System for connotative analysis of discourse6332143Abstract A computerized interactive language reference system includes a database of terms with associated denotative, connotative and human interest information. The system goes through each word of a passage and determines whether there is an entry in the database for such word. For each term where there is an entry, there is a check to see if the term has more than one denotative meaning. Where there is more than one denotative meaning, an appropriate one of the denotative meanings is selected. The system evaluates the passage for positive emotional connotations, negative emotional connotations, global emotional connotations, human interest, connotations of power, connotations of activity and connotations of abstractness/concreteness. Dominant emotional connotations and dominant words also are specifically identified and ranked. Claims What is claimed is: Description BACKGROUND OF THE INVENTION
TABLE 2
Connotative Database Fields
POSITIVE EMOTIONS:
Affection/ Amusement / Enjoyment/ Contentment/
Friendliness Excitement Elation Gratitude
01 Adoration 01 Amazement 01 Admiration 01 Appreci-
02 Affection 02 Amusement 02 Bliss ation
03 Amorousness 03 Astonishment 03 Cheer 02 Comfort
04 Devotion 04 Eagerness 04 Delight 03 Content-
05 Fondness 05 Enthusiasm 05 Ecstasy ment
06 Friendliness 06 Excitement 06 Elation 04 Gladness
07 Infatuation 07 Exhilaration 07 Enjoyment 05 Gratitude
08 Kindliness 08 Exuberance 08 Euphoria 06 Hope
09 Liking 09 Fun 09 Exultation 07 Peaceful
10 Love 10 Glee 10 Happiness ness
11 Lust 11 Hilarity 11 Joy 08 Relief
12 Passion 12 Merriment 12 Jubilation 09 Satisfaction
13 Tenderness 13 Mirth 13 Pleasure 10 Serenity
14 Trust 14 Surprise 14 Pride 11 Thankful-
15 Warmth 15 Thrill 15 Rapture ness
16 Wonder 12 Well-being
NEGATIVE EMOTIONS
Fear/ Anger Humiliation
Sadness/Grief Uneasiness /Loathing /Shame
01 Affliction 01 Alarm 01 Abhorrence 01 Chagrin
02 Agony 02 Anxiety 02 Acrimony 02 Contrition
03 Anguish 03 Apprehension 03 Aggravation 03 Degradation
04 Dejection 04 Desperation 04 Anger 04 Discredit
05 Demoralization 05 Distress 05 Animosity 05 Disgrace
06 Depression 06 Dread 06 Annoyance 06 Dishonor
07 Desolation 07 Fear 07 Antagonism 07 Disrepute
08 Despair 08 Horror 08 Antipathy 08 Disrespect
09 Despondency 09 Nervousness 09 Aversion 09 Embarrass-
10 Disappointment 10 Panic 10 Bitterness ment
11 Discourage- 11 Paranoia 11 Contempt 10 Guilt
ment 12 Stress 12 Creepiness 11 Humiliation
12 Disheartenment 13 Tension 13 Detestation 12 Indignity
13 Disillusionment 14 Terror 14 Dissatis- 13 Mortifi-
14 Dismay 15 Uneasiness faction cation
15 Distress 16 Worry 15 Disdain 14 Regret
16 Downhearted- 16 Disgust 15 Remorse
ness 17 Dislike 16 Shame
17 Forlornness 18 Enmity 17 Stigma
18 Gloom 19 Envy
19 Grief 20 Exasper-
20 Heartache ation
21 Heartbreak 21 Frustration
22 Heartsickness 22 Fury
23 Hopelessness 23 Hatred
24 Hurt 24 Hostility
25 Longing 25 Irritation
26 Melancholy 26 Indignation
27 Misery 27 Ire
28 Pain 28 Jealousy
29 Pity 29 Loathing
30 Sadness 30 Offense
31 Sorrow 31 Outrage
32 Suffering 32 Rage
33 Torment 33 Rancor
34 Unhappiness 34 Resentment
35 Wretchedness 35 Vexation
36 Yearning 36 Virulence
37 Wrath
In an embodiment including a set of human interest fields 48, the fields 48 relate a record and its denotative context to a human interest category. The purpose of incorporating the human interest fields is to permit the end user to easily retrieve special connotative content from the database by first selecting one or more human interest filters before querying the database. The human interest fields employed in a preferred embodiment of the invention are listed in Table 3. There are nine groupings of the human interest categories in such embodiment. Each human interest category is a field having a descriptor word. The categories and descriptors may vary from embodiment to embodiment.
TABLE 3
SET 3: Human Interest Database Fields
Non-emotional Connotations Spiritual Identity
Power Biblical Diction
Activity Christianity
Judaism
Rhythm Islam
Number of Syllables Hinduism
Accented Syllable Buddhism
Other Religious
Special Diction Myth and Legend
Question-starting Words Paranormal
Core Words Identified by S. I.
