Djalali, AJ et al."Inferencing underspecified natural language utterances in visual analysis."Proceedings of the 24th International Conference on Intelligent User Interfaces. 2019.BibTeXHandling ambiguity and underspecification in user utterances is a core challenge for natural language interfaces that support visual analytical tasks. This paper introduces a system that resolves partial utterances using syntactic and semantic constraints of the underlying analytical expressions, extending inferencing based on best practices in information visualization to generate useful visualization responses. Heuristics constrain the solution space of possible inferences, and ranking logic orders interpretations by relevancy and analytical usefulness.
Djalali, AJ."A constructive solution to the ranking problem in Partial Order Optimality Theory."Journal of Logic, Language & Information. 2017.BibTeXPartial Order Optimality Theory (PoOT) is a conservative generalization of classical Optimality Theory that enables the modeling of free variation and quantitative regularities without numerical parameters. Solving the ranking problem for PoOT—given a finite set of input/output pairs, how can a speaker learn the set of all grammars compatible with the data?—had remained an outstanding problem. This paper provides set-theoretic means for constructing the actual set of grammars compatible with an arbitrary dataset, thereby offering the first constructive solution to the PoOT ranking problem.
Djalali, AJ and Sven Lauer."A conceptual-epistemic perspective on model theory."Workshop on Models in Formal Semantics and Pragmatics, ESSLLI 2014, University of Tübingen. 2014.BibTeXThis talk argues that the mass/count distinction is neither directly observed in the world nor a matter of arbitrary linguistic convention, but rather a contrast in conceptualization. The analysis develops a conceptual-epistemic perspective on model-theoretic semantics, proposing that a level of conceptual representation mediates the relation between natural language and the world, with implications for how formal models should be interpreted.
Djalali, AJ."Synthetic logic."Linguistic Issues in Language Technology. 2014.BibTeXThis paper examines the logical foundations of MacCartney and Manning's natural logic system (NatLog), an algorithmic approach that combines insights from monotonicity calculi and syllogistic fragments to compositionally derive inferential relations between natural language sentences. The central contribution is a rigorous logical analysis showing that the system's table of lexical-semantic relations—including synonymy, antonymy, and hypernymy—has the underlying form of a syllogistic fragment and relies on a sort of generalized transitive reasoning. The work bridges proof-based semantic approaches with classical logical inference, providing theoretical grounding for computational natural language inference.
Djalali, AJ et al."Modeling expert effects and common ground using questions under discussion."Proceedings of the AAAI Workshop on Building Representations of Common Ground with Intelligent Agents. 2012.BibTeXThis paper presents a graph-theoretic model of discourse based on the Questions Under Discussion (QUD) framework, in which questions and assertions are treated as edges connecting discourse states in a rooted graph. The amount of common ground presupposed by interlocutors at any point corresponds to graphical depth. Using a new task-oriented dialogue corpus, the authors show that experts, presuming a richer common ground, initiate discourse at a deeper level than novices, enabling the quantification of speaker experthood relative to a fixed domain.
Djalali, AJ."If you own it, it exists; if you love it that says something about you not it: Semantically Conditioned Case in Finnish."Proceedings of the 30th West Coast Conference on Formal Linguistics. 2012.BibTeXThis paper addresses the apparent free variation between accusative and partitive case in Finnish, arguing that case assignment is a function of the lexical semantics of the verb, its object, and their composition. The key semantic distinction is existential commitment: verbs assigning accusative case entail the existence of their object's referent, while partitive-assigning verbs lack this commitment. The analysis shows that appeals to telicity and boundedness alone do not correctly predict Finnish stative behavior, and that existential commitment, encoded via meaning postulates, yields a more complete account of semantically conditioned case.
Djalali, AJ et al."Corpus Evidence for Preference-Driven Interpretation."Logic, Language and Meaning. 2012.BibTeXThis paper presents the Cards corpus of task-oriented dialogues and demonstrates how it can inform the study of goal- and preference-driven discourse interpretation. The authors report on experimental studies involving underspecified referential expressions and quantifier domain restriction, showing that the task itself gives rise to a notion of relevance that powerfully predicts both where speakers use underspecified expressions and how those expressions are interpreted. The findings support a model in which players' evolving approach to the central objective shapes their pragmatic strategies.
Djalali, AJ and Christopher Potts."A Formal Proof System for Natural Logic."Proceedings of the Workshop on Natural Logic, Proof Theory, and Computational Semantics, Stanford University. 2011.BibTeXThis paper establishes robust connections between MacCartney's NatLog system and linguistic theory by reformulating natural logic as a formal proof system. The authors rethink NatLog as a sequent calculus and prove completeness via representation, grounding the system's compositional approach to textual entailment in rigorous logical foundations. The work also explores redefining the semantics using large corpora, with applications to sentiment analysis and veridicality.
