Automated test item generation system and method

Abstract

A method and system for using a natural language generator for automatic assessment item generation is disclosed. The natural language generator includes a document structure generator that produces an abstract document specification defining a structure for an assessment item based on user input. The abstract document specification is input into a logical schema generator, which produces a logical schema specification that creates a more detailed specification for an assessment item. Finally, a sentence generator receives the logical schema specification and creates natural language for the assessment item based on the variables defined in the logical schema specification.

Description

FIELD OF THE INVENTION

[0002] The present invention relates to methods and systems for generating assessment items automatically. More particularly, this invention relates to a computer system and method for providing an interface for automatically generating word problems for use in testing.

BACKGROUND OF THE INVENTION

[0003] The use of Automatic Item Generation (AIG), the practice of creating assessment items algorithmically, is of increasing interest to educational assessors. AIG permits educational assessors to quickly generate assessment items and at least partially automate the generation of such assessment items. Furthermore, assessors can provide differing assessment items, each having a similar difficulty level, to improve test security. Moreover, adaptive testing having assessment items that vary systematically in difficulty can be produced. Such advantages have encouraged the research and development of AIG technologies.

[0004] However, AIG depends upon well-founded models of the cognitive abilities underlying performance. Where such models are lacking, AIG can only have heuristic usefulness. Conventional AIG research has been performed in areas where well-founded cognitive theories support the development of AIG algorithms, such as matrix completion and analytical reasoning. Such areas are generally restricted in content and highly structured.

[0005] In contrast, progress in AIG of verbal item types has been more limited, due to the openness of content and the considerable complexity of natural language. In open-ended verbal items, a strong preference exists for developing naturalistic items based upon actually published materials, and the most productive approaches have focused upon providing techniques to support test developers by supporting more efficient item selection and evaluation.

[0006] Where constrained item types have required natural language generation, the treatment of verbal materials has been straightforward and generally uses verbal templates to generate items. Typical template-based natural language generation includes two salient properties: 1) a list of phrases or sentences with open slots; and 2) the random or pseudo-random insertion of words from predetermined lists into particular slots. Template-based generation has the advantage of being straightforward, quick and dependent upon existing items. However, AIG from such simple templates is clearly limited because natural language complexities cannot be captured within a template format. Moreover, since the strings manipulated by template-based systems have no theoretical status, they do not support any principled analysis of the language employed in any particular problem type.

[0007] Conventional template-based AIG systems suffer from four distinct limitations: maintainability, output flexibility, output quality and an inability to easily produce multilingual outputs. In a template-based system, a large number of lists are stored and manipulated in order to generate textual output because each list is task or field specific. Accordingly, repetitive lists may be required to complete populate all task sets.

[0008] In addition, as the number of templates in a template-based system grows, it is more likely that the variety of templates disguises the systematic combination of a much smaller set of variables.

[0009] Moreover, systems must resolve context-dependencies inherent in language, such as subject-verb agreement, selection restriction (i.e., one drives a car, but flies an airplane), definite-indefinite selection (i.e., a student or the student), and the like. Such dependencies are handled ad hoc in a template-based system.

[0010] Finally, in order to produce a multilingual template-based system, a system maintainer must generate new templates for target language. Moreover, dependencies between templates and dependencies between entries in templates must be redefined for each template and/or combination of entries in the target language. As such, significant effort must be expended and significant resources must be dedicated to create a multilingual template-based system.

[0011] What is needed is a method and system for improving conventional automatic item generation by using non-template-based algorithms for generating assessment item text.

[0012] A need exists for a method and system for improving the maintenance of automatic item generation systems.

[0013] A further need exists for an automatic item generation system and method that increases output variety.

[0014] A still further need exists for an automatic item generation system and method that produces higher text quality.

[0015] A further need exists for a method and system of automatic test generation that more easily permits multilingual textual output.

[0016] The present invention is directed to solving one or more of the problems described above.

SUMMARY OF THE INVENTION

[0017] A Natural Language Generation (NLG) system according to the present invention may be used to perform automatic item generation. The NLG system may use computational linguistic techniques to automate text generation. The NLG system may receive input and determine textual content from the input. The textual content may be abstractly formulated without reference to details at the level of sentences, phrases or words. One or more sentences may be planned based on the abstract content. Following this process, the exact wording for each sentence may then be formed.