Hayakawa Physical Identity
Physical Appearance
Personal Identity Body
Gender Health
First Names (Baby Names)
Notorious Or Celebrated People Perception
Languages Abstract/Concrete Continuum
National Identity Place, General
Organizations of Note Place, Event
Home Place, Transportation
Personal Relationships Place, Cosmos
Intimacy Place, Noted
SET 3: Human Interest Fields
Perception (cont'd) Non-human Life
Color Animals
Hearing Plants
Touch Micro Organisms
Taste
Smell Argot/Vernacular
Time, General Slang
Time, Historical Taboo
Time, Calendar Offensive
Non-medical Drug Use Derogatory
Disgusting/Revolting
Euphemistic
Cliche
In a preferred embodiment there are also a set of miscellaneous fields 48 for storing respective ratings of power, activity and abstractness versus concreteness. In a preferred embodiment of the invention, the assignment of the fields and records as described above effectively links each traditional dictionary definition of each word in each context with more than 200 connotative, human interest and miscellaneous variables. The data base is organized into one or more tables, files or other units of organizing data. Regardless of the number of tables or files, there are entries from each table or file which correspond to a term (i.e., a word or phrase among the denotative fields). The data associated with any given term is referred to herein and in the claims as a `record`, whether or not coming from a single table or file, or from multiple tables or files. Each record includes a field which identifies a term and another field which identifies the denotative context for the term. The record also includes a field which identifies a connotative meaning. Thus, each record includes at least three fields allocated among 1 to 3 tables, files or other units of organization of the data base. In a preferred embodiment there are a plurality of connotative meaning fields, at least one for each of a plurality of emotional categories. The connotative meaning fields identify an emotional descriptor for a corresponding emotional category. Such identification may be an identification of a specific emotional descriptor within the corresponding category or an indication that there is no emotional descriptor for such category. An indication of no connotative meaning is still considered to be connotative meaning information. In varying embodiments additional fields are included for any one or more of the following types of information: parts of speech, intensity of connotative meaning, power rating, activity rating, abstract/concrete rating, human interest areas. When an intensity field is included the intensity is an intensity of a connotative meaning and is associated with a corresponding emotional category and the identified emotional descriptor for such category. User Interface 14 and Analyzing Processes 16 In a preferred embodiment a graphical user interface 14 is implemented, which provides an end user with the capability of analyzing a passage of discourse for connotative content. Although there are many ways in which a user interface may be implemented, in one embodiment a system with menus and windows is used. Referring to FIG. 3, in one embodiment the user interface 14 includes a window 52 having an area 55 for displaying denotative information from the database 12 and an area 57 with varying content. Referring to FIG. 4, a process 50 for analyzing a passage for connotative content is shown. At step 51, input discourse 17 (see FIG. 2) is loaded and a passage 54 (see FIG. 3) of discourse is selected. At step 56, an operator presses a button 53 to commencing scanning the passage 54. During the scanning, the passage 54 is evaluated to identify words and phrases in the passage that correspond to terms in the data base 12. As previously described, there is a separate record in the data base for each denotative context of each term. Thus, for some terms there may be multiple records in the data base 12. In one embodiment, one of the multiple records is automatically selected and associated with the term of the passage. Such selection is based on probabilities or a language distinctive context interpretation system. For example U.S. Pat. No. 5,794,050 to Dahlgren et al. discloses a system for interpreting natural language input. In other embodiments or at the user's option, the user is prompted to select the appropriate denotative context (and thus the appropriate record). Such user selection has the advantage of being more accurate, but the disadvantage of being slower. Referring to FIG. 3 the user is being prompted to select an appropriate denotative context for the term "wedding." As shown in area 55 there are three denotative contexts for "wedding." From viewing the passage 54, the user can determine that the first context is correct. In one embodiment the user simply clicks on the first definition to make their selection. The scanning process then continues. In some embodiments two passes are performed. During a first pass, all words and phrases in the passage that have only one denotative context are identified separate from those that have multiple denotative contexts. During a second pass, the system prompts the user to select the appropriate denotative context for each term having multiple denotative contexts in the database 12. Referring again to FIG. 4, at step 58 a set of analytical processes are executed to derive connotative information pertaining to the passage. The derived connotative information then is displayed at step 59 in area 57 of the display window 52 (see FIGS. 5-8). Referring to FIG. 5, there are three tabs 62, 64, 66 corresponding to three levels of information. FIG. 5 shows the information at a first level where tab 62 is selected. Display box 61 shows the title of file name of the input discourse 17 or passage 54. The high level information of FIG. 5 includes a rating of the passage for emotional content in terms of positive emotion, negative emotion and global emotion (i.e., positive not distinct from negative emotion). The high level information also includes a rating of the passage for human interest by words, sentences and global content. a rating of the passage for power, a rating of the passage for activity and a rating of the passage along a scale of abstractness versus concreteness. Emotional content of the passage is derived by counting the number of emotional connotations associated with all the terms in the passage. Each term in the passage corresponds to 0 or one record in the database. Each record in the database 12 has at least one connotative field for each one of a plurality of emotional categories. For an embodiment where only one emotional descriptor is associated with a record per emotional category, there can be 0 to n emotional connotations per record (or per term in the passage), where n corresponds to the number of emotional categories. In some embodiments there are two connotative fields per emotional category. In such embodiment there can be 0 to 2n emotional connotations per record or per term in the passage. A rating of global emotion is a cumulative count of the total number of emotional connotations for all terms of the passage. Referring to FIG. 5 a bar graph 68 is displayed indicating the global emotion rating of the passage. Such rating is based on a scale. The scale is shown to vary from slight to infinite, although the scale referencing may vary. The emotional connotations stored for each term in the database are associated with either a positive emotional category or a negative emotional category. Thus, the global count can be divided to rate the passage for positive emotions and for negative emotions. Referring again to FIG. 5 a bar graph 70 is displayed indicating the positive emotion rating of the passage. Further, the global count can be broken down more finely by emotional category. FIG. 5 shows the negative emotions separated into the categories of negative emotions with a bar graph 71 displayed for each category. In a preferred