Intelligent handwriting recording and analyzing system supporting translation process tracing and capability evaluation

Abstract

The invention relates to the technical field of computer data processing and mode recognition, and discloses an intelligent handwriting record analysis system supporting translation process traceability and capability assessment, which comprises the following steps: collecting multi-dimensional handwriting data generated when a user translates by using an intelligent pen; the Hidden Markov model decodes the handwriting into translation behavior primitive sequences such as smooth writing, hesitant pause and the like; automatically dividing semantic regions such as a main translation region and a draft region by a density clustering algorithm; performing space-time fusion attribution to generate a space-time attribution log for accurately recording behaviors, time and positions; and establishing a multi-dimensional capability index system containing translation efficiency, quality and strategies based on the log, and carrying out quantitative calculation and generating a visual user capability portrait. According to the method, the invisible translation cognition process is converted into traceable and quantifiable objective data, deep insight and comprehensive evaluation of the translation process are achieved, and a scientific and efficient analysis tool is provided for translation teaching and interpreter ability evaluation.

Description

Technical Field

[0001] This invention relates to the field of computer data processing and pattern recognition technology, specifically to an intelligent handwriting record analysis system that supports translation process tracing and ability assessment. Background Technology

[0002] Currently, translation competence assessment is a core component of translation teaching and research. Traditional assessment methods primarily rely on reviewing the final translation, a result-oriented, static evaluation. However, the translation process itself contains richer behavioral information about the translator at the cognitive level, such as decision-making, hesitation, and revision. With the widespread adoption of digital writing tools such as smart pens, it has become possible to fully record multi-dimensional handwriting data during the writing process, providing a new data foundation for objectively analyzing the dynamic process of translation.

[0003] Existing analytical applications of the acquired intelligent handwriting data primarily focus on the digitization and macro-level statistics of the writing results. These applications use handwriting recognition technology to convert handwritten translations into electronic text, facilitating automated comparison and evaluation with the original text. Simultaneously, the system can calculate global parameters such as the total time spent on the entire translation task, average writing speed, and total number of strokes. This method provides a certain degree of quantitative reference for translation evaluation, offering objective data support for assessing the translator's overall workload and fluency.

[0004] Existing macro-level analytical methods remain at a superficial level of process description, failing to distinguish the temporal differences between various cognitive activities. For example, whether an increase in total time stems from thoughtful conceptualization or repeated revisions is beyond the current system's ability to determine. Current technology lacks effective analysis of the writing space dimension, treating the entire writing plane as a homogenized area. The translator's exploratory writing in the draft area and the final strokes in the main translation area are conflated in the data. Ultimately, this separation of temporal and spatial information prevents the construction of models of users' actual translation strategies and hinders in-depth insights into the translator's skill set and limitations.

[0005] Therefore, this invention provides an intelligent handwriting record analysis system that supports translation process tracing and ability assessment, in order to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an intelligent handwriting record analysis system that supports translation process tracing and competence assessment. This system solves the problem that existing translation process analysis methods are unable to objectively and meticulously quantify and trace translators' cognitive behaviors and strategies in translation tasks.

[0007] To achieve the above objectives, the present invention provides an intelligent handwriting record analysis system that supports translation process tracing and competence assessment. This system includes:

[0008] The data acquisition module is used to collect smart handwriting data, including coordinates, pressure, and timestamp information, generated by the user when translating with a smart pen in real time.

[0009] The time dimension analysis module is configured to identify preset translation behavior primitives that represent the type of translation cognitive activity based on the intelligent handwriting data, and generate a sequence of translation behavior primitives with time information;

[0010] The spatial dimension analysis module is configured to segment the writing plane into one or more semantic regions through spatial clustering based on the coordinate information of the intelligent handwriting data.

[0011] The spatiotemporal fusion attribution module is configured to locate each translation behavior primitive in the sequence of translation behavior primitives to a specific one of the one or more semantic regions based on its spatiotemporal location information, and generate a spatiotemporal attribution log.

[0012] The capability profile generation module is configured to perform statistical analysis based on the spatiotemporal attribution logs to generate a multi-dimensional capability profile representing the user's translation ability.

[0013] In one specific embodiment, the specific workflow of the time dimension analysis module is as follows: First, each timestamp is calculated from the intelligent handwriting data. Dynamic feature vectors:

[0014] ;

[0015] The calculation method for each characteristic component is as follows:

[0016] Pen tip on timestamp instantaneous speed :

[0017] ;

[0018] in, For the pen tip at the time stamp Instantaneous velocity; and For timestamps The coordinates of the pen tip; and Previous time point The coordinates of the pen tip.

[0019] Pen tip on timestamp instantaneous acceleration :

[0020] ;

[0021] in, For the pen tip at the time stamp Instantaneous acceleration; For timestamps Instantaneous velocity; For the previous time point The instantaneous speed.

[0022] Pen tip on timestamp instantaneous jitter :

[0023] ;

[0024] in, For the pen tip at the time stamp The instantaneous jitter, i.e., the rate of change of acceleration; Timestamp Instantaneous acceleration; For the previous time point Instantaneous acceleration.

[0025] Handwriting traces in timestamps curvature :

[0026] ;

[0027] in, For handwriting trajectory in timestamp The curvature; and These are the pen tip coordinates. and timestamp The first derivative represents the velocity at... and Component of direction; and These are the pen tip coordinates. and Regarding time The second derivative of the acceleration represents the acceleration at... and The directional component.

