---
title: "Hebrew & Aramaic — papers — Elamur Research"
description: "Hebrew & Aramaic: 482 papers of the Elamur Research library, the most cited first. Among them: Colorless Green Recurrent Networks Dream…"
canonical: "https://research.elamur.ai/free/en/topics/hebrew-aramaic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/topics/hebrew-aramaic/)

[home](https://md.research.elamur.ai/free/en/) / [topics](https://md.research.elamur.ai/free/en/topics/) / Hebrew & Aramaic

# 🕎 Hebrew & Aramaic

482 papers · page 1 of 6 · 1–25 · the most cited first · [all topics](https://md.research.elamur.ai/free/en/topics/)

Biblical and Modern Hebrew, Aramaic and Syriac: the languages and the texts.

A survey · about 20 minutes to read · every paper named here has its own page in this library

**In this article**

1. [What it is](#a-what-it-is)
2. [History and milestones](#a-history-and-milestones)
3. [Main problems and tasks](#a-main-problems-and-tasks)
4. [Methods and schools](#a-methods-and-schools)
5. [Datasets, resources and benchmarks](#a-datasets-resources-and-benchmarks)
6. [Landmark papers in this library](#a-landmark-papers-in-this-library)
7. [People](#a-people)
8. [Institutions and places](#a-institutions-and-places)
9. [Open questions](#a-open-questions)
10. [Neighbouring topics](#a-neighbouring-topics)
11. [Key works in this library](#a-key-works-in-this-library)

Hebrew and Aramaic are two closely related Northwest Semitic languages whose histories have been intertwined for nearly three thousand years. Hebrew is the language of most of the Hebrew Bible, of the Mishnah and of a vast rabbinic, liturgical and poetic literature, and since its revival at the turn of the twentieth century it has been the everyday language of millions of speakers in Israel. Aramaic began as the speech of small states in Syria, became the administrative language of the Assyrian, Babylonian and Persian empires, and survived as the vehicle of large parts of Daniel and Ezra, of the Babylonian and Palestinian Talmuds, of the Targums, and, in its Syriac form, of a major Christian literature. Varieties known as Neo-Aramaic are still spoken today, though most are endangered.

This topic gathers research on both languages in all their periods: the grammar, phonology and textual history of Biblical Hebrew; the Hebrew of Qumran and the rabbis; the reading traditions that preserved the Bible's pronunciation; the syntax and semantics of Modern Hebrew; the dialects of Neo-Aramaic and the morphology of Syriac; and the computational processing of all of these, from finite-state analysers and treebanks to Hebrew language models and handwriting analysis of the Dead Sea Scrolls.

What unites them is a common set of problems. Both languages are written in consonantal scripts in which most vowels are left out, both build words from roots and patterns, and both attach prepositions, conjunctions and articles directly to the following word. These properties make a single written token stand for many possible readings, and much of the work collected here, ancient or modern, is about recovering the reading that a fluent reader would choose. For the wider family, including Arabic, Akkadian and Ethiopic, see the sibling topic [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/).

## What it is

### The languages and their stages

Hebrew is conventionally divided into several stages. **Biblical (Classical) Hebrew** is the language of the Hebrew Bible and of Iron Age inscriptions; scholars often distinguish an archaic poetic layer, Standard or Early Biblical Hebrew, and Late Biblical Hebrew found in books such as Chronicles, Ezra, Nehemiah and Daniel. **Qumran Hebrew**, known from the Dead Sea Scrolls discovered from 1947 onward, stands between the biblical and rabbinic languages. **Mishnaic or Rabbinic Hebrew**, the language of the Mishnah (compiled around 200 CE) and related works, has a markedly different verbal system and vocabulary. Medieval Hebrew served for poetry, philosophy, science and correspondence, and **Modern (Israeli) Hebrew**, revived as a spoken language in the late nineteenth and early twentieth centuries, combines biblical and rabbinic elements with new syntax and vocabulary shaped partly by contact with European languages.

Aramaic has a comparable sequence: Old Aramaic inscriptions from the early first millennium BCE; Imperial or Official Aramaic of the Persian period, which includes Biblical Aramaic; the Middle Aramaic of Qumran and the Targums; Late Aramaic dialects such as Jewish Babylonian Aramaic of the Talmud, Jewish Palestinian Aramaic, Samaritan Aramaic, Mandaic and **Syriac**, the literary language of Edessa and of Eastern Christianity; and the **Neo-Aramaic** dialects spoken until recently in northern Iraq, south-eastern Turkey, north-western Iran and Syria, many of them now mainly in diaspora communities.

