Information processing device, method, and program for performing pseudospace formation and reference frame inference based on text input including ASCII layout.

Abstract

This invention provides an information processing device and method that enables spatial reference responses based on a user reference frame (front, back, left, right, etc.) from text input including ASCII layout, without requiring sensor input or coordinate reconstruction, and that responds safely and explainably even when the input contains ambiguous, contradictory, or incomplete information. [Solution] The method extracts entities and relative relationships from text input to generate a normalized relational representation (CRF), and forms a pseudospatial representation that holds the CRF as a relational constraint network. The reference frame is primarily the UserFrame, and maintains a correspondence with the GlobalFrame only when orientation, etc., is explicitly stated in the input. Based on the interpretation policy, the ASCII layout estimates boundaries (walls), etc., and adds them to the CRF as movement constraints and / or occlusion constraints and directional cues. Depending on the uncertainty or safety importance of the inference result, confirmation requests, conditioning, or output suppression are performed.

Description

Technical Field

[0001] The present invention relates to virtual space formation, spatial reference inference, and dialogue response generation based on text input including natural language and / or ASCII layout.

Background Art

[0002] Many methods for spatial understanding and navigation are based on sensor input and coordinate restoration (map generation).

[0003] However, in environments where sensors are not available, environments with strong privacy and computing resource constraints, or environments where only text is given, it is necessary to realize a spatial reference response from text input. Furthermore, since text input may include ambiguity, omission, and contradiction, a conclusive response may be dangerous or inappropriate in some cases.

Prior Art Documents

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The present invention solves at least some of the following problems. (1) Generating a spatial reference response from only text without requiring coordinate restoration. (2) Performing inference that is not easily broken for ambiguous, contradictory, and insufficient information. (3) Enabling confirmation requests / conditioning / suppression according to uncertainty or safety importance. (4) Suppressing inference costs while maintaining explainability. (5) Including ASCII interpretation as a core function while making implementation optimization confidential and updatable. [Means for solving the problem]

[0006] One aspect of the present invention involves analyzing text input to extract entities and relative relationships and generating a normalized relational representation (CRF).

[0007] Based on CRF, coordinate reconstruction is not required, and a pseudospatial representation is formed as a set of relational constraints (constraint network).

[0008] The reference frame is managed primarily using the UserFrame, and maintains its association with the GlobalFrame only when the direction or other parameters are explicitly specified in the input.

[0009] Inference is performed based on relational constraints of the pseudospatial representation, and the response may present evidence (part of the constraints or methods of reference) in addition to the conclusion. Prior to output, confirmation requests, conditioning, or output suppression are performed depending on uncertainty or safety importance. Based on the interpretation policy, the ASCII layout estimates boundaries (walls), etc., and adds movement constraints and / or occlusion constraints and directional cues to the CRF. [Effects of the Invention]

[0010] According to the present invention, (1) spatial reference response can be achieved from text alone, without requiring sensor input or coordinate reconstruction.

[0011] Furthermore, (2) ambiguity, contradictions, and deficiencies can be addressed with multiple hypotheses, confidence levels, and additional questions to reduce failures; (3) erroneous conclusions can be suppressed by output control according to uncertainty / safety importance; (4) explainability can be easily provided by underlying relational constraints; and (5) inference costs can be reduced through primary estimation and reuse. [Brief explanation of the drawing]

[0012] [Figure 1] This is a functional block diagram of the PSR system.

[0013] [Figure 2]This is a processing flow diagram (input → CRF → pseudospace → reference frame → inference → response → output control).

[0014] [Figure 3] This is a conceptual diagram of the CRF schema and ASCII → constraint assignment (candidate generation → evaluation → confirmation / questioning). [Modes for carrying out the invention]

[0015] This specification describes the present invention in a way that allows for its implementation by explicitly defining the processing sequence (framework) and data structure (minimal schema). However, the combinations of rules, dictionaries, and learners that can be used to interpret the ASCII layout, the specific weights, thresholds, and priorities of the evaluation criteria, exception handling, calibration procedures, and operational management information can be appropriately modified depending on the input distribution and application. Therefore, this description is not limited to specific examples, but does not depart from the spirit of the invention.