[0028] Pen pressure on timestamp instantaneous rate of change :

[0029] ;

[0030] in, Pen tip pressure on time stamp The instantaneous rate of change; For timestamps Pressure value; Previous time point The pressure value.

[0031] Secondly, the time series of the dynamic feature vectors is used as the observation sequence and input into a pre-trained Hidden Markov Model (HMM), where the set of hidden states of the HMM is defined as corresponding to the set of types of the preset translation behavior primitives. Finally, the most probable hidden state sequence is solved using a decoding algorithm (such as the Viterbi algorithm) to generate the translation behavior primitive sequence.

[0032] Preferably, the types of translation behavior primitives include at least: fluent writing primitive, hesitant pause primitive, exploratory writing primitive, brute-force correction primitive, fine correction primitive, and backtracking positioning primitive.

[0033] In one specific embodiment, the specific workflow of the spatial dimension analysis module is as follows: First, extract the set of coordinate points of all pen stroke states (i.e., pressure values ​​greater than a preset threshold) in the intelligent handwriting data; second, process the set of coordinate points using a density-based spatial clustering algorithm to generate one or more initial handwriting region clusters; finally, label each initial handwriting region cluster to generate the one or more semantic regions.

[0034] Preferably, the semantic region labeling of each initial handwriting region cluster specifically refers to: statistically analyzing the distribution information of various translation behavior primitives falling within the initial handwriting region cluster, such as the total duration of various translation behavior primitives occurring; based on the distribution information and according to preset rules, labeling the initial handwriting region cluster as one of the main translation area, draft conception area, or alternative vocabulary area.

[0035] In one specific embodiment, the spatiotemporal attribution log generated by the spatiotemporal fusion attribution module includes, for each entry: the type of translation behavior primitive, the start and end times, and the semantic region identifier where it occurred.

[0036] Preferably, the system further includes a graph construction unit, used to: use the annotated semantic regions as nodes of a graph structure; identify the movement of the pen tip between different semantic regions as edges of the graph structure; thereby constructing a directed graph representing the user's translation strategy.

[0037] Preferably, the spatiotemporal fusion attribution module is further configured to: based on the spatiotemporal attribution log, identify, through a sequence pattern mining algorithm, a chain of behaviors consisting of multiple translation behavior primitives occurring sequentially in different semantic regions and having specific translation strategy significance.

[0038] In one specific embodiment, the capability profile generation module generates a multi-dimensional capability profile by quantifying a series of capability indicators, which include at least one or more of the following: cognitive fluency, syntactic organization ability, vocabulary extraction efficiency, and cognitive resource management.

[0039] Preferably, the quantification method of the ability indicator is: cognitive fluency. From the main translation area The total duration of fluent writing primitives occurring within the region, and all regions All translation actions occurring within the primitives The proportion of the total duration is determined as follows:

[0040] ;

[0041] in, As an indicator of cognitive fluency; The sum of the total duration of all fluent writing primitives that occur within the main translation area; The total translation time is the sum of the durations of all types of TBPs occurring in all regions.

[0042] The vocabulary extraction efficiency The specific behavioral chains related to lexical queries identified from the spatiotemporal attribution logs. The reciprocal of the average processing time is used to determine:

[0043] ;

[0044] in, As a metric for vocabulary extraction efficiency; To find the average value function; The duration of the word query-related behavior chain identified from the logs; This refers to the identified word query behavior chain pattern.

[0045] This invention provides an intelligent handwriting record analysis system that supports translation process tracing and competence assessment. It offers the following advantages:

[0046] 1. This invention, through its time-dimensional analysis module, transforms raw intelligent handwriting data into a dynamic feature vector sequence containing components such as speed and acceleration. It then uses a Hidden Markov Model to decode this into a sequence of translation behavior primitives composed of types such as fluent writing and hesitation / pause. This method objectively quantifies the translator's invisible cognitive activities into a series of analyzable data units, enabling precise, reproducible micro-behavioral analysis of the translation process down to the second level. This overcomes the subjectivity and ambiguity of traditional observation or interview methods, achieving objectification and fine-grained traceability of the translation cognitive process.

[0047] 2. This invention utilizes a spatial dimension analysis module and a density-based spatial clustering algorithm to process handwriting coordinates, dividing the writing plane into different physical regions. Crucially, it further analyzes the type and duration distribution of translation behavior primitives occurring within each region, automatically assigning functional semantic region labels such as the main translation area and the draft conception area to these regions. This enables the system to automatically reveal how translators plan and utilize their writing space, thereby inferring their specific translation strategies when handling complex sentence structures or vocabulary choices, and automatically identifying the translator's workspace organization patterns and translation strategies.