[0018] Thus, a major difference between an NLG system according to the present invention and a template-based system is that the specification of textual content is separated from the details of the wording. Building an NLG system entails constructing a model specifying possible content and separating it from a model specifying how that content is most naturally expressed.

[0019] An NLG system of the present invention is more maintainable than a template-based system because a large number of lists need not be stored and manipulated in order to generate textual output. In an embodiment, the present invention separates knowledge about language in general from knowledge about specific tasks. Accordingly, the present invention is easier to maintain and modify to suit changing assessment needs and demands.

[0020] In addition, the output of the present invention may be more flexible than a template-based model. The present invention may be used to reduce variety between template-based lists in order to enable more flexible test generation.

[0021] Moreover, systems must resolve context-dependencies inherent in language, such as subject-verb agreement, selection restriction (i.e., one drives a car, but flies an airplane), definite-indefinite selection (i.e., a student or the student), and the like. Unlike a template-based system, the present invention may include modules that resolve such dependencies automatically.

[0022] Furthermore, the present invention may permit text generation in multiple languages more easily than a template-based system. In an embodiment, low-level language-specific details are segregated into language-specific modules, which permit transposition into high-level knowledge without requiring a complete system redesign. Instead, adapting an NLG system for a target language may merely require adjustment of the language-specific modules to describe a new set of linguistic, grammatical, and/or semantic rules and translation of the entries in the template.

[0023] Considerable synergy between AIG within assessment theory and NLG within computational linguistics exists. Both AIG and NLG permit automatic generation of verbal material. Moreover, the verbal material is generated in complementary ways. While AIG focuses on controlling construct-relevant content, NLG focuses on lower-level document details.

[0024] Another advantage of the present invention is the ability to distinguish between factors that significantly affect item difficulty, known as radicals, and factors that do not, known as incidentals. Conventionally, no a priori way of determining radicals and incidentals is known. In general, the determination of radicals and incidentals depends on the cognitive content of the assessment item and the task the assessment item requires respondents to perform. However, in the present invention, the vast majority of variables are incidentals, such as the particular choice of words, grammatical constructions, and phrasing details.

[0025] Verbal tasks may be separated into two types of elements: those which involve decoding (i.e., determining the meaning of text as written) and those which involve content manipulation (i.e., performing inference or other thought processes on the content after decoding). The present invention specifies how to encode what test subjects decode, so a direct relationship exists between the encoding of the present invention and decoding by a respondent. In other words, radicals generally involve content manipulation and wording typically interacts with radicals at a content level.

[0026] The present invention is directed to a method and system of generating tasks that are drawn from a constrained universe of discourse, are susceptible to formalization, require little complex inference, and for which basic verbal comprehension (decoding/encoding) is not the tested subject matter. Such tasks may include, for example, mathematical word problems. Such mathematical word problems may include distance-rate-time problems, interest rate computation problems, taxation problems, production problems, physics problems, and the like.

BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Aspects, features, benefits and advantages of the embodiments of the present invention will be apparent with regard to the following description, appended claims and accompanying drawings where:

[0028] FIG. 1 depicts an exemplary high-level implementation of a system for performing natural language generation according to an embodiment of the present invention.

[0029] FIG. 2 depicts an exemplary screen shot of an existing word problem assessment item according to an embodiment of the present invention.

[0030] FIG. 3 depicts an exemplary variable modification screen for a distance-rate-time assessment item according to an embodiment of the present invention.

[0031] FIG. 4 depicts an exemplary detailed control screen for a distance-rate-time assessment item according to an embodiment of the present invention.

[0032] FIG. 5 depicts an exemplary task-relevant problem structure for an assessment item according to an embodiment of the present invention.

[0033] FIG. 6 depicts an exemplary measurement unit definition screen according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0034] Before the present structures, systems and methods are described, it is to be understood that this invention is not limited to particular structures, systems, methodologies or protocols described, as these may vary. It is also to be understood that the terminology used in the description is for the purpose of describing the particular versions or embodiments only, and is not intended to limit the scope of the present invention.