Djalali, AJ et al."What can be ground? Noun type, constructions, and the universal grinder."Proceedings of the 37th Annual Meeting of the Berkeley Linguistics Society. 2011.BibTeXThis paper examines constraints on the Universal Grinder, a coercion mechanism whereby count nouns surface as mass expressions in particular morphosyntactic contexts. The analysis argues against an unconstrained contextualist hypothesis, showing that noun type—defined by the ontological properties of a noun's denotation—systematically constrains which nouns can undergo grinding. The result is an account in which the interaction between noun type and construction determines the availability of mass interpretations for count nouns.
Djalali, AJ et al."Extension, ontological type, and morphosyntactic class: Three ingredients of countability."Conference on Empirical, Theoretical and Computational Approaches to Countability in Natural Language, Ruhr-Universität Bochum. 2010.BibTeXThis paper identifies three factors that jointly determine the mass/count status of nominals: the extension of a noun's denotation, the ontological type of the entities in that extension, and the morphosyntactic class of the noun. The analysis argues that no single factor is sufficient to predict countability, and that the interaction among all three ingredients is required to account for the full range of cross-linguistic variation in the mass/count distinction.
Djalali, AJ et al."Distinguishing the said from the implicated using a novel experimental paradigm."Semantics and Pragmatics: From Experiment to Theory, Palgrave Macmillan, pp. 74–93. 2009.BibTeXThis paper investigates whether speakers can systematically distinguish between what is said and what is implicated. Using a novel experimental paradigm in which subjects evaluate sentences from the perspective of a strictly literal interpreter, the authors show that inferences from gradable adjectives are less frequently incorporated into truth-conditional meaning than those from quantifiers, cardinal numerals, and rank orderings. The gradience of the data suggests that a more nuanced account of the relationship between generalized conversational implicatures and truth conditions is needed.
Djalali, AJ and Stefan Kaufmann."Probabilistic inference in dynamic semantics."Proceedings of the 10th Symposium on Logic and Language, pp. 99–107. 2009.BibTeXThis paper explores the integration of probabilistic reasoning into the framework of dynamic semantics, where meaning is characterized in terms of its potential to change the information state of a discourse context. The work develops a formal model that extends classical dynamic semantic frameworks with probability distributions, enabling the representation and computation of graded inferences in natural language. The approach provides a unified treatment of categorical and probabilistic inference within a single compositional system.
Djalali, AJ et al."The effect of focus on bridging inferences."LSA Annual Meeting, Chicago, IL. 2008.BibTeXThis paper investigates how prosodic focus affects bridging inferences in discourse comprehension. A perception experiment tests the effects of shifting prosody on potentially ambiguous bridging inferences, finding that listeners systematically vary their referent assignment based on heavy focus on specific lexical items. The results suggest that bridging and contrastive inferences pattern similarly, indicating an underlying connection between the two phenomena that may be captured by a formal theory of discourse and context.
Djalali, AJ et al."The effects of scale type and salience on the interpretation of scalar implicature."LSA Annual Meeting, Linguistic Society of America. 2008.BibTeXThis paper investigates whether scale type (open vs. closed) and salience of alternatives affect the strength of scalar implicature. The experimental results show that expression type—numerals, quantifiers and modals, and gradable adjectives—is a more reliable predictor of implicature strength than boundedness. The gradience of the data challenges the assumption that scalar implicature is a uniform phenomenon and motivates a more nuanced account of the relationship between generalized conversational implicatures and truth conditions.
Djalali, AJ et al."Distinguishing among contextually-determined aspects of utterance meaning: An empirical investigation."LSA Annual Meeting, Linguistic Society of America, Anaheim, CA. 2007.BibTeXThis paper addresses the theoretical distinction between context-dependent and context-independent aspects of utterance interpretation, a standard assumption in current theories of meaning. The authors empirically investigate whether speakers can systematically distinguish these aspects, examining possible contextual and grammatical factors that influence the degree of incorporation and the strength of conversational implicatures. The findings suggest that the distinction between what is said and what is implicated is gradient rather than categorical.
Djalali, AJ."The syntax and semantics of ordinary comparative constructions in English."PhD Dissertation, Department of Linguistics, Stanford University. 2015.BibTeXThis dissertation argues that standard degree-based analyses of comparative constructions in English are fundamentally flawed and develops an alternative account grounded in Barker and Shan's continuation semantics and Muskens' simplified Montague logic. Rather than treating degrees as proper objects in the semantic ontology, the work proposes a transparent and minimal semantic representation language that avoids the complex covert operators typically posited in comparative syntax. The resulting framework provides a more empirically adequate and formally elegant treatment of ordinary comparatives while dispensing with the unnecessary machinery of traditional degree semantics.