[0048] 3. This invention, through a spatiotemporal fusion attribution module, precisely locates each translation behavior primitive to its semantic region of occurrence, generating a spatiotemporal attribution log. Based on this log, the competency profile generation module can calculate a series of specific quantitative indicators, such as quantifying cognitive fluency by statistically analyzing the percentage of time spent on fluent writing primitives within the main translation area. This transforms the assessment of translators' abilities from a general qualitative judgment into a multi-dimensional competency profile composed of multiple objective indicators, allowing for longitudinal tracking and horizontal comparison. This provides reliable data support for precise translator training and development, and offers a data-driven, multi-dimensional quantitative assessment basis for translation capabilities. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0050] Figure 2 This is an internal flowchart of the time dimension analysis module of the present invention;

[0051] Figure 3 This is a schematic diagram of the state transition of the hidden Markov model of the present invention;

[0052] Figure 4 This is an internal flowchart of the spatial dimension analysis module of the present invention;

[0053] Figure 5 This is a schematic diagram illustrating the initial handwriting region cluster generation of the present invention;

[0054] Figure 6 This is a schematic diagram of the semantic region annotation process of the present invention;

[0055] Figure 7 This is a schematic diagram of the spatiotemporal attribution log generation process of the present invention;

[0056] Figure 8 This is a schematic diagram of the behavior chain mining process of the present invention;

[0057] Figure 9 This is a schematic diagram of the capability index system structure of the present invention;

[0058] Figure 10 This is a schematic diagram of the quantitative calculation process for the capability indicators of the present invention. Detailed Implementation

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see the appendix Figure 1 This invention provides an intelligent handwriting record analysis system that supports translation process tracing and capability assessment. The system may include: a data acquisition module, a time dimension analysis module, a spatial dimension analysis module, a spatiotemporal fusion attribution module, and a capability profile generation module.

[0061] The data acquisition module is used to collect, in real time, a continuous, timestamped sequence of handwriting dots generated by the user during the translation and writing process using a smart pen and a digital writing surface. This sequence of handwriting dots is formalized as smart handwriting data and serves as the system's unified input.

[0062] The intelligent handwriting data is transmitted to the time dimension analysis module and the spatial dimension analysis module for parallel analysis and processing.

[0063] The time-dimension analysis module receives the intelligent handwriting data and, based on the time-series information of the handwriting points, transforms the physical handwriting signal into a discrete sequence of translation behavior primitives representing cognitive activities by calculating dynamic features and performing sequence decoding. This sequence of translation behavior primitives is a series of behavioral units with start and end timestamps.

[0064] The spatial dimension analysis module also receives the intelligent handwriting data and, based on the spatial coordinate information of the handwriting points, segments the handwriting on the writing plane using a spatial clustering algorithm. It then labels the translation behavior primitive sequence output by the time dimension analysis module to generate one or more semantic regions with functional definitions.

[0065] The spatiotemporal fusion attribution module receives the translation behavior primitive sequence from the temporal dimension analysis module and the semantic region from the spatial dimension analysis module. Based on the coordinates of each translation behavior primitive, this module assigns it to a specific semantic region, thereby fusing the analysis results from both the temporal and spatial dimensions to generate a structured spatiotemporal attribution log.

[0066] The capability profile generation module receives spatiotemporal attribution logs from the spatiotemporal fusion attribution module. This module performs long-term statistical analysis and calculations on the data in the spatiotemporal attribution logs, and generates the final output—a multi-dimensional capability profile representing the user's translation ability—through a series of preset quantitative indicator models.

[0067] See attached document Figure 2 The data acquisition module in this embodiment of the invention functions to provide high-fidelity, structured input data for the subsequent analysis module. Physically, this module includes a smart pen with multiple built-in sensors and a digital writing surface used in conjunction with it.

[0068] To ensure the ability to capture subtle behavioral changes reflecting the user's cognitive activities, the smart pen has a data sampling rate of no less than 240Hz, meaning it can record at least 240 consecutive data points per second. The digital writing surface, for example, employs electromagnetic induction technology to locate the pen tip's position on a two-dimensional plane in real time.

[0069] The data acquisition module works as follows: when the user writes on the digital writing surface using the smart pen, the module continuously collects data from the pen tip at each time stamp using a preset sampling rate. The multi-dimensional physical state information is organized into a data point. .

[0070] like Figure 2 As shown, each data point The specific data structure can be defined as:

[0071] ;

[0072] in, This is the timestamp when the data point was collected, with a precision unit of milliseconds (ms), used to establish strict time series relationships; and For this data point, at the timestamp At that time, the pen tip is in the Cartesian coordinate system of the digital writing surface. axis coordinates and Axial coordinates; the accuracy of this coordinate system is determined by the resolution of the digital writing surface, for example, not less than 2540 lines per inch; For this data point, at the timestamp The pressure value applied by the pen tip to the digital writing surface. This pressure value is quantified as an integer, for example, within a range of 0 to 4095, where 0 indicates the pen tip is suspended and 4095 indicates the maximum pressure is reached; For this data point, at the timestamp At that time, the angle of inclination between the pen body and the normal of the writing surface of the number; For this data point, at the timestamp At that time, the azimuth angle of the pen body rotating around its own long axis.

[0073] The final output of the data acquisition module is a series of data points arranged in chronological order. A complete sequence of intelligent handwriting data The sequence It fully records all handwriting dynamics and morphological information throughout the entire translation and writing process, and serves as a unified data source for subsequent temporal and spatial dimension analysis modules.

[0074] See attached document Figure 3 The time dimension analysis module receives the intelligent handwriting data sequence provided by the data acquisition module. Next, the calculation of dynamic feature vectors is performed. The purpose of this step is to transform the original physical layer data, which is mainly composed of coordinates and pressure, into high-dimensional feature data that can more directly characterize the dynamic properties of the handwriting motion process.

[0075] For intelligent handwriting data sequences Each data point in Each time dimension analysis module calculates a corresponding dynamic feature vector. This dynamic feature vector It is a numerical vector containing multiple components, from which each timestamp is calculated. Dynamic feature vectors:

[0076] ;

[0077] The calculation method for each characteristic component is as follows:

[0078] Pen tip on timestamp instantaneous speed :

[0079] ;

[0080] in, For the pen tip at the time stamp Instantaneous velocity; and For timestamps The coordinates of the pen tip; and Previous time point The coordinates of the pen tip.