[0035] It must also be noted that as used herein, the singular forms “a,” “an” and “the” include plural references unless the context clearly dictates otherwise. Thus, for example, reference to a “variable” is a reference to one or more variables and equivalents thereof known to those skilled in the art, and so forth. Unless defined otherwise, all technical and scientific terms used herein have the same meanings as commonly understood by one of ordinary skill in the art. Although any methods, devices and material similar or equivalent to those described herein can be used in the practice of testing of embodiments of the present invention, the preferred methods, devices, and materials are now described. All publications mentioned herein are incorporated by reference. Nothing herein is to be construed as an admission that the invention is not entitled to antedate such disclosure by virtue of prior invention.

[0036] The term “semantic frame,” as used herein, refers to a method for organizing verbal content. In particular, a semantic frame makes it possible to analyze a particular type of word problem and isolate a series of variables with clear task-model relevance.

[0037] Semantic frames may be organized as a hierarchy of frames. Each semantic frame may contain linguistic detail including the schematic structure and the vocabulary associated with the particular semantic frame. Semantic frames may be used to denote particular patterns of vocabulary for certain types of assessment items that are highly systematic and have constrained vocabulary and syntax, such as mathematical word problems.

[0038] The term “schematic logical representation,” as used herein, refers to information contained in a file pertaining to the syntax and vocabulary used to generate natural language for a given assessment item. A schematic logical representation may define one or more variables for which text may be generated.

[0039] Based on the semantic frame that the schematic logical representation uses, the variables for an assessment item may be limited in their vocabulary (i.e., only a limited number of word choices may be available for each variable). For instance, in a Limousine variant of a distance-rate-time semantic frame, vehicle and driver variables may be constrained to be a limousine and a chauffeur, respectively. An exemplary schematic logical representation according to an embodiment of the present invention may represent the sentence “A car drove 600 miles” logically as “VEHICLE|W TRAVEL(DISTANCE=“600”|P).” In the schematic logical representation, lexical elements (VEHICLE, TRAVEL, and DISTANCE) may be represented as abstract labels, which are filled in by choosing a semantic frame and substituting appropriate lexical and grammatical materials. In addition, variables (|W and |P) and numeric values, such as the specification of distance, may permit the schematic logical representation to abstract itself from language-specific details of verbal expression while still providing a detailed specification of the content and the order in which it is to be presented. Each lexical element may represent a variable that ranges over a class of elements. The potential values for a lexical element may depend upon the selected semantic frame.

[0040] The term “mental model,” as used herein, refers to a structure with a limited number of component entities and relationships, subject to direct imagination and inspection. For example, assessment items typically vary along the following dimensions (exemplary values for distance-rate-time assessment items are provided):

[0041] The number of events (one or two, typically)

[0042] The semantic frame(s) associated with each event

[0043] The primary participants in each event (variations include operator v. passenger and the type of vehicle)

[0044] The identity of the primary participants (participants may be the same or different for each event and each participant may use the same vehicle or a different vehicle)

[0045] The secondary roles relevant to each event (e.g., distance, rate, time, route followed, start point, end point, start time and end time)

[0046] The identity of the secondary roles across events (i.e., specifying relationships among events)

[0047] FIG. 1 depicts an exemplary high-level implementation of a system for performing natural language generation according to an embodiment of the present invention. In the embodiment, three primary components are identified: the document structure generator 105, the logical schema generator 115, and the sentence generator 125. The document structure generator 105 receives as input a set of abstract choices from a user interface and produces, for example, an XML document 110. The XML document 110 may represent the underlying schematic content of an assessment item that is being generated. In an embodiment, the document structure generator 105 may perform, for example, the following tasks: building a mental model, outlining a document structure which expresses the mental model in a format compatible with the task-relevant problem structure, and passing choices about low-level wording to later modules. Building a mental model may include, for example, identifying which high-level semantic frames are to be used, setting up a list of events, and binding variables across events. Outlining a document structure may include, for example, determining the number of sentences required to express different parts of the mental model, determining the function of each sentence (e.g., stating a set of facts, querying a particular variable, and the like), and determining whether information is expressed in each sentence. Passing on the choices may include, for example, packaging the information for transmission to the logical schema generator 115.