[0016] (Overall Configuration) The information processing device comprises an input unit, a descriptive analysis and normalization unit, a pseudo-space formation unit, a reference frame management unit, an inference unit, a response generation unit, and an output control unit. Each unit may be implemented in a single device, or it may be distributed across a client / server or cloud / edge.

[0017] (Processing Flow) The system takes text input containing natural language and / or ASCII layout, extracts entities and relative relationships, and normalizes them into a CRF. An interpretation policy is applied to the ASCII layout, and boundaries (walls), etc., are estimated and added to the CRF as movement constraints, occlusion constraints, directional cues, etc. The relational constraints of the CRF are integrated as a constraint network to form a pseudospatial representation (coordinate reconstruction is not mandatory). UserFrames are managed as primary references, and the correspondence with GlobalFrames is maintained only when orientation, etc., is explicitly stated in the input. In response to queries, the system derives inference results by referring to the constraint network and generates a response statement that includes conclusions and justifications (part of the constraints or the method of reference). Prior to output, confirmation requests / conditions / suppressions are made depending on uncertainty or safety importance before output.

[0018] (CRF Minimum Schema) CRF can include at least Entity (entity ID, type), Predicate (relation predicate), Qualifier (at least one of distance, angle, time, modality, etc.), Frame (UserFrame, GlobalFrame if necessary), Confidence (confidence level), Conflict / Unknown (conflicting, uncertain, or missing information), Source (derived from natural text, ASCII, estimation, question and answer, etc.). The specific design of Confidence, Conflict / Unknown, and Source (weights, thresholds, classification granularity, calibration procedures, etc.) can be changed according to the application and input distribution and can be managed as operation optimization assets.

[0019] (ASCII → Constrained: Minimum Procedure) Detect line segment-like expressions, closed region-like expressions, partition boundary-like expressions, labels, and arrow-like expressions or asymmetric arrangements in ASCII. Generate boundary (wall) candidates, passage or opening candidates, and orientation (direction) candidates from the extracted primitives. Assign boundary candidates to CRF as movement constraints and / or occlusion constraints, and assign orientation candidates to CRF as direction cues for UserFrame. When multiple interpretation candidates are established, select candidates through evaluation including consistency, confidence level, and safety. If it cannot be determined, perform additional questions (confirmation requests) or conditional responses. The evaluation criteria (penalties, weights, thresholds, priorities), exception handling, and branching rules for additional question templates can be changed according to the input distribution and operation requirements.

[0020] (Inference: Minimum Basis) The inference part can refer to the constraint network and perform at least one of the following: visibility determination based on occlusion constraints, reachability determination based on movement constraints and passage candidates, and reinterpretation of relative relationships based on the viewpoint transformation of UserFrame. When there are contradictions or deficiencies, handle uncertainties through maintaining multiple hypotheses, prioritization based on confidence level, and requesting additional information.

[0021] (Output Control) The output control unit performs at least one of confirmation request, conditioning, or output suppression according to the uncertainty or safety importance of the inference result. The classification of safety importance, suppression threshold, etc. can be changed according to the application and can be managed as operating assets.

[0022] (Inference Cost Reduction) When sufficient reliability is obtained by primary estimation, high-cost processing is suppressed, or the inference cost is reduced by reusing intermediate representations / inference results. The conditions for reuse, cache strategy, threshold, etc. can be changed in the implementation.

[0023] (Compliance) One aspect of the present invention can be implemented as a configuration that satisfies predetermined requirements such as entity and relationship extraction, holding of reference frames (UserFrame), constraint network representation that does not require coordinate restoration, imposition of ASCII layout constraints, handling of uncertainty, etc. Furthermore, it can be implemented as a configuration that satisfies extended requirements including qualifiers such as distance or angle, inferences such as occlusion or reachability, contradiction detection, reliability assignment, etc. These specific test items, thresholds, evaluation criteria, and operation optimizations depending on the input distribution can be changed according to the application.