[0081] Pen tip on timestamp instantaneous acceleration :

[0082] ;

[0083] in, For the pen tip at the time stamp Instantaneous acceleration; For timestamps Instantaneous velocity; For the previous time point The instantaneous speed.

[0084] Pen tip on timestamp instantaneous jitter :

[0085] ;

[0086] in, For the pen tip at the time stamp The instantaneous jitter, i.e., the rate of change of acceleration; Timestamp Instantaneous acceleration; For the previous time point Instantaneous acceleration.

[0087] Handwriting trajectory at a point in time curvature :

[0088] ;

[0089] in, For handwriting trajectory in timestamp The curvature; and These are the pen tip coordinates. and timestamp The first derivative represents the velocity at... and Component of direction; and These are the pen tip coordinates. and Regarding time The second derivative of the acceleration represents the acceleration at... and The directional component.

[0090] Pen pressure on timestamp instantaneous rate of change :

[0091] ;

[0092] in, Pen tip pressure on time stamp The instantaneous rate of change; For timestamps Pressure value; Previous time point The pressure value.

[0093] See attached document Figure 4 In generating dynamic feature vector sequences Next, the time dimension analysis module continues to identify translation behavior primitives. This step is achieved through a pre-trained Hidden Markov Model (HMM), which is used to establish the probabilistic relationship between the observable dynamic feature vector sequence and the unobservable sequence of translation behavior primitives representing cognitive activities.

[0094] The Hidden Markov Model defines the following parameters:

[0095] ;

[0096] in, It is the initial state probability distribution vector, and its components are... This indicates that the first action at the start of the translation and writing process is the first... The probability of a translation behavior primitive; It is the state transition probability matrix, where the elements This indicates that the user's action at the current point in time is the [number]th [action / period]. Given a primitive translation behavior, the behavior at the next time point is the first... The probability of a translation behavior primitive. It is an observed probability distribution, whose elements This indicates that when the user's behavior is in the first position... When the translation behavior primitive state is reached, the specific value observed is: The probability of the dynamic feature vector. Due to the dynamic feature vector It is a multidimensional continuous variable, and the probability distribution of this observation is represented by a multidimensional Gaussian mixture model in a specific embodiment.

[0097] In embodiments of the present invention, the set of hidden states of the Hidden Markov Model is defined as corresponding one-to-one with a preset set of translation behavior primitives. For example, the set of translation behavior primitives may include: fluent writing primitive, hesitant pause primitive, exploratory writing primitive, brute-force correction primitive, fine correction primitive, and backtracking localization primitive.

[0098] The recognition process is as follows: The time dimension analysis module will analyze the dynamic feature vector sequence generated in the previous step. As an observation sequence, it is input into the pre-trained Hidden Markov Model. The goal of this module is to find a solution that generates the observation sequence with the maximum a posteriori probability. Hidden state sequence The mathematical expression for this solution process is:

[0099] ;

[0100] in, It is any possible sequence of hidden states; To maximize the parameter solution, this is a mathematical operator that finds the input parameters that maximize the subsequent objective function. It doesn't return the maximum probability value, but rather the specific sequence of translation behavior primitives that lead to that maximum probability value. For the posterior probability, this expression represents the probability in a Hidden Markov Model. and observation sequence Given all possibilities, this formula represents the probability of a specific sequence of primitives for a particular translation action occurring. The core of the formula is to find, among all possibilities, the one that maximizes this posterior probability.

[0101] To solve the above optimization problem, the time dimension analysis module employs the Viterbi algorithm. This algorithm uses dynamic programming to find a globally optimal path among all possible hidden state paths, i.e. .

[0102] Since the hidden states of the model have established a direct correspondence with the primitive types of the translation behavior, the optimal hidden state sequence decoded by the Viterbi algorithm is... This sequence of translation behavior primitives is directly used as the translation behavior primitive sequence. Each element in this sequence is labeled with the translation behavior primitive type at a specific point in time. This translation behavior primitive sequence is the final output of the temporal dimension analysis module, and is subsequently transmitted to the spatiotemporal fusion attribution module and the spatial dimension analysis module.

[0103] See attached document Figure 5 The spatial dimension analysis module receives the intelligent handwriting data sequence provided by the data acquisition module. Then, the analysis process is executed in parallel with the time dimension analysis module. Its first step is to generate an initial cluster of handwriting regions based on the spatial distribution of handwriting points.

[0104] The first step in this process is to extract the set of coordinate points used for clustering. The spatial dimension analysis module iterates through the intelligent handwriting data sequence. And based on each data point Pressure value in Perform a screening. A preset pressure threshold. It is used to distinguish between the pen's initial stroke and its suspended stroke. If and only if > At that time, the data point The coordinates of the point that is determined to be a valid pen stroke point The extracted coordinates are collectively formed into a two-dimensional coordinate point set. .

[0105] Obtain the set of coordinate points Next, the spatial dimension analysis module processes it using a density-based spatial clustering algorithm. In a specific embodiment, this algorithm is Density-Based Noisy Application Spatial Clustering (DBSCAN). Execution of this algorithm requires configuring two core parameters: neighborhood radius. , is used to define the scope of a point; and minPts, is used to define the minimum number of points required to form a dense region.