[0048] In an embodiment, the first intermediate representation 110 may be an XML document. The first intermediate representation 110 may specify the possible values for variables pertinent to the assessment item, indicate high-level items pertaining to the assessment item structure, and outline the sequence of sentences that will constitute the final assessment item. An exemplary first intermediate representation 110 is shown in Table 1.

TABLE 1
An XML Word Problem Specification
<?xml><body>
<variables>W,X,Y,Z,VONE</variables>
<frame id=“1” frameID=“ONE” type=“DRT” VONE=“VEHICLE”/>
<event id=“2” frameID=“ONE” eventID=“A” type=“DRT”
subj=“W@” rate=“30” rateVar=“Y” />
<event id=“3” frameID=“ONE” eventID=“B” type=“DRT”
subj=“X@” rate=“70” rateVar=“Z” />
<bindings id=“4” frameID=“ONE” type=“DRT” W=“VONE” X=“VONE” />
<proposition id=“2” frameID=“ONE” eventID=“A” role=“QCColumnA” type=“DRT”
distance=“QUERY” rate=“EXPRESSED” time=“UNKNOWN” />
<proposition id=“F” frameID=“ONE” eventID=“B” role=“QCColumnB” type=“DRT”
distance=“QUERY” rate=“EXPRESSED” time=“UNKNOWN” />
</body>

[0049] The information contained in Table 1 specifies that five variables are used (i.e., W, X, Y, Z, and VONE) and that the sentences in the resulting assessment item will use vocabulary from a single semantic frame of type DRT (distance-rate-time). The variable VONE is defined to be of type VEHICLE within the semantic frame. In addition, two events are described in the document, with different rate values and different subject variables. Moreover, the resulting document includes two sentences. Each sentence queries the DISTANCE variable and provides RATE information. Neither sentence defines any information regarding TIME. Furthermore, the first sentence is placed in column A of a word problem having a Quantitative Comparison format, and the second sentence is placed in column B under the same format.

[0050] Although the content of the first intermediate representation 110 is abstract, the representation may indicate, in a concise form, the information that is to be presented and the organization of the information. The succeeding modules may define further details for the resulting assessment item.

[0051] The logical schema generator 115 may perform, for example, two discrete tasks. First, the logical schema generator 115 may structure sentences in the resulting assessment item so that the sentences are directly tied to the appropriate semantic frame. For example, for a distance-rate-time problem, the logical schema generator 115 may restructure the input from the first intermediate representation 110 so that a series of schematic statements or questions about VEHICLEs, TRAVELing, RATE, TIME and DISTANCE are produced.

[0052] In addition, the logical schema generator 115 may decide the format of the information contained in each sentence. Particularly, this may include the type of verb to use and the ordering of elements. The logical schema generator 115 may produce a predicate calculus in which the arguments are specified by the underlying mental model and in which the predicates are variables to be filled by actual natural language expressions. However, the logical schema generator 115 may not engage in fine details regarding the phrasing of each sentence. Instead, the transformation of the first intermediate representation 110 to the second intermediate representation 120 may be based on detailed information about the semantic frames relevant to each problem type. In an embodiment, the language for the assessment item is stored in the second intermediate representation 120. The second intermediate representation 120 may result in an output divided into two sections. The first section may include a series of variable definitions, which identify the classes of vocabulary to be used whenever a particular variable occurs in the text. The second section may include a logical formula using these variables, which sets up a schema or template for the resulting assessment item.

[0053] In an embodiment, the second intermediate representation 120 may be an XML document. The XML document may specify variable definitions including a series of tags for indicating how particular logical expressions map to actual vocabulary. Exemplary mappings are shown in Table 2.