embodiment, a category by category analysis is performed to obtain the ratings for positive and negative emotions. For a given category, the level 1 rating for connotative content is derived from the following equation: ##EQU1## where E.sub.ip =Rating of connotative content of a passage, p, for emotional category i; n.sub.p =total number of words having connotative content in passage p (i.e., excludes articles, conjunctions, forms of the verb "to be", et cet.); and D.sub.i,j,p =intensity of word j in passage p for primary emotional descriptor in emotional category i. The total rating for all positive emotions is the average rating for the four positive emotional categories. Similarly the total rating for all negative emotions is the average rating for the four negative emotional categories. The total rating for global emotion is the average rating for the four positive emotional categories plus the average rating for the four negative emotional categories. The ratings can be presented in summary fashion (e.g., as for the FIG. 5 examples of global emotion and positive emotion) or by category (e.g., as for the FIG. 5 examples of negative emotion). In some embodiments a different rating scale is used for the judge's selections (e.g., 0 to 7) than for the displayed intensities (e.g., 0 to 10). The ratings in FIG. 5 are shown to be the same 0-10 rating as for the displayed intensities. Thus, to achieve a rating along such scale, the rating is converted to a 0 to 10 scale before being displayed. With regard to the human interest rating, the passage is analyzed in multiple ways. One way is to count the number of terms which identify people (i.e., terms which are names of people, personal pronouns, references by gender (man, woman, grandma), and non-gender references to people (clown, firefighter). FIG. 5 shows a rating for human interest by words. Such rating is derived from the proportion of words which relate to people to the total number of words in the passage, placed on a scale of 0 to 10. Another human interest rating is by sentence. Each sentence which includes quoted dialogue, a question, an exclamation, or a command, each incomplete sentence, and each sentence addressed directly to the audience (e.g., reader) are considered to be conversational in nature. The proportion of such conversational sentences to the total number of sentences in the passage is displayed along a scale of 0 to 10, or another scale. Such rating is used as the human interest rating by sentence. Still another method for rating human interest is a global rating. The global human interest rating is derived from the human interest rating for words and the human interest rating for sentences. The rating for words is weighted more heavily than the rating for sentences. The specific weighting factor may vary. In one embodiment the rating for words, on a scale of 0 to 10, is multiplied by 3.75, while the rating for sentences on a scale of 0 to 10, is multiplied by 0.3. The two weighted ratings then are summed and divided by 2 to get a weighted average. Such weighted average is displayed. In FIG. 5 the global human interest rating is displayed along a scale of 0 to 10, although the scale may vary. With regard to `power`, the passage is analyzed by deriving the average power rating for all the terms in the passage. In one embodiment only those terms having a power rating are included in the derivation of the average. In another embodiment, a word without an entry in the database is considered to have a power rating of 0. For terms of the passage which correspond to a record in the database, the power rating is read from the corresponding field of the record. The average power rating stored in one embodiment is along a scale of 0 to 7, where each term in the data base 12 has a field for storing the power rating of such term along the same scale. Note however that different scales may be used for displaying data than for rating the data by evaluators. For example, it may be more meaningful for an evaluator to use a scale of 0-5 to rate power. However, it may be more beneficial to display the same rating on a scale of 0-10, for either consistency among sales or for easier acceptance to a user. Thus, the derived power rating can be rescaled to comply with a display scale which differs from the data base scale for power. A similar method is used to derive an average `activity` rating for the passage and a rating of `abstractness versus concreteness` for the passage. In particular each record in the database may include a field which stores an activity rating for the term associated with such record. An activity rating is the degree to which the term is an action of relates to an action or activity. Each record in the database also may store a field which associates a term with a relative degree of abstractness or concreteness, where abstract is at an opposite end of a scale from concrete. FIG. 5 shows an example of a power rating, activity rating and abstract/concrete rating for the passage 54 using bar graphs 72, 73, 74. Of course other visual indicators may be used to convey the respective ratings. Referring to FIG. 6, a more detailed level 2 analysis of the connotative content of the passage 54 is provided by clicking on the level 2 tab 64. The level 2 information includes a listing of the most frequent (i.e., dominant) emotional connotations. In various embodiments such listing can be according to a global ranking for all emotional descriptors. Alternatively, the listing can be according to a ranking by positive emotions in one list and negative emotion in another list. The listing can be more precisely defined into a list for each emotional category. FIG. 6 shows one rank-ordered list of the most frequent positive emotional descriptors associated with the terms of the passage. FIG. 6 shows a set of rank-ordered lists of the most frequent emotional descriptors according to negative emotional categories. Lists for 2 negative emotional categories are shown, although lists for the other two negative emotional categories could be viewed by scrolling down in the window area 57. The terms of the passage associated with the rank-ordered emotional descriptors are shown in parentheses adjacent to the emotional descriptor in the example illustrated. The number of descriptors listed, the inclusion of passage terms and the method of ordering the descriptors may vary. Referring to FIGS. 7 and 8, an even more detailed level 3 analysis of the connotative content of the passage 54 is provided by clicking on the level 3 tab 66. The level 3 analysis identifies the dominant words in the passage 54 for emotional connotation, human interest, concreteness or abstractness, power and activity. To derive a dominant word, each word achieves a score based on the number of occurrences of the word (in the given context--so that it identifies a unique record in the database) times the intensity rating for such word. With regard to dominant words for emotional connotation, it is the average intensity among all emotional connotations for such word that is used. For example, consider a first word with intensity ratings of 1, 2, 3, 4, 3, 2, 1, 0 for the eight emotional categories. If such word occurs 10 times in the passage, then it receives a score of 20. The average intensity is 16/8 =2. The score then is 2*10 =20. Consider a second word that has intensity ratings of 3, 4, 3, 4, 3, 4, 3, 4 for the eight emotional categories. The average intensity rating is 3.5. If such word occurs 5 times in the passage, then it receives a score of 17.5. Thus, the first word is more dominant in emotional connotation than the second word, even though the average intensity associated with the first word is less. The number of occurrences causes the first word to be more dominant in this example. Consider a third word that has intensity ratings of 9, 9, 9, 9, 9, 9, 9, 9 for the eight emotional categories. The average intensity rating is 9. If such word occurs 3 times in the