{-# LANGUAGE DeriveGeneric #-} {-# LANGUAGE DuplicateRecordFields #-} {-# LANGUAGE NamedFieldPuns #-} {-# LANGUAGE OverloadedStrings #-} {-# LANGUAGE RecordWildCards #-} {-# LANGUAGE StrictData #-} {-# LANGUAGE NoImplicitPrelude #-} {-# OPTIONS_GHC -fno-warn-unused-imports #-} {-# OPTIONS_GHC -fno-warn-unused-matches #-} -- Derived from AWS service descriptions, licensed under Apache 2.0. -- | -- Module : Amazonka.Comprehend.Types.SyntaxToken -- Copyright : (c) 2013-2023 Brendan Hay -- License : Mozilla Public License, v. 2.0. -- Maintainer : Brendan Hay -- Stability : auto-generated -- Portability : non-portable (GHC extensions) module Amazonka.Comprehend.Types.SyntaxToken where import Amazonka.Comprehend.Types.PartOfSpeechTag import qualified Amazonka.Core as Core import qualified Amazonka.Core.Lens.Internal as Lens import qualified Amazonka.Data as Data import qualified Amazonka.Prelude as Prelude -- | Represents a work in the input text that was recognized and assigned a -- part of speech. There is one syntax token record for each word in the -- source text. -- -- /See:/ 'newSyntaxToken' smart constructor. data SyntaxToken = SyntaxToken' { -- | The zero-based offset from the beginning of the source text to the first -- character in the word. beginOffset :: Prelude.Maybe Prelude.Int, -- | The zero-based offset from the beginning of the source text to the last -- character in the word. endOffset :: Prelude.Maybe Prelude.Int, -- | Provides the part of speech label and the confidence level that Amazon -- Comprehend has that the part of speech was correctly identified. For -- more information, see -- -- in the Comprehend Developer Guide. partOfSpeech :: Prelude.Maybe PartOfSpeechTag, -- | The word that was recognized in the source text. text :: Prelude.Maybe Prelude.Text, -- | A unique identifier for a token. tokenId :: Prelude.Maybe Prelude.Int } deriving (Prelude.Eq, Prelude.Read, Prelude.Show, Prelude.Generic) -- | -- Create a value of 'SyntaxToken' with all optional fields omitted. -- -- Use or to modify other optional fields. -- -- The following record fields are available, with the corresponding lenses provided -- for backwards compatibility: -- -- 'beginOffset', 'syntaxToken_beginOffset' - The zero-based offset from the beginning of the source text to the first -- character in the word. -- -- 'endOffset', 'syntaxToken_endOffset' - The zero-based offset from the beginning of the source text to the last -- character in the word. -- -- 'partOfSpeech', 'syntaxToken_partOfSpeech' - Provides the part of speech label and the confidence level that Amazon -- Comprehend has that the part of speech was correctly identified. For -- more information, see -- -- in the Comprehend Developer Guide. -- -- 'text', 'syntaxToken_text' - The word that was recognized in the source text. -- -- 'tokenId', 'syntaxToken_tokenId' - A unique identifier for a token. newSyntaxToken :: SyntaxToken newSyntaxToken = SyntaxToken' { beginOffset = Prelude.Nothing, endOffset = Prelude.Nothing, partOfSpeech = Prelude.Nothing, text = Prelude.Nothing, tokenId = Prelude.Nothing } -- | The zero-based offset from the beginning of the source text to the first -- character in the word. syntaxToken_beginOffset :: Lens.Lens' SyntaxToken (Prelude.Maybe Prelude.Int) syntaxToken_beginOffset = Lens.lens (\SyntaxToken' {beginOffset} -> beginOffset) (\s@SyntaxToken' {} a -> s {beginOffset = a} :: SyntaxToken) -- | The zero-based offset from the beginning of the source text to the last -- character in the word. syntaxToken_endOffset :: Lens.Lens' SyntaxToken (Prelude.Maybe Prelude.Int) syntaxToken_endOffset = Lens.lens (\SyntaxToken' {endOffset} -> endOffset) (\s@SyntaxToken' {} a -> s {endOffset = a} :: SyntaxToken) -- | Provides the part of speech label and the confidence level that Amazon -- Comprehend has that the part of speech was correctly identified. For -- more information, see -- -- in the Comprehend Developer Guide. syntaxToken_partOfSpeech :: Lens.Lens' SyntaxToken (Prelude.Maybe PartOfSpeechTag) syntaxToken_partOfSpeech = Lens.lens (\SyntaxToken' {partOfSpeech} -> partOfSpeech) (\s@SyntaxToken' {} a -> s {partOfSpeech = a} :: SyntaxToken) -- | The word that was recognized in the source text. syntaxToken_text :: Lens.Lens' SyntaxToken (Prelude.Maybe Prelude.Text) syntaxToken_text = Lens.lens (\SyntaxToken' {text} -> text) (\s@SyntaxToken' {} a -> s {text = a} :: SyntaxToken) -- | A unique identifier for a token. syntaxToken_tokenId :: Lens.Lens' SyntaxToken (Prelude.Maybe Prelude.Int) syntaxToken_tokenId = Lens.lens (\SyntaxToken' {tokenId} -> tokenId) (\s@SyntaxToken' {} a -> s {tokenId = a} :: SyntaxToken) instance Data.FromJSON SyntaxToken where parseJSON = Data.withObject "SyntaxToken" ( \x -> SyntaxToken' Prelude.<$> (x Data..:? "BeginOffset") Prelude.<*> (x Data..:? "EndOffset") Prelude.<*> (x Data..:? "PartOfSpeech") Prelude.<*> (x Data..:? "Text") Prelude.<*> (x Data..:? "TokenId") ) instance Prelude.Hashable SyntaxToken where hashWithSalt _salt SyntaxToken' {..} = _salt `Prelude.hashWithSalt` beginOffset `Prelude.hashWithSalt` endOffset `Prelude.hashWithSalt` partOfSpeech `Prelude.hashWithSalt` text `Prelude.hashWithSalt` tokenId instance Prelude.NFData SyntaxToken where rnf SyntaxToken' {..} = Prelude.rnf beginOffset `Prelude.seq` Prelude.rnf endOffset `Prelude.seq` Prelude.rnf partOfSpeech `Prelude.seq` Prelude.rnf text `Prelude.seq` Prelude.rnf tokenId