Levels of Language Understanding in Natural Language Processing Systems

Natural Language Processing (NLP) systems often explain “language understanding” as a stack of interacting linguistic analysis levels, from the smallest units of sound to meaning in context across multiple sentences. Many courses and surveys describe these levels as phonological/phonetic processing, morphological word-structure processing, syntactic sentence-structure processing, semantic meaning interpretation, and pragmatic/discourse interpretation using context and world knowledge. A common framing is: acoustic/phonetic → morpho-syntactic → semantic-pragmatic, with additional emphasis on discourse coherence beyond the sentence level.3

Mermaid view of the idea of “levels” (not a strict implementation requirement, but a useful conceptual model):

Key takeaway: each higher level depends on representations produced by lower levels, but modern systems may learn these jointly rather than as hand-engineered pipelines.2

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩

2. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩ ↩²

3. NLP and Linguistics 1 — Just How Much ... (levels of linguistic analysis) - Lists phonetics/phonology/morphology/syntax/semantics/pragmatics/discourse as linguistic analysis levels. ↩

4. The 2022 Definitive Guide to Natural Language Processing (NLP) - Provides a level-by-level description including phonetic, morphological, lexical, syntactic, semantic, discourse, pragmatic. ↩

A “levels of analysis” lens for NLP

Below are the main levels typically taught in NLP and linguistics. Some curricula merge or subdivide levels (e.g., separating pragmatics from discourse), but the conceptual responsibilities remain similar.2

Core levels (common in NLP):

• Phonetics / Phonology: sound realization and sound-pattern modeling.2
• Morphology: word formation from morphemes; handles inflection/derivation.2
• Lexicon (often paired with morphology): part-of-speech and lemma/word sense cues.
• Syntax: how words combine into structures/relations in sentences.2
• Semantics: literal meaning of words/phrases/sentences.2
• Pragmatics: meaning shaped by speaker intent and situational context.2
• Discourse: coherence and relations across sentences/utterances (e.g., reference, progression).2

Important definitions

• Phonology
• Morphology
• Syntax
• Semantics
• Pragmatics
• Discourse

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. NLP and Linguistics 1 — Just How Much ... (levels of linguistic analysis) - Lists phonetics/phonology/morphology/syntax/semantics/pragmatics/discourse as linguistic analysis levels. ↩ ↩²

3. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩ ↩² ↩³ ↩⁴ ↩⁵

4. Natural language processing: state of the art, current trends ... (discourse level focus) - Explains that syntax/semantics operate over sentence-length units while discourse deals with more than one sentence for coherence. ↩

Level 1 — Phonetics and Phonology (speech/sound understanding)

Goal. Convert raw speech or text-adjacent signals into linguistic units that can support downstream processing. In many “three-area” descriptions, phonological/acoustic processing is separated from later syntactic/semantic interpretation.

• Phonetics deals with the physical/articulatory/acoustic properties of speech sounds.
• Phonology deals with how those sounds function as abstract units (e.g., patterns/rules within a language).

Typical NLP tasks

• Automatic speech recognition (ASR) and phoneme/word alignment
• Prosody cues (stress, rhythm) that can influence segmentation and meaning (often treated alongside “phonetic” processing in curricula)

Where this matters for understanding If the system mis-recognizes sound/word boundaries, later levels (morphology → syntax → semantics) receive corrupted inputs.

Keywords Phoneme ASR Prosody

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩ ↩²

Level 2 — Morphology (word structure)

Goal. Analyze word-internal structure—how words are built from morphemes—to recover grammatical and lexical information.

A course-style description notes that morphological knowledge relates to word construction from morphemes (small meaning-bearing units).2

Examples of what morphology enables

• Determining tense/aspect/number from inflection (e.g., “walk” vs “walked”)
• Handling derivation that changes meaning (e.g., “happy” → “unhappiness”)
• Supporting robust handling of productive word formation

Typical NLP tasks

• Stemming/lemmatization
• Morphological tagging (POS+features)
• Subword tokenization strategies (often used in modern neural NLP)

Keywords Morpheme Lemma Inflection

Note. Some teaching materials treat morphology and lexical aspects together; others keep them separate but both feed syntax/semantics.

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩ ↩²

2. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩

Level 3 — Lexicon (word identity cues)

Goal. Provide a bridge from surface word forms to structured representations like part-of-speech, lemmas, and (sometimes) word senses.

