IVR system <b>1</b>.

[0051] A log obtaining part <b>27</b> for obtaining a speech log and the like stored in the log data recording part <b>17</b> in the IVR system <b>1</b> is added to the grammar update system <b>2</b>, and the speech newly-recognizing part <b>21</b> newly recognizes the obtained speech log.

[0052] The log obtaining part <b>27</b> obtains speech data as a speech log, log data representing an interaction state and a circumstance, and information on a recognition result in the speech recognizing part <b>12</b>, reliability of the recognition result in the speech recognizing part <b>12</b>, and the like. Furthermore, there is no particular limit to a data structure of log data. An example thereof includes a data structure in which the state of interaction, grammar to be used, an interaction time, a caller No., a user ID, speech data, and the like are organized hierarchically.

[0053] More specifically, the following application is considered. FIG. 4 shows a configuration of a grammar update system of another example according to the present invention. In FIG. 4, the case will be described where grammar being used is updated in real time, using a speech log, an interaction state log, and a recognition result log in the IVR system <b>1</b>.

[0054] Referring to FIG. 4, the IVR system <b>1</b> includes a speech log data recording part <b>18</b> for storing a log of speech data, an interactive log data recording part for storing a log regarding an interaction state, and the grammar to be updated recording part <b>16</b> for storing grammar data to be updated.

[0055] The interaction log data recording part <b>19</b> stores, together with time stamp information, an input speech data ID uttered by a user, a recognition result corresponding to the input speech data ID, a system response, an interaction state (n an interaction scenario), and a grammar ID used at that time. The speech data and the grammar data to be updated, stored respectively in the speech log data recording part <b>18</b> and the grammar to be updated recording part <b>16</b>, are also managed based on the input speech data ID and the grammar ID.

[0056] First, in an interaction log obtaining part <b>28</b>, interaction logs are read in the order of a time sequence by referring to the interaction log data recording part <b>19</b>, and input speech data IDs are obtained successively. Then, the obtained input speech data ID is given to the speech data obtaining part <b>24</b>. Simultaneously, a recognition result corresponding to the input speech data ID is extracted to be given to the new-recognition result determining part <b>22</b>, and the grammar ID used at this time is given to the grammar obtaining part <b>26</b>.

[0057] The speech data obtaining part <b>24</b> obtains speech data corresponding to the specified input speech data ID from the speech log data recording part <b>18</b> in the IVR system <b>1</b> and gives it to the speech analyzing part <b>25</b>. The speech analyzing part <b>25</b> converts the obtained speech data to a feature value suitable for speech recognition by performing acoustic analysis processing such as FFT, and gives the feature value to the speech newly-recognizing part <b>21</b>. The speech newly-recognizing part <b>21</b> recognizes the obtained speech data without using the grammar data and gives the new-recognition result to the new-recognition result determining part <b>22</b>. The subsequent processing is the same as described above, so that the detailed description thereof will be omitted here.

[0058] Furthermore, it is also considered to set restriction matter regarding the update of grammar. The purpose for this is to allow special dialect, the difference between male and female colloquial expressions, and the like to be reflected. For example, in the case where a unique phrase of particular dialect is not set as any restriction matter, a phrase in accordance with the standard language may be accepted using the grammar data, whereas a unique phrase of dialect may not be accepted. In such a case, there is no guarantee that a system manager is always familiar with the special phrase of dialect. Therefore, by previously preparing the special phrase of dialect as restriction matter in the grammar data, it becomes possible to prevent the recognition precision with respect to the special phrase of dialect from decreasing.

[0059]FIG. 5 illustrates grammar data stored in the grammar to be updated recording part <b>16</b>. As shown in FIG. 5, the relationship between ‘#action’ and ‘#object’ is recorded one by one based on an exemplary sentence structure as shown in FIG. 6. According to the method for updating grammar, for example, unrecorded ‘#action’ is added, and a method for connecting an article to a noun in ‘#object’ is newly defined. For example, it is considered to add ‘copy’ to ‘#action’, and the like.