TABLE 2
Logical Expression Mappings
<frameset id=“DRT” type=“9998” />
<frameset id=“TEMP” type=“10002” />
<vardef name=“DRIVE” id=“DRT” frameRole=“27” />
<vardef name=“VEHICLE” id=“DRT” frameRole=“3” />
<vardef name=“TIME” id=“TEMP” frameRole=“2” unit=“5” />

[0054] The definitions in Table 2 may define two semantic frames by reference to each semantic frame's location in a database containing the equivalent of a dictionary specification of all semantic frame elements. The variable DRIVE may be defined to use vocabulary from role 27 in the DRT semantic frame definition, the variable VEHICLE may be defined to use vocabulary from role 3 in the DRT semantic frame definition, and the variable TIME may be defined to use vocabulary from role 2 in the TEMP semantic frame definition. Based on these and other variable definitions, the remainder of the output may include a precise schematic specification of the content to be output. Table 3 shows a specification instance.

TABLE 3
Logical Schema Representation
<given>
((W and X each)|SB :: PERSON|SB
DRIVE(SEPARATE : VEHICLE|OB, SAME : DISTANCE|DI, SAME : NUMBER : TIME|TI)
</given>
<given>
PERSON|W LEAVEONE(ORIGIN|I, CLOCKTIME|M) and PERSON|W
ARRIVEONE(DESTINATION|K, CLOCKTIME|O),
DRIVEONE((VONE|Y :: VEHICLE|Y), (METER=“50”|S :: DISTANCE|S), STRAIGHT :
ROUTE|G)
</given>
<columnA>
AVERAGE : NUMBER : RATE :: PERSON|X
DRIVEONE((VONE|Z :: VEHICLE|Z), (TI|U :: TIME|U), EVENT|F :: PERSON
TRAVELONE|EVENT((DI|S :: DISTANCE|S), STRAIGHT : ROUTE|H, ORIGIN|J,
DESTINATION|L)) RANGE(START|N, END|P)
</columnA>
<columnB>
RATE=“3”
</columnB>

[0055] The representation in Table 3 varies from standard logical representations in order to make the translation to actual natural language text more straightforward. The translation is made more straightforward by, for example, separating the subject from the predicate; treating variables, such as |N, |S and |P, as labels attached to the elements that are verbally expressed; by treating modifiers as being attached to the argument that they modify; and the like.

[0056] The sentence generator 125 may translate the logical representations in the second intermediate representation 120 into assessment item text. The sentence generator 125 may parse the logical representations, annotate the resulting parse tree with grammatical information, find words and word forms in a dictionary based related to a selected language to fit the grammatical information and output the resulting text in order to complete the translation.

[0057] The sentence generator 125 may cover a wide range of linguistic details that are generally unrelated to the substance of the assessment item. For example, the sentence generator 125 may record the number of times an entity has been mentioned so that the choice of determiners between “a”/“an” and “the” is appropriate. Other grammatical phenomena may also be tracked, such as subject/verb agreement, preposition and case marking, and the like.

[0058] The output representation 130 may be an XML document having all text converted to the specified language. The output representation 130 may retain XML tags identifying the role that each text chunk performs in the structure of the assessment item. The output representation 130 may then be formatted for display and presented as an assessment item. The resulting assessment item may be used in connection with a test creation system.

[0059] FIGS. 2-6 depict exemplary screen shots of a graphical user interface used to generate assessment items according to an embodiment of the present invention. In the embodiment disclosed in FIGS. 2-6, a distance-rate-time assessment item is generated. Other embodiments include interest accumulation problems, tax problems, problems associated with producing items and the like. The variables used in each embodiment may be selected based on the problem type to be solved. Accordingly, the embodiment described below and in reference to FIGS. 2-6 is not meant to be limiting, but merely exemplary of the generation of one type of an assessment item.

[0060] Generation of the distance-rate-time assessment item may be sub-divided into five tasks: assigning mental model structure variables, defining identity variables in the mental model structure, determining a task-relevant problem structure, describing a document format and structure, and determining language variations.

[0061] The assignment of mental model structure variables may include defining the number of events; the number of distinct frames; the types of participants; the type of trip; the default unit of measure; and the actual distance, rate and time units used in each event. In an embodiment, a distance-rate-time problem has one or two events. Alternately, more events may be used for a distance-rate-time problem. If two or more events are used, the semantic frame for each event may differ. For example, a speed of a truck may be at issue in a first event, and a speed of a train may be at issue in a second event. Alternately, the semantic frames for each event may be the same. The type of participant may be used to specify whether people and/or vehicles are mentioned in the assessment item. If more than one event is included in the assessment item, the primary participants may include different people using the same vehicle or different vehicles, or the same person using the same vehicle or different vehicles. The trip type may be used to determine the values of one or more secondary variables. The trip type may include a round trip, a one way trip, two people or vehicles heading towards each other, two people or vehicles heading away from each other, and numerous other types of trips.

