Novice programmers may not quickly understand a new or unfamiliar programming language. In order to help them understand the source code, existing studies have proposed approaches that translate a programming language into a pseudo-code. However, to the best of our knowledge, no studies have proposed translating Java, one of the well-known object-oriented programming languages, to pseudo-code. Many novice programmers learn the Java language because it is relatively less complicated than C++ and versatile in terms of practical use. Furthermore, the educational curriculum includes a Java course frequently. In this paper, we propose an approach to translate Java into pseudo-code at the code fragment level. We expect the proposed approach could help novice programmers to learn the Java language with the translated pseudo-code.
to line-by-line translating Java source code to A pseudo-code presents the logic of an algorithm pseudo-code using context for matching code patin a natural language imitating a programming terns and a template for translating matched patlanguage. The pseudo-code forms of natural lan- terns to pseudo-code. This approach does not need guage are more readable and understandable than the efort to gather data like machine learning techinterpreting the programming language directly. niques.
Therefore, novice programmers could easily un- This paper is organized as follows. Section 2 derstand the meaning of the actual source code describes related works for our study. Section 3 through pseudo-code. introduces our approach. Section 4 demonstrates
For this reason, several studies have proposed the demo results. Section 5 describes the discussion.
approaches to translating source code to pseudo- Finally, section 6 concludes this paper.
code [
Furthermore, most studies used machine learning 2. Related Works
techniques [
Although machine learning techniques are strong forward, we hypothesize the programming language is formal language enough to generate pseudo-code using rules like a compiler process.
Oda et al.[
These studies treated source code as natural lan- terns shown in Figure 2 express a unique sequence guage. However, their approaches have a potential of symbols and IDs. Finally, the templates, as shown shortcoming in that such an approach regards dis- in Figure 3, describe each pattern. tinct contexts despite the same ones with diferent identifiers. For instance, from a pseudo-code per- 4. Demonstration spective, the contexts "int a" and "int b" are equally treated as variable declarations. Treated as natu- 4.1. Approach Demonstration ral language, however, diferent vectors are created because of the diferent words ’a’ and ’b’.
Two study used contexts and templates similar
to our study [
To demonstrate our approach, we first depict following the steps. For step 1, as shown in Figure 4, we remove comments and blank lines from the entered source code and tokenize them. For step 2, shown in Figure 5, we replace all the code tokens with pre-defined symbols. For step 3, we replace the symbols with unique IDs, as shown in Figure 3. Approach 6. For step 4, as shown in Figure 7, we use the predefined patterns to find the longest pattern from Our approach, inspired by a compiler, to translating left to right in the symbol ID sequence, and if found, Java source code to pseudo-code is as follows: iteratively search from the following ID. If the pattern is not found, leave the symbol ID and search again from the following ID. Finally, as shown in Figure 8, we generate the pseudo-code with the pre-defined templates. • step-1) Remove all comments and blank lines
and tokenizes the source code • step-2) Changes code tokens to pre-defined symbols and keeps the original code tokens of each symbol • step-3) Changes the symbols to specific sym- 4.2. Prototype
bols by checking the original tokens We implemented a prototype of our translation ap• step-4) Finds patterns line-by-line from pre- proach as a web. Figures 9-11 show the demonstradefined patterns through the longest symbol tion of our approach. Figure 9 shows the textarea for inputting the Java source code. When click- sequence patterns based on pre-defined symbol ing the translate button, the inputted source code patterns. For example, the symbolic code above changes to symbolic code through the approach’s changes to [’(id id =)’, ’(keyword id)’, ’( ( id . id )’ ’( steps 1 to 4. For example, source code line “Scan- ) )’] which as pre-defined symbol sequence. Finally, ner scan = new Scanner (System.in);” is changed pseudo-code generates by extracting pseudo-code to “##Identifier## ##Identifier## = ##Keyword## for each pattern from a pre-defined pseudo-code ##Identifier## ( ##Identifier## . ##Identifier ## ) template, as shown in figure 11. For example, the ;”, as shown in Figure 10. Then, the symbols are (id id =) changed to “create Scanner type object transformed into specific symbols. After that, By variable input and assign with”, the (keyword id) using the longest symbol sequence-first match, our changed to “Scanner object”, and (id . id) changed approach replaces the symbolic code with symbol to “with parameters ( System . in )”. - method call lines (17.1%) (e.g., method();) - package/import lines (16.0%) (e.g., import java.util.*;) - variable definition lines (7.6%) (e.g., int a=1; except object) - method definition lines (6.6%) (e.g., int method () ) - if or else-if lines (5.5%)