[0060] Furthermore, the reliability of the new-recognition result in the speech newly-recognizing part <b>21</b> can be obtained with respect to an update portion of the grammar data. Therefore, a method for determining whether or not the grammar data should be automatically updated in accordance the reliability is also considered. Such reliability can be determined, for example, as follows.

[0061] First, speech data for evaluation is prepared, and a new-recognition result is obtained by processing the speech data in the speech newly-recognizing part <b>21</b>. Then, an evaluation score on the basis of one utterance or one word during use of speech data for evaluation is calculated.

[0062] Then, the contents of utterance of the speech data for evaluation are compared with the evaluation score on the basis of one utterance or one word during use of the new-recognition result and the speech data for evaluation to presume a relationship formula of the evaluation score and the correct answer ratio (criterion representing to which degree correct answers are obtained), whereby the reliability of the new-recognition result is obtained. Therefore, the reliability may be obtained with respect to the input entire speech data or on the basis of a word contained in the speech data.

[0063] Furthermore, low reliability means that the new-recognition result in the speech newly-recognizing part <b>21</b> is not reliable. Therefore, by determining whether or not the grammar data is updated in accordance with the reliability, the grammar data can be updated with good precision. FIG. 7 shows a configuration of a grammar update system of the present example. Hereinafter, only the difference with respect to FIG. 2 will be described.

[0064] In FIG. 7, the speech newly-recognizing part <b>21</b> provides reliability to the new-recognition result with respect to the entire result or on the basis of a word and outputs the resultant new-recognition result. Furthermore, in the case where the reliability of the entire new-recognition result in the speech newly-recognizing part <b>21</b> is low, even if the new-recognition result cannot be recognized by the grammar to be updated, the new-recognition result determining part <b>22</b> sends an instruction signal indicating that it is not necessary to update the grammar data to the grammar adding and updating part <b>23</b>.

[0065] The grammar adding and updating part <b>23</b> calculates the reliability of a grammar portion to be updated based on the reliability of a word that has not been matched with the obtained grammar data. Then, the grammar adding and updating part <b>23</b> sends an update original grammar ID, updated grammar data, an update portion of the grammar data, and the reliability of the grammar portion to be updated to an automatic/manual determining part <b>29</b>.

[0066] In the case where the reliability of the grammar portion to be updated is equal to or higher than a predetermined threshold value, the automatic/manual determining part <b>29</b> sends the update original grammar ID, the updated grammar data, and a replacement order signal to a grammar replacing part <b>30</b>. In the case where the reliability of the grammar portion to be updated is less than the predetermined threshold value, the automatic/manual determining part <b>29</b> sends an update portion of the grammar data and the reliability of the grammar portion to be updated to an update information presenting part <b>41</b> of a system manager terminal <b>3</b>, and is placed in an input stand-by state.

[0067] The update information presenting part <b>41</b> outputs the received information on a display or the like, receives an input of the system manager from an update instruction input part <b>42</b>, and sets the contents of the instruction (update/non-update) as a response signal to the automatic/manual determining part <b>29</b>. In the case where the response signal from the update information presenting part <b>41</b> indicates update, the automatic/manual determining part <b>29</b> sends an instruction signal for updating the grammar data as well as the updated grammar data to the grammar replacing part <b>30</b>.

[0068] Furthermore, in the grammar replacing part <b>30</b> receiving the instruction signal for updating the grammar data, the following configuration is also considered: in the case where it is determined that the grammar to be updated specified by the update original ID is manually replaced by the updated grammar data received together with the instruction signal, speech data based on which the grammar data is updated is sent to the update information presenting part <b>41</b>, and which contents of utterance is determined not to be recognized to necessitate updating of the grammar data is presented in the update information presenting part <b>41</b> together with the grammar update portion. In this case, related information accompanying the speech data can contain personal information on a user. Therefore, the system manager can change setting based on the determination of whether or not the user is member having the right to access the personal information.