Djalali, AJ et al."Using natural language expressions to define data visualization calculations that span across multiple rows of data from a database."US-11550853-B2. Issued January 10, 2023.BibTeXDescribes a method for processing natural language commands to generate data visualizations involving table calculations. The system identifies relevant data fields, aggregates values across specified time periods, computes differences between consecutive period values, and generates visualizations showing these computations. An intermediate language called ArkLang applies syntactic and semantic constraints to resolve ambiguous or incomplete natural language expressions.
Djalali, AJ et al."Methods and systems for inferring intent and utilizing context for natural language expressions to modify data visualizations in a data visualization interface."US-11314817-B1. Issued April 26, 2022.BibTeXDescribes a system that enables data visualization modifications through natural language commands by inferring user intent. The system handles both explicit requests and implicit intent, adjusting visual variables such as data field mappings and chart characteristics accordingly. It maintains continuity by preserving relevant elements from previous visualizations while incorporating new analytical requests, creating an interactive analytical workflow without dead ends.
Djalali, AJ et al."Visually correlating individual terms in natural language input to respective structured phrases representing the natural language input."US-11301631-B1. Issued April 12, 2022.BibTeXDescribes a system that provides incremental visual feedback as users type natural language queries for data visualization. The system generates tokens from entered terms, maps them to analytical concepts in a lexicon, and displays corresponding structured phrases while simultaneously emphasizing matching terms through visual effects. Unrecognized terms are de-emphasized, reducing cognitive burden and creating a more efficient human-computer interface for data exploration.
Djalali, AJ et al."Analyzing underspecified natural language utterances in a data visualization user interface."US-11244114-B2. Issued February 8, 2022.BibTeXDescribes a system that addresses the challenge of incomplete or vague natural language instructions in data visualization applications. The system parses input into an intermediate representation using context-free grammar rules, automatically infers missing details such as data fields and aggregation methods, and translates the complete specification into database queries. The intermediate language ArkLang uses syntactic and semantic constraints to resolve ambiguity, supporting visualizations ranging from maps and bar charts to scatter plots.
Djalali, AJ et al."Determining levels of detail for data visualizations using natural language constructs."US-11055489-B2. Issued July 6, 2021.BibTeXDescribes a system that accepts natural language commands and creates intermediate expressions using context-free grammar and a semantic model of the data source's fields. The system identifies aggregation type, target data fields, and grouping terms within the command, then translates these into database queries to retrieve appropriately aggregated datasets. The approach supports multiple aggregation levels within a single visualization, enabling intuitive data exploration through natural language.
Djalali, AJ et al."Analyzing natural language expressions in a data visualization user interface."US-11048871-B2. Issued June 29, 2021.BibTeXDescribes a method for automatically updating natural language expressions in data visualizations to maintain consistency. When a user modifies a term in one phrase, the system automatically updates dependent phrases to prevent errors—for example, adjusting filter conditions when switching between data fields of different types. The approach reduces cognitive burden by eliminating the need for users to manually track dependencies between different portions of their analytical input.
Djalali, AJ et al."Natural language interface for building data visualizations, including cascading edits to filter expressions."US-10902045-B2. Issued January 26, 2021.BibTeXDescribes a system for building and modifying data visualizations using natural language with automatic cascading updates. The system analyzes natural language input to identify distinct phrases and their relationships, and when detecting a modification to one phrase, automatically updates any dependent phrases without manual intervention. When users switch data fields in filter expressions, the system intelligently replaces comparison values appropriate to the new field's data type.
Djalali, AJ et al."Inferring Intent and Utilizing Context For Natural Language Expressions in a Data Visualization User Interface."US-20220253481-A1. Filed April 25, 2022.BibTeXDescribes a system where a computing device receives natural language commands to modify an existing data visualization. The system extracts keywords, determines the user's intended modifications, generates an updated visual specification, and executes queries to retrieve relevant data. The technology distinguishes between explicit user intent and implicit intent inferred from context, with explicit intent taking priority, enabling coherent iterative analytical workflows.
Djalali, AJ et al."Using natural language constructs for data visualizations."US-20210319186-A1. Filed June 25, 2021.BibTeXDescribes a system that converts natural language input into an intermediate expression using a context-free grammar and semantic data model, bridging natural language queries and formal database operations. The system identifies key components within commands including aggregation types, target data fields, and grouping terms, then translates these into executable database queries. The intermediate language ArkLang reduces user cognitive burden during data exploration while supporting multiple aggregation levels within a single visualization.