Many instructional framings list “lexical” cues alongside syntactic/semantic processing; for instance, some explain separate levels for lexical and syntactic organization.2

Typical NLP tasks

• Part-of-speech tagging (POS)
• Lemmatization (often overlaps with morphology)
• Word-sense disambiguation (more common when semantics is explicitly modeled)

Keywords POS tagging Word sense disambiguation Tokenization

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩

2. The 2022 Definitive Guide to Natural Language Processing (NLP) - Provides a level-by-level description including phonetic, morphological, lexical, syntactic, semantic, discourse, pragmatic. ↩

Level 4 — Syntax (sentence structure)

Goal. Determine how words combine and relate in a sentence (e.g., subject/object relations, phrase structure).

Instructional material emphasizes that syntactic knowledge is about organizing words into meaningful and correct sentences.2

Typical NLP tasks

• Parsing (constituency or dependency parsing)
• Grammar/structure prediction (e.g., relation extraction that uses syntax)
• Syntactic tagging/labeling (e.g., treebank-style annotations)

Why syntax is “understanding-relevant” Semantics depends on structure: “The dog chased the cat” differs from “The cat chased the dog,” even if the same words appear. Syntax determines which roles attach to which arguments.

Keywords Parse tree Dependency parsing Constituency parsing

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩

2. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩

Level 5 — Semantics (literal meaning)

Goal. Map structured inputs (from syntax and lexicon) to meaning representations.

A standard description states that semantic processing deals with the literal meaning of words/phrases/sentences.2

Typical NLP tasks

• Named entity recognition (NER) and entity linking (semantic categorization)
• Semantic role labeling (who did what to whom)
• Relation extraction (meaning-level relations)
• Textual entailment / semantic similarity
• Generating meaning representations (e.g., predicate-argument views in discourse tasks)

Keywords Semantic role labeling Entailment Predicate-argument

Mermaid: semantics as “meaning from structure”

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩

2. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩

3. Recognizing Implicit Discourse Relations in the Penn ... (Penn Discourse Treebank overview) - Introduces PDTB as discourse-level annotation, including implicit relations, using a predicate-argument view. ↩

Level 6 — Pragmatics (meaning in context)

Goal. Infer intended meaning using context, speaker goals, and general/world knowledge—not just sentence-internal meaning.

Course notes describe pragmatics as high-level knowledge about how sentences are used in different contexts and how context affects meaning.2

What pragmatics explains

• Indirect speech acts: “Can you pass the salt?” as a request, not a real question about ability
• Deixis and reference: “this/that,” “here/there,” and who “I/you” refer to
• Politeness strategies and implied commitments

Typical NLP tasks

• Dialogue understanding
• Intent detection (often blends semantics+pragmatics)
• Contextual question answering
• Contingent reasoning over conversational cues

Keywords Speech act Implicature World knowledge

Callout: Pragmatics is where “understanding” becomes human-like Type: tip

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect) - Defines phonological, morphological, syntactic, pragmatic components and describes levels of linguistic analysis. ↩

2. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩

Level 7 — Discourse (beyond the sentence)

Goal. Model how multiple sentences/utterances connect to create coherence, track entities, and resolve references.

Research descriptions emphasize that discourse-level NLP analyzes units larger than a sentence and focuses on coherence by interpreting relations between sentences. Resources like the Penn Discourse Treebank explicitly annotate discourse relations across spans, including implicit relations where cues are not overt.

Typical NLP tasks

• Discourse relation classification (e.g., because, contrast, elaboration)
• Coreference resolution (link “it/they” to antecedents)
• Coherence modeling and summarization with discourse structure
• Multi-sentence reading comprehension

Keywords Coreference resolution Discourse relation Coherence

Callout: Discourse often determines “what it’s about” Type: info

Footnotes

1. Natural language processing: state of the art, current trends ... (discourse level focus) - Explains that syntax/semantics operate over sentence-length units while discourse deals with more than one sentence for coherence. ↩

2. Recognizing Implicit Discourse Relations in the Penn ... (Penn Discourse Treebank overview) - Introduces PDTB as discourse-level annotation, including implicit relations, using a predicate-argument view. ↩

How these levels interact in an NLP system (conceptual pipeline)

Even when modern models are end-to-end, thinking in levels helps interpret what each component must solve. Some instructional material explicitly divides processing into broader stages: acoustic-phonetic, morphological-syntactic, semantic-pragmatic.

Mermaid “stage view” (high-level):

Footnotes

1. CSME 206A – Natural Language & Speech Processing Spring Semester (NLSP lect), three-level framing - States a divide into acoustic-phonetic, morphological-syntactic, and semantic-pragmatic processes. ↩

Summary map: what each level “answers”