passage, then it receives a score of 27. Thus, the third word is more dominant in emotional connotation than the first word or the second word, even though the number of occurrences is less than for either the first word or the second word. The average intensity rating causes the third word to be more dominant in this example. A similar approach is implemented to rank the words for dominance in the passage based on power, activity and concreteness. For these ranking however there is only one rating stored for each word for the specific ranking. Thus we do not need to calculate an average. The most dominant word is the word with the highest score, where the score is based on the number of occurrence times the ranking. Thus, for a power ranking, the score is the number of occurrences of a word in the passage times the power rating for such word. To rank the dominant words in the area of human interest, all the words that identify people are ranked by number of occurrences. Such words include names of people, personal pronouns, references by gender (man, woman, grandma), and nongender references to people (clown, firefighter). In addition a rating of the percent of words in the passage that have connotative meaning is derived. There may be some words in the passage which do not have any connotative meaning and others that do have connotative meaning. The percentage of words that do have connotative meaning is a useful indicator for a writer to determine how dramatic or interesting a work is as distinct from being dull or mildly interesting. Another rating shown in FIG. 8 is a rating of the percent of words in the passage that are Hayakawa words. Hayakawa words are a set of core words for any given language, as determined by the Hayakawa method known by those of ordinary skill in the art of linguistics. The percentage of words that are Hayakawa words relative to the total number of words in the passage is the displayed Hayakawa rating. Referring again to FIG. 5, summary and recommendation messages are displayed in area 55 of window 52. Such messages are part of a support database of writing prescriptions and summary information messages. The different prescriptions and messages are displayed when a corresponding trigger occurs. Such corresponding triggers are numerical data from the connotative database. Various ratings of positive and negative emotions result in corresponding messages. In FIG. 5, the mix of positive and negative emotions results in a message "Your text communicates a good mix of positive and negative emotional connotations." Another message would say a substantial mix rather than good mix for a higher rating of positive and negative emotional categories. For a lower rating the message may say a poor mix or even indicate that the writing does lacks connotative value. Similarly messages are present relating to human interest rankings, power ranking, and abstractness or concreteness. For example, messages are included which recite that the writing is abstract, or is somewhat abstract, that the writing conveys a high level of human interest or is devoid of human interest. Recommendation messages also are displayed in area 55. Such messages are triggered by similar profiles of various emotional, human interest, power and abstract ratings. Exemplary recommendations included in the support data base in one embodiment, include: All sentences are declarative. for more human interest consider recasting some of your material in the form of questions, commands or dialogue. Consider using more examples or illustrations incorporating concrete nouns. This will reduce abstractness and increase connotative values. Do not distribute this text. Rewrite this text from scratch. In addition, other messages are included. The tone and content of the messages may vary for differing embodiments. In a best mode embodiment the data base is stored on a computer readable storage media, such as embedded memory, RAM, ROM, a hard disk, an optical disk, a floppy disk, magneto-optical disk, electro-optical disk, or another known or to be implemented transportable or non-transportable computer readable storage media. The user interface, processing filters and other executable instruction code for maintaining and accessing the data base also is stored on the same or another computer readable storage media of the types listed above. Under various computing models, the connotative language analysis system 10 is installed at an end user computer or accessed remotely from an end user computer. For resident computing models, the executable instructions may be loaded onto the computer and the data base accessed from a transportable storage media. Alternatively the data base also may be installed locally. One skilled in the art will appreciate the many computing configurations that may be implemented. For non-resident computing models, the data base may be stored at a server computer on a public or private, local or wide area network, or even on a global computer network. The executable instructions may be run either at the end user computer or at a server computer with the data being displayed at the end user's display device. Identifying Connotative Meaning According to a preferred embodiment of the invention the connotative meanings associated with the terms found in the connotative language reference system 10 are derived by subjective responses from a plurality of evaluators. In a best mode embodiment the evaluators are a panel of persons having objective credentials or accepted expertise in connotative analysis. However, in some embodiments the evaluators may be selected at random. Such persons are referred to herein as connotative judges. In a preferred embodiment, the Internet is used as a recruitment medium to recruit 100 to 200 individuals who are not known to each other to act as independent connotative judges. In one embodiment, the connotative judges are screened for the following characteristics listed below in Table 4: TABLE 4 Characteristics and Qualifications of Connotative Judges 1. 25% of all judges aged 40 or older and female 2. 25% of all judges under the age of 40 and female 3. 25% of all judges aged 40 or older and male 4. 25% of all judges under the age of 40 and male 5. All judges having at least 2 years of post-secondary education 6. All judges having an above-average vocabulary and command of whichever language is being used to practice the invention. 7. All judges having a substantial interest and some experience in the craft of writing, preferably creative writing. 8. Judges geographically dispersed over the area of interest for the language of interest. While the above qualifications are used in one embodiment, the invention may be practiced using any number of judges having any qualifications of one's choosing. For example, connotative judges may be only women, or only men, or only individuals of a defined age or ethnic group, or only people who reside in a certain geographical location. The nature and quality of data captured will of course vary with the demographic profile of connotative judges, as well as with the number of judges used when practicing the invention, their geographical locations, and the linguistic qualifications of the judges. The connotative judges evaluate the meaning of given words and phrases for connotative content using a questionnaire. The questionnaires preferably are distributed as database software files, although they may also be distributed in paper document form. The responses of the connotative judges are processed using either custom-designed database software or database software commercially available from manufacturers such as Inprise, Oracle, and Microsoft. As the data are analyzed, a database of connotative meaning is constructed, which is linked with each context of each word in the connotative language reference system. Each questionnaire is, in effect, a small database table containing four data fields, preferably five data fields, as summarized in Table 5.