[0106] The clustering algorithm processes the following steps: from the set of coordinate points... Select any unprocessed point, calculate and find its... All other points within the neighborhood. If the number of points in the neighborhood is greater than or equal to minPts, the point is marked as a core point, and a new cluster is created. Subsequently, the algorithm recursively merges all density-reachable points in the neighborhood of this core point (i.e., the core point and other core points in its neighborhood) into the current cluster. If the number of points in the neighborhood of a point is less than minPts, the point is temporarily marked as a noise point. This process is repeated until a set is formed. All points were visited and categorized.

[0107] After the clustering algorithm is executed, the original set of coordinate points is obtained. A collection of points divided into one or more sets. Each set of points, consisting of core points and boundary points, is defined as an initial handwriting region cluster. These initial handwriting region clusters physically correspond to individual, densely written regions formed by the user on the writing plane. These generated initial handwriting region clusters will serve as the direct processing objects for the next step of semantic region annotation.

[0108] See attached document Figure 6 After generating the initial handwriting region clusters, the spatial dimension analysis module proceeds to annotate the semantic regions. The purpose of this step is to assign a semantic label representing the translation function to each initial handwriting region cluster that is purely based on spatial location segmentation.

[0109] This annotation process is achieved by associating the translation behavior primitive sequence output by the temporal dimension analysis module. The spatial dimension analysis module first performs annotation on each initial handwriting region cluster. Construct a regional behavior feature vector This vector is used to quantify the distribution of various cognitive behaviors occurring within a specific physical area.

[0110] Specifically, for the first A cluster of initial handwriting regions Its regional behavioral feature vector The generation process is as follows: The system traverses the complete sequence of translation behavior primitives. For each translation behavior primitive in the sequence, the system obtains all coordinate points during its occurrence. If these coordinate points spatially fall into the initial handwriting region cluster... Within the boundary, the duration of the translation action primitive is accumulated. The vector corresponds to the component of its type. The formal definition of this vector is:

[0111] ;

[0112] in, Representing the Types of translation behavior primitives in the region The total duration of events occurring within the period, It is the total number of types of translation behavior primitives.

[0113] After calculating the corresponding regional behavior feature vectors for all initial handwriting region clusters, the spatial dimension analysis module labels each region using a pre-defined rule engine. This rule engine contains a series of conditional judgments that classify regions based on the numerical distribution of their regional behavior feature vectors. In one specific embodiment, the rule engine includes the following rule: if a region... Within this context, the percentage of the total duration of fluent writing primitives in the total duration of all behaviors exceeds a preset threshold for the main region. (For example, 0.7), then this area is marked as the main translation area. If a region Within this context, the total time spent on exploratory writing primitives, violent correction primitives, and fine correction primitives exceeds a preset draft area threshold as a percentage of the total time spent on all actions. (For example, 0.6), then this area is marked as the draft concept area. If an area The total handwriting duration is relatively short, and the ratio of the duration of hesitation and pause primitives to the total handwriting duration exceeds a preset vocabulary threshold. If so, that area is marked as the candidate vocabulary area.

[0114] Through the above process, the final output of the spatial dimension analysis module is a series of semantic regions with clearly defined functional labels. These semantic regions are then transmitted to the spatiotemporal fusion attribution module for subsequent analysis and processing.

[0115] See attached document Figure 7 The spatiotemporal fusion attribution module integrates the analysis results from the temporal and spatial dimensions, assigning precise spatial location attributes to each identified translation behavior primitive. This module receives a sequence of translation behavior primitives from the temporal dimension analysis module and a set of labeled semantic regions from the spatial dimension analysis module.

[0116] The core processing procedure of this module is as follows: First, the spatiotemporal fusion attribution module traverses each translation behavior primitive in the sequence of translation behavior primitives. Each translation behavior primitive records its own type and start timestamp. and end timestamp .

[0117] Secondly, for the current translation behavior primitive, the module retrieves all timestamps from the original intelligent handwriting data sequence S. satisfy data points .

[0118] Then, the module extracts the coordinates of these retrieved data points. The semantic region to which each coordinate point belongs is determined. By statistically analyzing the distribution of all coordinate points in each semantic region, the semantic region in which the translation behavior primitive occurs is determined. In one specific embodiment, the translation behavior primitive is assigned to the semantic region containing the vast majority of its coordinate points.

[0119] Finally, the spatiotemporal fusion attribution module creates a log entry for this translation behavior primitive, which records its temporal and spatial attribution information. (A log entry) The structure can be defined as:

[0120] ;

[0121] in, The type of the primitive for this translation behavior; and These are the start and end timestamps of the translation action primitive, respectively; This is the identifier of the semantic region where the translation action primitive occurs, determined after the above attribution determination.

[0122] The spatiotemporal fusion attribution module repeats the above process until all primitives in the translation behavior primitive sequence have been processed. The final output of this module is a spatiotemporal attribution log organized in chronological order and containing multiple log entries. This log is then sent to the capability profile generation module as the data basis for subsequent statistical analysis and capability assessment.

[0123] See attached document Figure 8 In an optional embodiment, the spatiotemporal fusion attribution module can be further configured to perform behavior chain mining on the generated spatiotemporal attribution logs. This step aims to identify frequently occurring sequence patterns of multiple behavior events from a continuous stream of behavior events that can characterize a user's specific work habits or translation strategies; these patterns are defined as behavior chain patterns.