[0062] Defining the identity variables in the mental model structure may include choosing variables to express one or more of the following in each event: people, vehicles, time, distance, rate, starting locations, ending locations, starting times, ending times, the route and the event as a whole. While the distance, rate and time values are central to determination of the problem, other variables may be purely incidental to the determination of the answer to the problem.

[0063] Determination of a task-relevant problem structure may include determining which variable is being determined for each event, the answer for each event, numeric values for the mathematical variables (i.e., numeric values for distance, rate and time in each event), whether some information is provided in summary form, and whether time is expressed as a sequence of clock times. Most problems within a Transportation semantic frame may solve the equation d=rt. In such a case, the equation may map to the semantic roles Distance, Rate and Time. Another problem type that may be within a Transportation semantic frame may include a fuel efficiency problem. In such a problem, the equation to solve may be f=ed, where f is fuel consumption, e is fuel efficiency, and d is distance. If a unique answer is required for solution to an assessment item, only one variable may be unspecified. Accordingly, values may be assigned to all but one variable to provide for a unique solution. Implicit values may be assigned to complicate an assessment item. For example, the amount of time may not be explicitly specified. Rather, the assessment item may state that the person/vehicle traveled from 3:00 to 6:00.

[0064] Describing a document format and structure may include determining the format of the problem (i.e., quantitative comparison vs. problem solving) and the arrangement of content between the options and the stem. The problem format may determine the type of language used for the query of the assessment item. For example, a problem-solving (multiple choice) assessment item may query information in the form of a question: “Which of the following?” “How much?” or “How many?” In contrast, a quantitative comparison question may query information in a phrasal form: “The number of hours required . . . ,” “The time taken to . . . ,” “The speed at which . . . ” Moreover, content may be arranged in different sections of an assessment item. For instance, in a quantitative comparison assessment item, both background information and the questions may be posed in the columns. Alternatively, background information may be posed in a question stem and the questions may be posed in the columns. Such arrangements are merely exemplary. Other embodiments may arrange information in alternate arrangements without limitation.

[0065] Determining language variations may include, for example, selecting a language, selecting a sentence structure, selecting referent identification types (identifying people by name or generically, identifying object with labels or by description), selecting detailed phrasing, determining whether quantities are in integer format, determining whether rates are identified as constant or average, and whether events are described in the present or past tense. The language may be selected based on a variable assigned by a user. In an embodiment, different data structures are stored for each language. The data structures may include, without limitation, words (in the selected language) available for use in assessment items and rules pertaining to sentence structure and grammar. The sentence structure may include an active or passive verb form. Referent identification may include calling a person in an event, for example, “Fred,” “a driver,” “the driver,” or “driver A.”

[0066] FIG. 2 depicts an exemplary screen shot of an existing word problem assessment item according to an embodiment of the present invention. The menu bar may be used to modify one or more variables for the assessment item.

[0067] FIG. 3 depicts an exemplary variable modification screen for a distance-rate-time assessment item according to an embodiment of the present invention. In an embodiment, the variable modification screen permits a user to alter the variable for which the assessment item solves 305, the basic document format 310, the core participant type 315, and the number of events 320.

[0068] FIG. 4 depicts an exemplary detailed control screen for a distance-rate-time assessment item according to an embodiment of the present invention. In an embodiment, the detailed control screen permits a user to alter the number of distinct vehicle types 405, the name used for each person 410, the type of trip 415, and numerous other parameters pertaining to the syntactic structure and the secondary grammatical content.

[0069] FIG. 5 depicts an exemplary task-relevant problem structure for an assessment item according to an embodiment of the present invention. In an embodiment, at least one of the distance, rate and time variables 505 and 510 may be set for each event in the assessment item. If desired, the user may assign constant values 515 to the assessment item. In addition, an answer 520 to the assessment item may be supplied if desired. If an answer is supplied, values may be selected from the ranges defined for each event to ensure that the specified answer is correct for the resulting assessment item. If any value is unspecified, a random value may be selected for the variable.