TABLE 5
Data Fields for Questionnaire Tables to Capture Connotative Data
Field 1 A field containing a term selected at random from the term
field of the main database
Field 2 A field containing the denotative context for the term in Field 1
Field 3 A field containing the part of speech for the term in Field 1
(optional, but preferred)
Field 4 A blank field assigned for the connotative judge to record data
identifying emotional connotations associated with the term and
context in Fields 1 and 2
Field 5 A blank field assigned for the connotative judge to record data
corresponding to intensity of emotion associated with the term,
context and part of speech in Fields 1, 2 and 3
Field 4 is used for identifying connotational meaning. Field 5 is used for identifying the intensity of the connotation. The use of Field 5 is described below in the section, Quantifying Intensity of Connotative Meanings. In one embodiment, a distributed computing model is employed, in which the connotative judges use their own computers in their own homes or offices to receive questionnaire tables over the Internet (via e-mail or from a World Wide Web site) that are extracted from the main database 12. The connotative judges complete their work on the questionnaire tables, and then return the data tables over the Internet. Referring to FIG. 9, a pool 80 of connotative judges are recruited to evaluate records of the connotative data base 12 for connotative associations of corresponding words and phrases. A sample of judges from the pool 80 forms a panel 82 used to evaluate a set of records. The same or different panels are formed to evaluate other sets of records. The number of records in a set may vary. For purposes of illustration a panel of 24 judges is described which evaluates a set of 500 records. In a preferred embodiment, each judge typically receives a questionnaire table 84 covering approximately 500 records, each record consisting of the four fields identified in Table 4. The questionnaire also includes instructions for selecting a code number to fill in the blank Field 4 for each record. Each connotative judge is also supplied with one or more of the eight category lists of code-numbered emotional descriptors identified in Table 2. In a preferred embodiment for a given questionnaire each judge is supplied with only one of the eight category lists of emotional descriptors found in Table 2. Thus, in a given questionnaire a judge evaluates the terms for connotative meaning in only a specific emotional category. To complete the questionnaire table, the connotative judge inputs one of three codes or code types for each record in Field 4, as listed in Table 6. TABLE 6 Coding Alternatives for Capturing SET 2 Connotative Data 1. Code the two-digit number (see Table 2 above) associated with one and only one emotional descriptor that most closely matches the emotional connotation that the connotative judge associates with the word or phrase, considering the denotative context and part of speech. 2. Code "00" if the connotative judge understands the word or phrase and its denotative context, but does not associate any of the emotional descriptors from the supplied list of emotional descriptors with the word or phrase and its denotative context. 3. Code "99" if the connotative judge does not know the word or phrase, or the specific associated denotative context. In a preferred method of practicing the invention, only one judgment is required for each record in a questionnaire table. However, in other embodiments more than one judgment may be allowed or required, if, for example, one wishes to capture the connotative judge's first choice and also the connotative judge's second choice of emotional descriptor. To effect such data capture, the number of connotative fields 46 (see FIG. 2) would need to be expanded accordingly, and the questionnaire table structure modified to include additional data capture fields. The connotative judge repeats this procedure for all 500 records in the questionnaire, then returns the completed questionnaire table 84 via the Internet. Thereafter, the judge may receive another questionnaire table 86, or 88 to evaluate. The next questionnaire table received by the connotative judge may contain exactly the same set of records that was just evaluated, but accompanied by a different category list of emotional descriptors to be used for coding. Alternatively, the next questionnaire table may contain a completely different selection of records. The exchange of questionnaire tables continues iteratively for the duration of connotative data collection. In a preferred embodiment of the invention, each block of 500 records is evaluated in this manner eight times (corresponding to the eight category lists of connotative descriptors listed in Table 2), each time by 24 different connotative judges selected at random from the pool of 100 to 200 available connotative judges, using a judge-selection technique that stratifies sampling to ensure equal representation according to the guidelines summarized in Table 4. Note that the number of judges selected, the size of the pool and the number of records processed in a given questionnaire may vary. Typically a plurality of panels 82, 83 are formed to evaluate the database 12 records for connotative associations. Different panels 82, 83 receive either the same or different questionnaires 84-89. For the exemplary embodiment where 24 judges evaluate each of 500 records in a given questionnaire, the same 24 judges may or may not evaluate all eight categories of emotional connotations for such 500 records. Quantifying Intensity of Connotative Meanings As previously described each judge receives a questionnaire. In one embodiment the questionnaire is in table format. Each record in the table has multiple fields as listed above in Table 5. Fields 1, 2 and 3 are already complete and correspond to the term, a denotative context for such term, and a part of speech. Field 4 is filled in as described above to identify connotative meaning for the word or phrase. Field 5 is to be filled in to record the intensity associated with the connotative meaning provided in Field 4. When a judge indicates that there is no connotative meaning for the word or phrase or that the judge does not know the term or the specific denotative context, then there is no need to quantify an intensity in Field 5. Where a judge is permitted to provide two connotative meanings (e.g., a primary and a secondary connotative meaning) then additional fields are included for each record in the questionnaire (e.g., two fields corresponding to Field 4 and two fields corresponding to Field 5). Typically a judge quantifies the intensity of the connotative meaning when selecting the connotative meaning itself. The intensity is a scaled value judgement of the judge. Table 7 lists the ratings scale for one embodiment.