[0124] The first step in this data mining process is data serialization. The spatiotemporal fusion attribution module iterates through each log entry in the spatiotemporal attribution log 40 times. Furthermore, a binary tuple consisting of the translation behavior primitive type and the semantic region identifier in which it occurs is extracted from it. This binary tuple is defined as a spatiotemporal behavior event. All spatiotemporal behavioral events are arranged in their original temporal order, forming a spatiotemporal behavioral event sequence. This sequence is the direct input for executing subsequent mining algorithms.

[0125] The second step in this mining process is to apply a sequence pattern mining algorithm. The spatiotemporal fusion attribution module analyzes the spatiotemporal behavioral event sequences. A pre-defined sequence pattern mining algorithm, such as the Generalized Sequence Pattern (GSP) algorithm, is applied. The execution of this algorithm requires a pre-defined minimum support threshold, min_support. The support of a sequence pattern refers to the support of that pattern across the entire spatiotemporal sequence of behavioral events. The frequency of occurrence of a subsequence in the sequence.

[0126] The execution process of the GSP algorithm includes: First, processing the sequence... The process involves scanning and calculating the support of each independent spatiotemporal behavioral event. Events with support no less than `min_support` are identified and grouped into frequent 1-itemsets. Then, an iterative process begins. In the `m`-th iteration, all frequent (m-1) sequences identified in the `m-1`-th iteration are used to generate a candidate set of `m` sequences. Subsequently, the sequences are scanned again. First, calculate the support of all candidate m-sequences. Finally, remove all sequences with support below `min_support` from the candidate m-sequence set; the remaining sequences are the frequent m-sequences. This iterative process continues until no new frequent sequences can be generated.

[0127] The final output of this data mining process is a set containing all identified behavioral chain patterns with a support level of at least `min_support`. Each behavioral chain pattern is itself an ordered sequence of spatiotemporal behavioral events. For example, an identified pattern could be <(hesitation and pause, drafting area), (exploratory writing, drafting area), (fluent writing, main translation area)>. This set provides structured data for further analysis of users' macro-level translation strategies.

[0128] See attached document Figure 9 After receiving the spatiotemporal attribution log generated by the spatiotemporal fusion attribution module, the capability profile generation module's core function is to perform a series of quantitative calculations based on this log to generate a multi-dimensional capability profile. This calculation process is based on a preset capability indicator system.

[0129] The capability indicator system is a set of multiple independent, quantifiable indicators designed to describe user behavior in translation tasks from different dimensions. In one specific embodiment, the capability indicator system includes three primary dimensions: translation efficiency indicators, translation quality indicators, and translation strategy indicators.

[0130] Translation efficiency metrics measure the speed and fluency with which users complete translation tasks. This dimension can include the following secondary metrics:

[0131] Net writing rate The calculation formula is as follows:

[0132] ;

[0133] in, It is the total number of fluently written primitives in the spatiotemporal attribution log. It is the first The physical stroke length corresponding to a single smooth writing primitive. It is the total duration of the entire translation task.

[0134] Percentage of suspended behaviors The calculation formula is as follows:

[0135] ;

[0136] in, It is the total number of hesitant and paused primitives. It is the first The duration of a hesitation or pause primitive.

[0137] Translation quality metrics are used to reflect the stability and revision costs of user-generated translations through indirect behavioral representation. This dimension can include the following secondary metrics:

[0138] Correcting behavior frequency The calculation formula is as follows:

[0139] ;

[0140] in, It is the total number of two types of behavioral primitives: violent modification primitives and fine modification primitives.

[0141] Modify the intensity of behavior The calculation formula is as follows:

[0142] ;

[0143] in, It is the total duration of all modification-type behavior primitives. It is the total duration of all fluent writing primitives.

[0144] Translation strategy metrics describe how users allocate cognitive resources across different functional areas. This dimension can include the following secondary metrics:

[0145] Draft area utilization rate The calculation formula is as follows:

[0146] ;

[0147] in, It is the total number of all translation action primitives that occur within the draft conception area. It is the first The duration of each of these primitives.

[0148] The capability profile generation module continuously calculates and records the values ​​of the above indicators, ultimately generating a structured data report that reflects the user's long-term capability characteristics.

[0149] See attached document Figure 10After the capability indicator system is established, the core task of the capability profile generation module is to perform quantitative calculations of the indicators. This process transforms discrete, qualitative behavioral events in the spatiotemporal attribution log into continuous, quantitative values ​​that can objectively evaluate a user's capabilities.

[0150] The calculation process is executed by a built-in computing engine. This engine uses the spatiotemporal attribution log as its core data input and is configured to traverse the log, aggregate information, and perform precise calculations based on preset indicator formulas.

[0151] Specifically, the detailed process of quantitative calculation for each defined capability indicator is as follows:

[0152] 1. Calculation of translation efficiency indicators:

[0153] Net writing rate The calculation of this metric requires correlating the spatiotemporal attribution log with the original intelligent handwriting data sequence. The calculation engine first initializes a total handwriting length accumulator and a total duration timer. Total duration The engine determines this by calculating the difference between the end time of the last event and the start time of the first event in the log. Then, the engine iterates through the log, filtering out all log entries of type fluent writing primitives. For each such entry, the engine extracts information from the original data sequence based on its recorded start and end timestamps. The corresponding handwriting point subsequence is extracted. Then, the physical handwriting length corresponding to the fluent writing primitive is calculated by summing the Euclidean distances between all adjacent coordinate points in the subsequence. The physical handwriting lengths of all fluent writing primitives are summed and finally divided by the total duration. The net writing rate is obtained.

