ternlang-core 0.3.3

Compiler and VM for Ternlang — balanced ternary language with affirm/tend/reject trit semantics, @sparseskip codegen, and BET bytecode execution.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35

// Module:  stdlib/nlp/ner.tern
// Purpose: Named Entity Recognition
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Extracts entities. 'tend' handles uncertain boundaries.

struct NERTagger {
    model: trittensor<4 x 4>
}

fn bio_encode(token_state: trit) -> trit {
    // B-Begin (affirm), I-Inside (tend), O-Outside (reject)
    match token_state {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}

fn tag_trit(tagger: NERTagger, seq: trittensor<4 x 1>) -> trit {
    @sparseskip
    let logits: trittensor<4 x 1> = tagger.model * seq;
    return bio_encode(logits[0, 0]);
}

fn entity_gate_trit(confidence: trit) -> trit {
    // Uncertain spans = tend
    if confidence == tend { return tend; }
    match confidence {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}