TABLE 7
General Coding Model for Capturing Field 5 Data
1 2 3 4 5 6 7
slightly very
intense intense
0=automatically coded, corresponding to `00` code entry in Field 4 9=automatically coded, corresponding to `99` code entry in Field 4 In a preferred embodiment, a guiding set of anchor terms are included which are prerated for intensity. The judge's review of such guiding set improves consistency and accuracy among many judges completing the questionnaires. Specifically, the anchor terms are terms that are associated with scale numbers that represent the average intensity scores that other people have provided for various terms. Each connotative judge is expected to disagree with some of the scores represented by some of the anchor terms. For this reason, connotative judges are instructed to either, (i) highlight only those anchor terms with which they are comfortable, or (ii) alternatively, cross out those anchor terms with which they are uncomfortable. The connotative judge then compares the term he or she is scaling in the questionnaire table with his or her intensity of feeling associated with the anchored scales, choosing the scale number with the closest match. The anchor terms are representative averages; they do not imply "correctness," but rather provide the connotative judge with indicators of the relative strength of feeling that the connotative judge associates with the chosen emotional descriptor that is meant to be associated with each score number. In effect the anchor terms serve as a way for the judges to calibrate themselves to a scale of intensity. Table 8 presents a typical list of anchor terms associated with a group of emotional descriptors subsumed under the emotional category `Sadness,` which is one of the eight emotional categories previously described. TABLE 8 Anchor Terms for Scaling Intensity of Connotative Feeling for the Emotional Category"Sadness" (NOTE: the numbers 1 through 12 for the anchor words are merely for reference purposes. NO RANK ORDER is implied by the numbering within a scale level.) SCALE LEVEL 7 (Very Intense Sadness) 1. Holocaust--n the mass murder by the Nazis of the Jews of continental Europe between 1940 and 1945. 2. Auschwitz--n Polish town, site of a Nazi death camp during World War II. 3. child abuse n physical, sexual, or emotional ill-treatment or neglect of a child by parents or other adults. 4. AIDS--n acquired immune (or immuno-) deficiency syndrome. 5. rape victim--n a person who has suffered rape/sexual attack 6. starving person--n a person whose health is deteriorating from lack of food 7. murder--n the unlawful premeditated killing of one human being by another 8. cancer--n malignant growth or tumor; uncontrolled cell division 9. suicide--n the act or an instance of killing oneself intentionally 10. death--n the permanent end of all functions of life 11. blind--adj unable to see; sightless 12. insane--adj mentally deranged; crazy; of unsound mind SCALE LEVEL 6 1. slave--n a person having no freedom and forced to work for another. 2. abused person--n a person who is maltreated, esp. physically or sexually. 3. leukemia--n an acute or chronic disease characterized by a gross proliferation of leucocytes; cancer of the blood. 4. abandon--v to forsake completely; desert; leave behind 5. child pornography--n sexually explicit writings, pictures, films, etc., of children designed to stimulate sexual excitement. 6. depression--n a mental disorder characterized by extreme gloom, feelings of inadequacy, and inability to concentrate 7. divorce--n the dissolution of a marriage by judgment of a court. 8. starve--v to die or cause to die from lack of food. 9. lonely--adj unhappy as a result of being without the companionship of others 10. heartless--adj unkind or cruel; hard-hearted 11. hurt--adj injured or pained physically or emotionally 12. suicidal--adj involving, indicating, or tending towards suicide SCALE LEVEL 5 1. beggar--n a person who begs, esp. one who lives by begging. 2. sufferer--n a person who is undergoing pain, punishment, etc. 3. leper--n a person who has leprosy. 4. victimize--v to punish or discriminate against selectively or unfairly. 5. lose--v to be without, as through theft, accident, negligence, etc. 6. degrade--v to reduce in worth, character, etc.; disgrace; dishonor. 7. desert--v to leave or abandon, esp. in violation of a duty, promise. 8. deprive--v to prevent from possessing or enjoying; dispossess (of). 9. alienate--v to cause to become indifferent, unfriendly, or hostile; estrange. 10. demoralize--v to undermine the morale of; dishearten. 11. pain--n emotional, mental, or physical suffering or distress. 12. missing--adj not able to be traced and not known to be dead. SCALE LEVEL 4 1. inmate--n a person confined to an institution such as a prison or hospital. 2. drunkard--n a person who is frequently or habitually drunk. 3. addict--n a person who is addicted, esp. to narcotic drugs. 4. haunt--v to intrude upon or recur to (the memory, thoughts, etc.) 5. condemn--v to express strong disapproval of; censure. 6. refuse--v to decline to accept (something offered). 7. skid row--n a dilapidated section of a city inhabited by vagrants, etc. 8. hopeless--adj having or offering no hope. 9. alone--adj apart from another or others; solitary. 10. persecuted--adj oppressed, harassed, or maltreated. 11. unemployed--adj without remunerative employment; out of work. 12. deformed--adj disfigured or misshapen. SCALE LEVEL 3 1. wino n a person who habitually drinks wine as a means of getting drunk. 2. wretch n a person pitied for their misfortune. 3. tracks n needle marks on the skin of an injection drug user. 4. ostracize v to exclude or banish (a person) from a particular group, society. 5. forsake v to give up (something valued or enjoyed). 6. jail n a place for the confinement of prisoners. 7. lonely adj unhappy as a result of being without companionship of others. 8. hungry adj experiencing pain, weakness, or nausea through lack of food. 9. rejected adj not accepted, acknowledged, used, believed, etc. 10. pitiful adj arousing or deserving sympathy or sorrow. 11. helpless adj unable to manage independently. 12. let down adj unfulfilled in expectations ; disappointed. SCALE LEVEL 2 1. drop-out--n a student who fails to complete a school or college course. 2. lush--n a heavy drinker, esp. an alcoholic. 3. underestimate--v to think insufficiently highly of. 4. flophouse--n a cheap lodging house, esp. one used by tramps. 5. God Bless the Child--n a song written by Billie Holiday and Arthur Herzog. 6. homesick--adj depressed or melancholy at being away from home and family. 7. lost--adj confused, bewildered, or helpless. 8. empty--adj without purpose, substance, or value. 9. heavy hearted--adj sad; melancholy. 10. disenchanted--adj disillusioned. 