[0154] Percentage of suspended behaviors The calculation engine initializes a pause duration accumulator. The engine iterates through the logs, filtering out all log entries of type "hesitation pause". For each such entry, the engine calculates its duration (…). Then sum them up. After the traversal is complete, divide the summed total pause times by the total duration. The percentage of actions that were suspended was [not specified].

[0155] 2. Calculation of translation quality indicators:

[0156] Correcting behavior frequency The computation engine initializes a correction behavior counter. The engine iterates through the logs, determining the type of each entry. If the type is a brute-force correction primitive or a fine-grained correction primitive, the counter is incremented. After iteration, the total number of correction behaviors is divided by the total duration. (In minutes or hours) to obtain the frequency of corrective actions occurring per unit of time.

[0157] Modify the intensity of behavior The calculation engine initializes two duration accumulators: the corrected total duration and the fluent writing total duration. The engine iterates through the logs; if a log entry is a corrected primitive, its duration is added to the corrected total duration; if it's a fluent writing primitive, its duration is added to the fluent writing total duration. After iteration, the corrected total duration is divided by the fluent writing total duration to obtain the ratio. To prevent the denominator from being zero, when the fluent writing total duration is zero, this metric can be recorded as a preset maximum or invalid value.

[0158] 3. Calculation of translation strategy indicators:

[0159] Draft area utilization rate The calculation engine initializes a draft area activity duration accumulator. The engine iterates through the log, determining the spatial ownership of each log entry. If the entry occurs in a semantic region... For the draft / concept area, the engine adds its duration to the draft area activity duration accumulator. After traversal, the total accumulated draft area activity duration is divided by the total duration. This yields the utilization rate of the draft area.

[0160] By executing the above calculation process, the final output of the computing engine is a structured capability profile report. This report clearly lists the names of all capability indicators and their corresponding quantitative calculation results in key-value pairs, providing objective and accurate data support for subsequent user evaluations or feedback.

[0161] After completing the quantitative calculation of all capability indicators, the final step of the capability profile generation module 50 is to present these structured numerical results through a multi-dimensional, visual user interface. This presentation process aims to transform abstract data into intuitive insights, providing users with actionable feedback.

[0162] In one specific embodiment, the capability profile presentation interface consists of the following core visualization components:

[0163] 1. Comprehensive Ability Radar Chart: This component provides a macro-level, bird's-eye view of the user's translation capabilities. A radar chart is displayed at the center of the interface, with each vertex corresponding to one or more core primary indicators in the capability index system, such as translation efficiency, translation quality, and translation strategy. The system normalizes the calculated indicator values ​​(e.g., mapping to 0-100 points) and plots them on the corresponding coordinate axes of the radar chart. All plotted points are connected to form a polygonal region, whose shape and area visually demonstrate the balance and overall level of the user's various capabilities. As an optional feature, this radar chart can be overlaid with a preset benchmark polygon representing a "standard translator" or "user average level," allowing users to clearly identify their strengths and areas for improvement.

[0164] 2. Indicator Evolution Time Series Chart: This component tracks the trends of various user abilities over time. The interface provides drop-down menus or checkboxes, allowing users to select one or more secondary indicators of interest (e.g., net writing speed, correction behavior frequency). Once selected, the system retrieves the corresponding indicator values ​​for each of the user's translation tasks from historical data and plots a line graph with time on the horizontal axis and indicator values ​​on the vertical axis. This chart clearly reveals the user's progress, regression, or fluctuations in specific abilities, providing data support for targeted training.

[0165] 3. Behavioral Pattern Visualization Panel: This component will be activated if the system performs the aforementioned behavioral chain pattern mining. It visualizes the user's most frequently used translation workflow. In one embodiment, this visualization can take the form of a Sankey diagram. Nodes in the diagram represent different spatiotemporal behavioral events (e.g., hesitating in the draft area, writing fluently in the main translation area), and the lines connecting the nodes represent the temporal relationship between these events. The width of the lines is proportional to the frequency (i.e., support) of the behavioral chain pattern. Through this diagram, it is easy to identify whether the user prefers a macro-level strategy pattern such as brainstorming before writing or revising as they write.

[0166] 4. Handwriting-Behavior Correlation Retrospective Player: This component provides a highly sophisticated diagnostic tool. It deeply binds the final translated handwriting to a spatiotemporal attribution log. Users can replay the entire writing process like a video. During playback, the timeline highlights the current translation behavior primitive type with different colored markers (e.g., red for correction, green for fluent writing). Users can not only drag the timeline to jump to any point in time, but also directly locate the specific behavioral event and its context when a particular stroke occurred by clicking on it in the translated text. This feature enables reverse tracing from result to process, providing unprecedented convenience for review and reflection.

[0167] Through the organic combination of the above visualization components, the present invention ultimately generates and presents a dynamic, multi-layered, and interactive capability profile report, rather than a static data list.

[0168] In summary, this invention discloses a translation process analysis system and method based on intelligent handwriting. This system and method innovatively employs a technical approach that combines parallel analysis of temporal and spatial dimensions, ultimately performing spatiotemporal fusion attribution, by real-time collection of multi-dimensional intelligent handwriting data generated by users during translation using an intelligent pen. The aim is to achieve an objective, refined deconstruction and quantitative evaluation of the translation process.