[0024] (Privacy and Security) Since the input is text, it can be implemented as a configuration that does not handle personal information derived from sensors. When the input text may contain personal information, privacy protection can be achieved by at least one of anonymization or masking, local processing or distributed processing, restriction of storage period, access control. When performing storage or reuse of CRF and inference results, security can be ensured by at least one of forgery detection, integrity verification, audit logs, etc. These specific methods and operation parameters can be changed according to the application, jurisdiction, and operation requirements and can be managed as operating assets.

Explanation of Signs

[0025] 11 Input Unit 12 Description Analysis / Normalization Unit (CRF Generation Unit) 14 Pseudo-Space Formation Unit 15 Reference Frame Management Unit 16 Reasoning part 17 Response generation unit 18 Output control unit CRF (Canonical Relational Form) UserFrame (User-defined reference frame) GlobalFrame Azimuth Reference Frame (Optional)

Claims

1. An information processing device that forms a pseudospace based on text input including an ASCII layout and generates a spatial reference response based on the pseudospace, wherein at least, (A) An input unit that acquires the text input, (B) A descriptive analysis and normalization unit that extracts entities and relative relationships from the text input, generates a normalized relational representation (CRF: Canonical Relational Form), estimates boundaries (walls) from at least some of the line segments, sections, repeating symbols, or boundary representations included in the ASCII layout based on an interpretation policy for the ASCII layout, assigns these as movement constraints and / or occlusion constraints to the CRF, and handles arrow-like symbols, direction indicator symbols, or asymmetry of arrangement as clues for orientation, (C) A pseudo-space formation unit that forms a pseudo-space representation as a set of relational constraints, without requiring coordinate restoration based on the CRF, (D) A reference frame management unit that manages reference frames including at least a UserFrame and maintains an association with a GlobalFrame only when the orientation is explicitly stated in the text input, (E) An inference unit that derives an inference result from the pseudo-spatial representation in accordance with the reference frame, (F) A response generation unit that generates a response sentence based on the inference result, An information processing device characterized by comprising:

2. The information processing device according to claim 1, wherein the CRF includes at least one of the following: (a) an entity, (b) a relational predicate, (c) a frame of reference information, (d) a confidence level, and (e) information relating to a contradiction or deficiency.

3. An information processing apparatus according to claim 1 or 2, further comprising an output control unit that, prior to outputting (G), performs at least one of a confirmation request, conditioning, or output suppression depending on the uncertainty or safety importance of the inference result.

4. The information processing device according to Claim 1, wherein the inference unit includes at least one of (a) visibility determination based on occlusion constraints, (b) reachability determination based on movement constraints, and (c) reinterpretation of relative relationships based on viewpoint transformation of UserFrame.

5. An information processing method comprising an information processing device forming a pseudospace based on text input including an ASCII layout and generating a spatial reference response based on the pseudospace, wherein the information processing device (A) The process of obtaining the text input, (B) A step of extracting entities and relative relationships from the text input and generating a CRF, and based on the interpretation policy for the ASCII layout, estimating boundaries (walls) from at least some of the line segments, sections, repeating symbols, or boundary representations included in the ASCII layout, and assigning them to the CRF as movement constraints and / or occlusion constraints, and treating arrow-like symbols, direction indicator symbols, or arrangement asymmetry as clues for orientation, (C) A step of forming a pseudo-spatial representation as a set of relational constraints based on the CRF, (D) A step of managing a reference frame that includes at least a UserFrame, and maintaining the association with the GlobalFrame only when the orientation is explicitly stated in the text input, (E) A step of deriving an inference result from the pseudo-spatial representation in accordance with the reference frame, (F) A step of generating a response sentence based on the inference result, An information processing method characterized by including

6. A program for causing a computer to execute the information processing method described in claim 5.

7. An information processing apparatus according to Claim 1, wherein the inference unit reduces computational load and / or inference cost by (a) suppressing the execution of high-cost processing when the reliability is determined to be above a predetermined level by lightweight first estimation, or (b) reusing intermediate representations or inference results for the same or similar inputs.

Images