11. unlucky--adj characterized by misfortune or failure. 12. blue--adj depressed, moody, or unhappy. SCALE LEVEL 1 (Light Sadness) 1. wallflower--n a person who stays on the fringes of a dance or party. 2. gambler--n a person who risks or bets (money) on games, sports, etc. 3. orphan--n a child, one or (more commonly) both of whose parents are dead. 4. runaway--n a person who takes flight or escapes. 5. dim--v to cause to seem less bright, as by comparison. 6. coal mine--n a system of excavations made for the extraction of coal. 7. mobile home--n living quarters mounted on wheels and capable of being towed. 8. Monday--n the second day of the week; first day of the working week. 9. comb-over--n a hairstyle in which long hair from the fringes of the scalp is arranged to cover and hide a bald portion of the scalp. 10. colorless--adj grey or pallid in tone or hue. 11. indifferent--adj showing no care or concern; uninterested. 12. resigned--adj acquiescent or submissive. A judge looks at each anchor term for a given rating in a given emotional category of the categories listed in Table 2. The judge selects one or more anchor terms under a given rating for a given emotional category which the judge feels most closely relates to the intensity rating subjectively felt by the judge. In a preferred embodiment, the anchor terms are updated over time based upon many judges' response entries into Field 5 for each record. In particular the Field 4 and 5 entries are analyzed to identify terms which consistently are judged by many different judges to have the same connotative meaning and the same intensity. Such terms become reliable anchor terms. This is done on an ongoing basis in order to build up a large, reliable database of anchor terms. As indicated above, an 8-point scale (including zero, indicating absence of the specified connotative feeling) is used to capture data for Field 5 in a preferred embodiment of the invention. However anchored scales of smaller or larger size, such a 3, 5, or 9 point scales may be used. Also, the number of anchor words or phrases may be greater than the 12 used in the preferred method, or fewer than 12. The number of anchor terms should be large enough to allow a choice permitting the connotative judge to select only those with which he or she is comfortable. The connotative judge repeats the above steps to input data for Fields 4 and 5 for all records in the questionnaire, then returns the completed questionnaire table via the Internet, then receives another questionnaire table to evaluate. The next questionnaire table received by the connotative judge may contain exactly the same set of terms that was just evaluated, but accompanied by a different list of emotional descriptors to be used for coding Field 4, and a correspondingly different set of anchors for coding Field 5. Alternatively, the next questionnaire table may contain a completely different selection of terms, with a corresponding emotional descriptor list for Field 4 and an anchor term list for Field 5. The exchange of questionnaire tables continues iteratively for the duration of connotative data collection. In a preferred embodiment of the invention, each block of 500 records is evaluated in this manner eight times, corresponding to eight emotional categories, each time by 24 different connotative judges. In this manner, a full-language dictionary database in any language, associating every context of every word with a very broad range of identified emotional connotations and their individual intensity levels, is constructed. Data Integrity Comparatively analyzing the connotative data associated with each block of records being processed serves to check for data integrity. Checking the data for integrity is part of an automated questionnaire processing function 90 (see FIG. 8). An initial integrity processing step is to determine whether any of the 24 sets of data should be rejected as invalid because of anomalous data. This is accomplished by statistically comparing the score set of each individual judge with the combined score sets of the other 23 judges who evaluated the same set of words using the same lists of emotional descriptors. If the scores between any given judge's data and the aggregate data of the other judges in the panel are not statistically related, then the data set for the anomalous judge is rejected. Anomalous data may arise if, for example, a connotative judge is filling in random data to avoid the mental work involved in providing genuine connotative data, or if a judge is coding a large number of double zeros and ninety-nines, or if a judge's experience is so far out of the mainstream that his or her connotative associations are not representative of the larger population. In a preferred method of practicing the invention, a minimum correlation level of 0.6 is used as a data rejection threshold. Further analysis includes determining how many valid non-zero scores remain after purging invalid scores and after accounting for 00 and 99 scores. A determination is then made to ascertain which emotional connotations the judges most often associate with each word or phrase. This is a function of four factors: 1. The number of valid scores remaining after data purging; 2. The number of emotional connotative descriptors in the list the judges had to choose from; 3. The number of judges who selected the same emotional descriptor; and 4. The probability that the same emotional descriptor was selected by more than one judge merely by chance. The multinomial probability distribution below in equation (I) embodies the above factors: ##EQU2## where: n is the total number of independent connotative judges evaluating the record; y is the number of judges selecting a particular emotional descriptor; p is the probability of the emotional descriptor being selected if the selection occurs by chance; q is the probability of an emotional descriptor being excluded if the selection occurs by chance; and P(y) is the probability of the emotional descriptor being selected by y judges if the selections occurred by chance. Tables may be constructed of the probabilities P(y) of connotative judges independently selecting the same emotional descriptors by chance for various panel sizes (e.g., increasing incrementally up to 24, and/or additional panel sizes of 36, 72, 96, 120 or any other panel size), and emotional connotative descriptors available for selection (e.g., increasing incrementally up to 24, with additional category group sizes of 36, 72, 96, 120 or other corresponding group size). As an example, consider the following set of connotative judgments for one word evaluated by 24 connotative judges on the Amusement/Excitement emotional category, which subsumes 16 emotional descriptors. The total number of valid judgments after purging is 21 (Table 9).