[0169] In the temporal dimension, this invention overcomes the limitation of traditional post-translation evaluation methods that cannot provide insight into the translation process. By introducing a Hidden Markov Model, this invention can decode unobservable, discrete sequences of cognitive states—that is, sequences of translation behavior primitives—from observable, continuous, dynamic writing features. This makes it possible to automatically identify and label cognitive activities such as hesitation, correction, and exploration within seconds.

[0170] In the spatial dimension, this invention eliminates the reliance on pre-defined answer areas. Through a density-based unsupervised clustering algorithm, it automatically identifies physically clustered handwriting regions from the user's freehand writing. Furthermore, by projecting the behavioral distribution features identified in the temporal dimension onto these physical regions, this invention assigns a clear functional label to each region, forming semantic regions such as the main translation area and the draft conception area. This provides crucial contextual information for understanding how users organize their cognitive activities in physical space.

[0171] The core innovation of this invention lies in its spatiotemporal fusion attribution mechanism. By generating a spatiotemporal attribution log, this invention, for the first time, precisely and data-drivenly binds each micro-level translation behavior primitive to the macro-level semantic region in which it occurs. This log not only serves as a bridge connecting the process and the result but also forms the data foundation for subsequent in-depth analysis.

[0172] Ultimately, based on this detailed spatiotemporal log, this invention establishes a multi-dimensional capability indicator system and, through visualized capability profiles, transforms complex behavioral data into intuitive insights into user translation efficiency, quality, and strategies. From macroscopic radar charts to microscopic handwriting-behavior retrospective players, this invention provides a novel, data-driven, process-oriented objective analysis tool for the fields of translation capability assessment, translator training, and research on translation cognitive processes, possessing significant practical value and technological advancement.

Claims

1. An intelligent handwriting record analysis system supporting translation process tracing and ability assessment, characterized in that, include: The data acquisition module is used to collect smart handwriting data, including coordinates, pressure, and timestamp information, generated by the user when translating with a smart pen in real time. The time dimension analysis module is configured to identify preset translation behavior primitives that represent the type of translation cognitive activity based on the intelligent handwriting data, and generate a sequence of translation behavior primitives with time information; The spatial dimension analysis module is configured to segment the writing plane into one or more semantic regions through spatial clustering based on the coordinate information of the intelligent handwriting data. The spatiotemporal fusion attribution module is configured to locate each translation behavior primitive in the sequence of translation behavior primitives to a specific one of the one or more semantic regions based on the spatiotemporal location information of the translation behavior primitives, and generate a spatiotemporal attribution log. The capability profile generation module is configured to perform statistical analysis based on the spatiotemporal attribution logs to generate a multi-dimensional capability profile representing the user's translation ability.

2. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 1, characterized in that, The time dimension analysis module is specifically used for: The dynamic feature vector at each time point is calculated from the intelligent handwriting data; The time series of the dynamic feature vector is used as the observation sequence and input into a pre-trained Hidden Markov Model. The set of hidden states of the Hidden Markov Model is defined as corresponding to the set of types of the preset translation behavior primitives. The most probable hidden state sequence is obtained by decoding algorithm to generate the translation behavior primitive sequence.

3. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 2, characterized in that, The types of translation behavior primitives include at least: fluent writing primitive, hesitant pause primitive, exploratory writing primitive, brute-force correction primitive, fine correction primitive, and backtracking positioning primitive.

4. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 1, characterized in that, The spatial dimension analysis module is specifically used for: Extract the set of coordinate points for all pen stroke states from the intelligent handwriting data; The set of coordinate points is processed using a density-based spatial clustering algorithm to generate one or more initial handwriting region clusters; Each initial handwriting region cluster is labeled to generate one or more semantic regions.

5. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 4, characterized in that, The annotation of semantic regions for each initial handwriting region cluster specifically refers to: The distribution information of various translation behavior primitives falling within the initial handwriting region cluster is statistically analyzed; Based on the distribution information and according to preset rules, the initial handwriting area clusters are labeled as one of the main translation area, draft concept area, or alternative vocabulary area.

6. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 1, characterized in that, Each entry in the spatiotemporal attribution log includes: the type of translation behavior primitive, the start and end times, and the semantic region identifier where it occurred.

7. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 5, characterized in that, The intelligent handwriting record analysis system that supports translation process tracing and competence assessment also includes a graph construction unit, used for: The annotated semantic regions are used as nodes in the graph structure; The movement of the pen tip between different semantic regions is identified as edges of a graph structure; This allows us to construct a directed graph that represents the user's translation strategy.

8. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 1, characterized in that, The spatiotemporal fusion attribution module is also configured as follows: Based on the spatiotemporal attribution log, a sequence pattern mining algorithm is used to identify behavioral chains that consist of multiple translation behavior primitives occurring consecutively in different semantic regions and have specific translation strategy significance.

9. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 1, characterized in that, The multi-dimensional capability profile generated by the capability profile generation module is obtained by quantifying a series of capability indicators, which include at least one or more of the following: cognitive fluency, syntactic organization ability, vocabulary extraction efficiency, and cognitive resource management.

10. The intelligent handwriting record analysis system supporting translation process tracing and ability assessment according to claim 9, characterized in that, The quantification method for the aforementioned capability indicators is as follows: The cognitive fluency is determined by the proportion of the total time of fluent writing primitives occurring in the main translation area to the total translation time. The vocabulary extraction efficiency is determined by the average processing time of the vocabulary query-related behavior chain identified from the spatiotemporal attribution log.

Images