[0069] Furthermore, in the update information presenting part <b>41</b>, signals (containing update information and the like) requesting an input by the system manager can be accumulated and processed as a single unit when the system manger accesses the system manger terminal <b>3</b>.

[0070] Furthermore, in the case of a configuration connectable to the network, a mobile phone, a PDA, or the like may be adopted as the system manager terminal <b>3</b>, and the system manager may be notified of the arrival of an instruction stand-by case via mail or the like.

[0071] Furthermore, in the case of an IVR system and the like in which grammar data is discriminated on the user basis, a user is considered to be suitable for determining whether or not the grammar data should be updated. In this case, it may also be possible to allow a user to update the grammar data by using a mobile phone or a web page.

[0072] In the present embodiment, a method for updating grammar data every time an update portion of grammar data is extracted has been described. However, a method for processing log data as a single unit to detect update portions as a single unit may also be used.

[0073] Next, a processing flow of a program for realizing the grammar update system of the embodiment according to the present invention will be described. FIG. 8 shows a flow diagram of the processing of a program for realizing the grammar update system of the embodiment according to the present invention.

[0074] In FIG. 8, first, input speech data is successively obtained (Operation <b>801</b>). Grammar data used for recognizing the obtained input speech data is obtained (Operation <b>802</b>).

[0075] Then, the speech data obtained without using the grammar data is converted to a feature value suitable for speech recognition by acoustic analysis processing such as FFT for new-recognition (Operation <b>803</b>).

[0076] Then, it is determined whether or not the new-recognition result can be accepted based on the obtained grammar data (Operation <b>804</b>). In the case where it is determined that the new-recognition result can be accepted (Operation <b>804</b>: Yes), an instruction signal indicating that it is not necessary to update the grammar data is sent (Operation <b>805</b>). In the case where it is determined that the new-recognition result cannot be accepted (Operation <b>804</b>: No), the new-recognition result in the speech newly-recognizing part and an instruction signal indicating that the grammar data should be updated are sent (Operation <b>806</b>).

[0077] Finally, only in the case where an instruction signal for updating the grammar data is received, the grammar data obtained based on a grammar ID is matched with the new-recognition result in the speech newly-recognizing part (Operation <b>807</b>). The unmatched portion is determined to be a grammar update portion (Operation <b>808</b>). The grammar data is complemented with the update portion to be updated (Operation <b>809</b>).

[0078] A program for realizing the grammar update system of the embodiment according to the present invention can be stored not only in a portable recording medium <b>92</b> such as a CD-ROM <b>92</b>-<b>1</b> and a flexible disk <b>92</b>-<b>2</b>, but also another storage apparatus <b>91</b> provided at the end of a communication line and a recording medium <b>94</b> such as a hard disk and a RAM of a computer <b>93</b>, as shown in FIG. 9. In execution, the program is loaded and executed on a main memory.

[0079] Furthermore, various log data used by the grammar update system of the embodiment according to the present invention, generated updated grammar data, and the like may be stored not only in a portable recording medium <b>92</b> such as a CD-ROM <b>92</b>-<b>1</b> and a flexible disk <b>92</b>-<b>2</b>, but also another storage apparatus <b>91</b> provided at the end of a communication line and a recording medium <b>94</b> such as a hard disk and a RAM of a computer <b>93</b>, as shown in FIG. 9. Such data is read by the computer <b>93</b> when the grammar update system of the present invention is used.

[0080] As described above, according to the present invention, a grammar creation/tuning operation in an IVR system, of which recognition precision has been difficult to enhance while requiring a considerable number of processes in a conventional example, can be executed at an arbitrary timing in accordance with a use situation by a user. Because of this, an IVR system can be configured and operated at a low cost with high performance, and at the same time, high recognition precision to a user can be realized easily.

[0081] The invention may be embodied in other forms without departing from the spirit or essential characteristics thereof. The embodiments disclosed in this application are to be considered in all respects as illustrative and not limiting. The scope of the invention is indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are intended to be embraced therein.