[0070] FIG. 6 depicts an exemplary measurement unit definition screen according to an embodiment of the present invention. Default units may be selected for one or more of a rate, a time and a distance either globally 605 or on a per event basis 610 and 615.

[0071] Using the graphical user interface described in FIGS. 2-6, new assessment item instances may be generated from an underlying model. Alternatively, parameters may be altered in an existing assessment item to generate a new assessment item.

[0072] Although this invention has been illustrated by reference to specific embodiments, it will be apparent to those skilled in the art that various changes and modifications may be made which clearly fall within the scope of the invention. The invention is intended to be protected broadly within the spirit and scope of the appended claims.

Claims

1. A method for automatically generating a word problem, the method comprising: receiving a designation of a problem type from a user; identifying one or more variables based on the problem type; assigning at least one value to at least one variable; receiving, from a user, a designation of a variable for which to solve; and generating an assessment item of the problem type based on the variables.

2. The method of claim 1 wherein identifying one or more variables comprises identifying a language used to generate the assessment item.

3. The method of claim 1 wherein the problem type comprises a mathematical word problem.

4. The method of claim 3 wherein assigning at least one value to at least one variable comprises determining a unit of measurement for a variable.

5. The method of claim 1 wherein assigning at least one value to at least one variable comprises assigning a range of values to a variable from which a value for the variable is selected when the assessment item is generated.

6. The method of claim 1 wherein assigning at least one value to at least one variable comprises assigning a variable used to determine a format for the assessment item.

7. A method of automatically generating an assessment item, the method comprising: receiving one or more inputs from a user; generating one or more variables based on the one or more inputs; determining one or more relationships between at least two of the variables; and generating an assessment item based on the one or more variables and the one or more relationships.

8. The method of claim 7 wherein the one or more relationships comprise relationships based on one or more of word order; word choice; word format; sentence structure; grammar; and language.

9. The method of claim 7 wherein the assessment item is a mathematical word problem.

10. A method of automatically generating an assessment item, the method comprising: receiving one or more inputs pertaining to the format of an assessment item, wherein the format of the assessment item comprises at least one event; selecting one or more variables for use in the assessment item; determining a relationship between variables assigned to each event; determining a format for the assessment item; and generating an assessment item based on at least the format for the assessment item and the relationship between variables assigned to the at least one event.

11. The method of claim 10 wherein the one or more inputs comprise one or more of the following: a number of events; a number of distinct frames; a type of participant in each event; a type of assessment item to generate; and a unit of measure for each variable.

12. The method of claim 10 wherein determining a relationship for the variables assigned to each event comprises one or more of the following: determining a variable for which to solve for each event; determining an answer for each event; determining a value for one or more variables; and determining a variable format.

13. The method of claim 10 wherein determining a format for the assessment item comprises: determining a problem format having one or more sections; and determining content to place within each section.

14. The method of claim 10 wherein generating an assessment item comprises one or more of the following: selecting a language for the assessment item; selecting a sentence structure for each sentence in the assessment item; selecting identification types for one or more of the variables; determining a numerical format for each of the one or more variables; and determining a verb tense to use for each event.

15. The method of claim 14 wherein selecting identification types comprises determining to identify a variable denoting a person by using a proper name.

16. The method of claim 14 wherein selecting identification types comprises determining to identify a variable denoting a person generically.

17. The method of claim 14 wherein selecting identification types comprises determining to identify a variable denoting an object by using a label.

18. The method of claim 14 wherein selecting identification types comprises determining to identify a variable denoting an object by using a description of the object.

19. A system for automatically generating an assessment item, the system comprising: a processor; and a computer-readable storage medium operably connected to the processor, wherein the computer-readable storage medium contains one or more programming instructions for performing a method of automatically generating an assessment item, the method comprising: receiving a designation of a problem type from a user, determining one or more variables based on the problem type, assigning at least one value to at least one variable, receiving, from a user, a designation of a variable for which to solve, and generating an assessment item of the problem type based on the variables.