### Core notions

A few notions recur throughout this topic:

- **Root and pattern.** Most content words are formed by interleaving a consonantal root (often three consonants) with a vowel pattern and affixes. The Hebrew verbal templates, the *binyanim* (qal, nifal, piel, pual, hifil, hofal, hitpael), and their Aramaic and Syriac counterparts, carry voice, causation and reflexivity.
- **Unvocalized script.** Everyday Hebrew is written without the vowel points (*niqqud*) devised by medieval scribes, so a written form like a three-letter string may correspond to several distinct words.
- **Proclitics and clitics.** Prepositions, the article, the conjunction "and" and the relativizer are written as part of the next word; pronominal suffixes attach to nouns, prepositions and verbs. Word segmentation is therefore a genuine linguistic task, not just whitespace splitting.
- **The construct state.** A noun can form a tight genitive chain with a following noun (Hebrew *smixut*), which matters for syntax, definiteness and parsing alike.
- **The Masoretic tradition.** The standard text of the Hebrew Bible carries vowel signs and cantillation accents added by the Masoretes of Tiberias in the second half of the first millennium CE; these marks encode pronunciation and a prosodic phrasing of every verse.

### How it differs from its neighbours

The topic overlaps heavily with [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/), which covers Arabic dialects, Akkadian, Ge'ez and comparative Semitic linguistics; this page keeps to Hebrew and Aramaic. It also borders on [Morphology](https://md.research.elamur.ai/free/en/topics/morphology/), because so much of the computational work is about segmentation and morphological disambiguation, and on [Historical linguistics & language change](https://md.research.elamur.ai/free/en/topics/language-change/), because the dating of biblical texts and the history of reading traditions are central debates. Classical philology of other ancient languages has its own pages on [Ancient Greek](https://md.research.elamur.ai/free/en/topics/ancient-greek/), [Latin](https://md.research.elamur.ai/free/en/topics/latin/) and [Akkadian and cuneiform NLP](https://md.research.elamur.ai/free/en/topics/akkadian-cuneiform-nlp/).

## History and milestones

### Philology before computers

Grammatical study of Hebrew began with medieval Jewish scholars. Modern comparative grammar placed Hebrew and Aramaic within Semitic and debated the meaning of the verbal forms, above all the "consecutive" forms *wayyiqtol* and *weqatal*. The Dead Sea Scrolls, the Cairo Genizah and fieldwork on Neo-Aramaic later added new evidence.

A distinctive thread in this library is the study of how the Bible was read aloud. [The Tiberian Pronunciation Tradition](https://md.research.elamur.ai/free/en/papers/khan-2020-tiberian-pronunciation-tradition-biblical/) by [Geoffrey Khan](https://md.research.elamur.ai/free/en/authors/geoffrey-khan/) reconstructs the pronunciation that the Tiberian vowel and accent signs were designed to record, a reading that fell out of use within a few centuries and was replaced in European grammars by Sephardi pronunciation. Related collections, such as [Studies in Semitic Vocalisation](https://md.research.elamur.ai/free/en/papers/hornkohl-2020-studies-semitic-vocalisation-reading/) and [New Perspectives in Biblical and Rabbinic Hebrew](https://md.research.elamur.ai/free/en/papers/hornkohl-2021-new-perspectives-biblical-rabbinic/), extend this work to Greek and Latin transcriptions, Babylonian vocalisation and the Hebrew verb.

### The first computational work

Computational treatment of Hebrew began in the 1970s and 1980s with concordances and morphological databases of the Bible, notably the long-running Amsterdam database of the Hebrew Bible associated with Eep Talstra. For Modern Hebrew, work at the Technion and elsewhere produced morphological analysers for unpointed text. A landmark of the statistical turn is [Learning Morpho-Lexical Probabilities from an Untagged Corpus](https://md.research.elamur.ai/free/en/papers/levinger-1995-learning-morpho-lexical-probabilities-untagged/) by Moshe Levinger, [Uzzi Ornan](https://md.research.elamur.ai/free/en/authors/uzzi-ornan/) and [Alon Itai](https://md.research.elamur.ai/free/en/authors/alon-itai/), which showed that the unambiguous relatives of an ambiguous word could be counted in raw text to estimate which analysis is most likely.

### The workshop era (2000–2010)

The early 2000s brought a burst of activity. The ACL workshops on Computational Approaches to Semitic Languages, starting in 2002, gave Hebrew, Aramaic and Arabic researchers a shared venue. The Hebrew Treebank, built from newspaper text by Khalil Sima'an and colleagues around 2001, made statistical parsing possible. Finite-state grammars, HMM taggers and diacritizers, root learners and a first Hebrew–English transfer system date from these years. The Knowledge Center for Processing Hebrew (MILA) released lexicons and analysers that many later systems used.