TABLE 9
Example of Field 4 and 5 Questionnaire Table Scores
Emotional Field 4 "Votes" Received Field 5
Descriptors from Connotative Judges Intensity Scores
Amazement 0
Amusement 3 5, 3, 5
Astonishment 0
Eagerness 2 5, 4
Enthusiasm 0
Excitement 1 4
Exhilaration 1 5
Exuberance 1 4
Fun 0
Glee 5 3, 4, 6, 4, 4
Hilarity 3 4, 4, 5
Meriment 1 5
Mirth 3 6, 4, 5
Surprise 0
Thrill 1 4
Wonder 0
The associated probabilities of chance selection of the same emotional descriptor by independent connotative judges, according to equation (I), are as follows:
Number of
Judges Selecting Probability of
the Same Category Chance Selection
0 0.258
1 0.361
2 0.241
3 0.102
4 0.030
5 0.007
In this example, only one emotional descriptor, "Glee," has been selected by enough independent connotative judges (5 judges) to meet the test of statistical significance, and is retained in the main database 12 as a connotative association for the term being evaluated. For any given term, selection of emotional descriptors from one emotional category does not preclude selection of emotional descriptors from other emotional categories. Any given term is apt to evoke several kinds of emotional response simultaneously. Therefore, the same term is also evaluated in an identical manner on the other seven categories of emotional connotations listed in Table 2. Thus, the term may, or may not, finish with more connotative emotional descriptors added when the data collection procedure has been completed. In a preferred embodiment of the invention, terms that receive no votes from the connotative judges on any of the connotative groupings, or too few votes on all eight connotative groupings to meet the test of statistical significance, are tagged as "non-connotative," so that such terms may be optionally excluded from further analysis or database querying. As for connotative intensity, all 21 scores in the above example are valid, not just the 5 scores for the specific emotional descriptor "Glee," because the 21 Field-5 scores represent the general emotional category, "Amusement/Excitement," which subsumes the specific emotional descriptor, "Glee." By completion of data analysis, each of the eight emotional variables contains one mean (i.e., average) intensity score for each word or phrase. An unbiased estimate of the variance of the sample of 21 anchored intensity scores in Table 9 is calculated according to the following variance formula: ##EQU3## from which the standard error of the mean for the sample is estimated in accordance with the following formula: ##EQU4## where: X is an independent connotative judge's score X is the sample mean n is the number of scores in the sample s.sup.2 is an unbiased estimate of the variance of the sample s is the sample standard deviation s.sub.X is the standard error of the mean. In the present example, the average of the 21-score sample of Field-5 data presented in Table 9 is 4.4. The standard deviation of the 21-score sample is approximately 0.81, which, when divided by 21 yields a standard error of the mean of about 0.177, for a 95% confidence level about the mean of .+-.0.35. Further accuracy is obtained by programming the computer to identify and purge "outlier" scores. This is accomplished by comparing each score with the mean and purging scores that are higher or lower than a statistically specified distance from the mean. In a preferred embodiment of the invention, terms that receive no Field-4 emotional descriptor votes from the connotative judges on any of the eight emotional categories (and therefore no Field-5 intensity scores) are tagged as "non-connotative," so that, at the user's option, such words may be excluded from further analysis or database querying. Identifying Human Interest Area Relating to a Record The Human Interest fields 48 may be defined in the same manner as the Field 4 data of Table 5. However, because the human interest fields are less subjective and relate more directly to denotative context, in a preferred embodiment assigned editors are used to define most of the human interest fields. However several variables on the Table 3 list of human interest fields, such as the miscellaneous fields for "Abstract-Concrete," "Power," and "Activity" are better left to evaluation by panels of connotative judges. These fields are defined as described above for the field 4 data and are subject to the same or similar data integrity procedures. A preferred embodiment of the invention such as the one described herein is both human-judgment based and dynamic, reflecting the human and dynamic nature of language. Since the data provided by the connotative judges are key to the system and method, one may wish to establish a program of continuous update of the database, either at prescribed intervals or on an ongoing basis, such as through a World Wide Web site. In this way, connotative judges would be able to supply data continuously, with turnover of connotative judges easily managed, and the database, particularly the connotative component, kept completely up to date allowing for new or changing connotative associations. In one embodiment participating judges periodically or aperiodically receive a mini-database via e-mail or by logging onto a web site. The mini-database serves as the questionnaire allowing the judge to enter a code for the connotative association (see table 6) for a given emotional category (see table 2). The results are then processed as described above for data integrity (see questionnaire processing 90 of FIG. 8 and related description). By practicing the above method and system of the present invention, a complete and accurate connotative language reference database 12 is constructed in any language, which then can be used to construct interactive connotative language reference tools, such as connotative dictionaries, connotative thesauruses, and connotative text analysis tools. Meritorious and Advantageous Effects One advantage of the system for identifying connotative meanings is that reliable associations, including connotative descriptions and intensities, are identified for given words and phrases in each of their denotative contexts. Another advantage is that the associations are maintained over time with changes in the vernacular or other changes/occurrences affecting connotative association. Another advantage of the invention is that an author, editor or another is able to automatically analyze the discourse to view the connotative aspects of the work. Such author editor or another interested party then can go back and alter the choice of some words, phrases, or even passages to fit more closely to a desired connotative profile desired for the discourse. In particular, by iteratively editing and analyzing the discourse a more desirable and effective discourse can be obtained for a given target audience. Although a preferred embodiment of the invention has been illustrated and described, various alternatives, modifications and equivalents may be used. Therefore, the foregoing description should not be taken as limiting the scope of the inventions which are defined by the appended claims.
|
Same subclass Same class Consider this |
||||||||||