20. A system for automatically generating an assessment item, the system comprising: a processor; and a computer-readable storage medium operably connected to the processor, wherein the computer-readable storage medium contains one or more programming instructions for performing a method of automatically generating an assessment item, the method comprising: receiving one or more inputs from a user, generating one or more variables based on the one or more inputs, determining one or more relationships between at least two of the variables, and generating an assessment item based on the one or more variables and the one or more relationships.

21. A system for automatically generating an assessment item, the system comprising: a processor; and a computer-readable storage medium operably connected to the processor, wherein the computer-readable storage medium contains one or more programming instructions for performing a method of automatically generating an assessment item, the method comprising: receiving one or more inputs pertaining to the format of an assessment item, wherein the format of the assessment item comprises at least one event, selecting one or more variables for use in the assessment item, determining a relationship between variables assigned to each event, determining a format for the assessment item, and generating an assessment item based on at least the format for the assessment item and the relationship between variables assigned to the at least one event.

22. A method of automatically generating an assessment item, the method comprising: receiving one or more input parameters; generating a document structure based on the one or more input parameters; producing a logical schema using the document structure; and generating an assessment item based on the logical schema.

23. The method of claim 22 wherein generating a document structure comprises: building a mental model; and outlining the document structure based on the mental model.

24. The method of claim 23 wherein building a mental model comprises: selecting one or more semantic frames; generating a list of one or more events; and binding one or more variables across the one or more events.

25. The method of claim 23 wherein outlining the document structure comprises: generating one or more sentences for the mental model; determining a function for each sentence; and determining information to express in each sentence.

26. The method of claim 22, further comprising: storing the document structure in a file.

27. The method of claim 26 wherein the file is an Extendable Markup Language (XML) file.

28. The method of claim 22 wherein the document structure includes one or more of the following: one or more variables; one or more values for at least one of the one or more variables; a mental model structure; and an outline of a sequence of one or more sentences for the assessment item.

29. The method of claim 22 wherein producing a logical schema comprises: outlining a sentence structure for one or more sentences; and determining an information format for each sentence.

30. The method of claim 29 wherein determining an information format comprises one or more of the following: determining a verb type for each sentence; determining an ordering of one or more elements for each sentence; and determining one or more vocabulary sets to use for each element.

31. The method of claim 22, further comprising: storing the logical schema in a file.

32. The method of claim 31 wherein the file is an XML file.

33. The method of claim 22 wherein generating an assessment item comprises: parsing the logical schema; annotating the parsed logical schema with grammatical information; determining words and word forms based on the grammatical information; and outputting text representing the assessment item.

34. The method of claim 22 wherein an input parameter determines a language used for the assessment item.

35. A method of automatically generating an assessment item, the method comprising: assigning one or more mental model structure variables; defining one or more identity variables for a mental model structure; determining a task-relevant problem structure; defining a document format; and determining language variations.

36. The method of claim 35 wherein assigning one or more mental model structure variables comprises defining one or more of the following: one or more events; one or more distinct semantic frames; one or more participant types; and an event type for each event.

37. The method of claim 36 wherein determining a task-relevant problem structure comprises: determining a variable for which to solve for each event; determining an answer for each event; and determining one or more values for each variable.

38. The method of claim 35 wherein determining language variations comprises: selecting a sentence structure for each of one or more sentences; selecting a referent identification type for each of one or more participants; and determining a tense for each of one or more events.

39. The method of claim 35, further comprising: assigning a variable determining a language in which to generate the assessment item.

40. A system for automatically generating an assessment item, the system comprising: a processor; and a computer-readable storage medium operably connected to the processor, wherein the computer-readable storage medium contains one or more programming instructions for performing a method of automatically generating an assessment item, the method comprising: receiving one or more input parameters, generating a document structure based on the one or more input parameters, producing a logical schema from the document structure, and generating an assessment item from the logical schema.

41. A system for automatically generating an assessment item, the system comprising: a processor; and a computer-readable storage medium operably connected to the processor, wherein the computer-readable storage medium contains one or more programming instructions for performing a method of automatically generating an assessment item, the method comprising: assigning one or more mental model structure variables, defining one or more identity variables for a mental model structure, determining a task-relevant problem structure, defining a document format, and determining language variations.