Custom AMR Graph Builder

This Java project is designed to construct Abstract Meaning Representation (AMR) Graphs by parsing a specific, custom "Penman-like" notation.

Unlike standard Penman notation (which uses parentheses for nesting), this project is designed to process data exported from Google Sheets to CSV. It relies on indentation levels (represented by empty CSV cells) to determine the parent-child relationships between concepts.

📂 Project Structure

• AMRGraph.java: The main data structure representing a complete graph (sentence). It holds a map of Nodes to Edges and allows for deep cloning.
• AMRNode.java: A Java Record representing a single concept (e.g., "run", "boy").
• AMREdge.java: Represents a directed edge between nodes, containing the target node and the relation label (e.g., "ARG0", "frequency").
• IndentNode.java: An intermediate helper record. It captures the raw data from the CSV (Name, Relation, and Indentation Depth) before the graph is constructed.
• FileReader.java: The core parsing logic. It reads the CSV, calculates indentation based on empty delimiters, and uses a stack-like logic to reconstruct the tree structure.
• Main.java: Entry point for testing manual graph creation or running the file reader.

📝 Input Format (The "Google Sheets" Notation)

The parser expects a CSV file (exported from Google Sheets). The logic separates graphs by headers and determines hierarchy via column indentation.

1. Header Line

Each new graph must start with a line containing the ID (ending in .train or .test) and the full sentence.

• Format: ID, SENTENCE
• Example: 0463.train, SÜREKLİ İLGİLENDİ .

2. Node Lines

The lines following the header represent the nodes. The hierarchy is defined by how many empty commas precede the data.

• Syntax: Concept:Relation (The parser splits the string by :).
• Logic:
  • Level 0: The root of the graph.
  • Level 1: Direct children of the root.
  • Level 2: Children of the immediate Level 1 parent.

Example CSV Structure

Raw CSV View:

0463.train, SÜREKLİ İLGİLENDİ .
2/ilgilendi,
,1/sürekli:frequency
,o:ARG0

Spreadsheet View (Visual)

Column A (ID/Root)	Column B (Lvl 1)	Column C (Lvl 2)	Note
0463.train	SÜREKLİ İLGİLENDİ .		Header Line
2/ilgilendi			Root Node (Indent 0)
	1/sürekli:frequency		Child of ilgilendi (Indent 1)
	o:ARG0		Child of ilgilendi (Indent 1)

Resulting Logic:

1. The parser reads 2/ilgilendi at Indent 0.
2. It reads 1/sürekli at Indent 1 with relation frequency. It links ilgilendi -> frequency -> sürekli.
3. It reads o at Indent 1 with relation ARG0. It links ilgilendi -> ARG0 -> o.

⚙️ How It Works (Parsing Logic)

The FileReader.java uses a depth-based reconstruction algorithm (buildGraphNodes):

1. Intermediate Parsing: It reads the CSV line by line. It counts the empty strings in the split array to determine the indent integer. It parses the text to separate the Concept from the Relation (split by :) into an IndentNode.
2. Graph Construction:
   • It iterates through the list of IndentNodes.
   • If the nextNode has an indent exactly +1 greater than the currentNode, a direct edge is added.
   • If the nextNode has the same or smaller indent, the algorithm backtracks up the list to find the nearest node with indent == nextNode.indent - 1 to find the correct parent.

🚀 Usage

1. Preparing the Data

Ensure your .csv file is located in the accessible directory (default in code is files/amrtest7.csv).

2. Running the Code

You can run the FileReader main method to parse a CSV and print the resulting graph structures to the console.

// In FileReader.java
public static void main(String[] args) throws IOException {
    String csvFile = "files/amrtest7.csv";
    ArrayList<AMRGraph> graphs = processCSVFile(csvFile);
    for (AMRGraph graph : graphs) {
        graph.printGraph();
    }
}

Output

The printGraph() method displays the adjacency list for the graph:

Index: 0463.train
Sentence: SÜREKLİ İLGİLENDİ .
2/ilgilendi's children:
Relation: frequency, To: 1/sürekli
Relation: ARG0, To: o