### Morphologically rich parsing (2010–2019)

From around 2010 the focus shifted to parsing. Hebrew became a standard test case for the workshops and shared tasks on Statistical Parsing of Morphologically Rich Languages (SPMRL), where it showed that segmentation and syntax must be resolved together. Lattice parsing, joint transition-based models and new dependency treebanks followed, and Hebrew entered [Universal Dependencies](https://md.research.elamur.ai/free/en/topics/universal-dependencies/). In parallel, digital humanities projects turned the Hebrew Bible into openly shared, richly annotated data, and stylometry began to address questions of biblical authorship.

### Pre-trained models and large language models (2019–today)

With BERT came a series of Hebrew encoders (HeBERT, AlephBERT, AlephBERTGimmel, HeRo, DictaBERT), Hebrew-specific benchmarks for question answering, summarization, sentiment and named entities, and, more recently, Hebrew-adapted generative models such as DictaLM. Historical Hebrew gained models of its own, such as BEREL for rabbinic texts.

Landmark papers on this topic: the 12 most cited, in time The 12 most cited papers on this topic span 2000–2021; the dot grows with the citations.

**The numbers**

| Paper | Citations |
| --- | --- |
| [2000 — Multitiered Nonlinear Morphology Using Multitape Finite Automata: A Case Study on Syriac and Arabic](https://md.research.elamur.ai/free/en/papers/kiraz-2000-multitiered-nonlinear-morphology-multitape-finite/) | 96 |
| [2002 — An HMM Approach to Vowel Restoration in Arabic and Hebrew](https://md.research.elamur.ai/free/en/papers/gal-2002-hmm-approach-vowel-restoration-arabic/) | 106 |
| [2006 — An Unsupervised Morpheme-Based HMM for Hebrew Morphological Disambiguation](https://md.research.elamur.ai/free/en/papers/adler-2006-unsupervised-morpheme-based-hmm-hebrew/) | 99 |
| [2006 — Integrated Morphological and Syntactic Disambiguation for Modern Hebrew](https://md.research.elamur.ai/free/en/papers/tsarfaty-2006-integrated-morphological-syntactic-disambiguation-modern/) | 67 |
| [2008 — Late Biblical Hebrew and the Qumran Pesher Habakkuk](https://md.research.elamur.ai/free/en/papers/young-2008-late-biblical-hebrew-qumran-pesher/) | 67 |
| [2014 — Typology of case alignments in NENA dialects](https://md.research.elamur.ai/free/en/papers/barotto-2014-typology-case-alignments-nena-dialects/) | 66 |
| [2015 — A Graph-based Lattice Dependency Parser for Joint Morphological Segmentation and Syntactic Analysis](https://md.research.elamur.ai/free/en/papers/seeker-2015-graph-based-lattice-dependency-parser/) | 66 |
| [2018 — Colorless Green Recurrent Networks Dream Hierarchically](https://md.research.elamur.ai/free/en/papers/gulordava-2018-colorless-green-recurrent-networks-dream/) | 760 |
| [2018 — Missing objects in Hebrew: Argument ellipsis, not VP ellipsis](https://md.research.elamur.ai/free/en/papers/landau-2018-missing-objects-hebrew-argument-ellipsis/) | 89 |
| [2020 — Cross-Linguistic Syntactic Evaluation of Word Prediction Models](https://md.research.elamur.ai/free/en/papers/mueller-2020-cross-linguistic-syntactic-evaluation-word/) | 97 |
| [2020 — The Tiberian Pronunciation Tradition of Biblical Hebrew](https://md.research.elamur.ai/free/en/papers/khan-2020-tiberian-pronunciation-tradition-biblical/) | 75 |
| [2021 — HeBERT & HebEMO: a Hebrew BERT Model and a Tool for Polarity Analysis and Emotion Recognition](https://md.research.elamur.ai/free/en/papers/chriqui-2021-hebert-hebemo-hebrew-bert-model/) | 68 |

Papers on this topic by year 480 dated papers from 1969 to 2026; the busiest year is 2024 with 46, and 61% of them appeared in 2018 or later.

**The numbers**

| Year | Papers |
| --- | --- |
| 2026 | 35 |
| 2025 | 38 |
| 2024 | 46 |
| 2023 | 26 |
| 2022 | 33 |
| 2021 | 31 |
| 2020 | 33 |
| 2019 | 27 |
| 2018 | 22 |
| 2017 | 17 |
| 2016 | 29 |
| 2015 | 9 |
| 2014 | 15 |
| 2013 | 6 |
| 2012 | 18 |
| 2011 | 7 |
| 2010 | 9 |
| 2009 | 8 |
| 2008 | 10 |
| 2007 | 5 |
| 2006 | 9 |
| 2005 | 2 |
| 2004 | 8 |
| 2003 | 2 |
| 2002 | 5 |
| 2001 | 2 |
| 2000 | 6 |
| 1999 | 1 |
| 1998 | 4 |
| 1996 | 1 |
| 1995 | 2 |
| 1994 | 3 |
| 1993 | 1 |
| 1991 | 3 |
| 1990 | 2 |
| 1988 | 1 |
| 1982 | 1 |
| 1974 | 1 |
| 1969 | 2 |

## Main problems and tasks

### Reading and writing the script

- **Vowel restoration (diacritization).** Adding niqqud to unpointed text is needed for text-to-speech, language teaching, children's books and poetry. It is hard because the vowels depend on morphology and context. Work ranges from hidden Markov models to dictionary-backed neural systems such as Nakdan and dictionary-free models.
- **Segmentation.** Splitting a written token into its proclitics, stem and suffixes; this is the first step of almost every pipeline and the source of many downstream errors.
- **Morphological analysis and disambiguation.** Listing every possible analysis of a token and choosing the right one in context. Each Hebrew token typically has several candidate analyses, far more than in English.
- **Root extraction and lexicon building.** Recovering the consonantal root from inflected or weak forms, and assembling computational lexicons that cover the inflectional paradigms.

### Syntax and semantics

- **Parsing.** Constituency and dependency parsing of Modern Hebrew, with special difficulty in coordination, prepositional attachment, construct chains and free word order; treebanking for ancient Hebrew and Syriac.
- **Theoretical syntax and semantics.** Ellipsis, reflexive and middle verbs, the structure of the noun phrase, differential object marking with *et*, reciprocals, and the acquisition of movement structures by Hebrew-speaking children.
- **Biblical Hebrew grammar.** The meaning of the verbal forms (*qatal*, *yiqtol*, *wayyiqtol*, *weqatal*, the participle), word order and information structure, left dislocation, verbless clauses, negation, and the semantics of individual words and stems.

### Texts and their history

- **Dating and diachrony.** Whether the language of a biblical book can show when it was written, and how Late Biblical Hebrew, Qumran Hebrew and Mishnaic Hebrew relate.
- **Authorship and composition.** Separating authorial layers in composite texts, identifying scribes, and dating manuscripts by their handwriting.
- **Textual reuse and reconstruction.** Finding parallel passages in rabbinic literature, filling gaps in damaged manuscripts, and correcting OCR errors in historical prints.
- **Neo-Aramaic documentation.** Recording and describing endangered spoken dialects, and analysing how contact with Kurdish, Persian, Turkish and Arabic shaped them.

### Applications for Modern Hebrew

Translation, sentiment, offensive-language detection, summarization, question answering, speech, and legal and medical applications.

## Methods and schools

### Philological and linguistic approaches

The philological tradition reads texts closely, compares manuscripts and versions, and uses Semitic comparison. Since the late twentieth century, Biblical Hebrew study has absorbed approaches from general linguistics. Typological and grammaticalization-based work, exemplified by [Alexander Andrason](https://md.research.elamur.ai/free/en/authors/alexander-andrason/), treats verb forms as points on cross-linguistic paths of change. Cognitive and functional approaches, as in the work of [Josh Westbury](https://md.research.elamur.ai/free/en/authors/josh-westbury/) on left dislocation, model constructions as networks of schemas tied to discourse function. Generative syntax, represented by [Jacobus A. Naudé](https://md.research.elamur.ai/free/en/authors/jacobus-a-naude/) and [Cynthia L. Miller-Naudé](https://md.research.elamur.ai/free/en/authors/cynthia-l-miller-naude/), brings tools such as clitic theory and prosodic phrasing to biblical clauses, often using the Masoretic accents as evidence of prosody.

The most persistent debate concerns **linguistic dating**. The chronological model holds that Early Biblical Hebrew gave way to Late Biblical Hebrew around the Babylonian exile, so linguistic features can date texts. Scholars such as [Ian Young](https://md.research.elamur.ai/free/en/authors/ian-young/) and [Robert Rezetko](https://md.research.elamur.ai/free/en/authors/robert-rezetko/) have argued that early and late features represent coexisting styles and that the fluidity of manuscript transmission makes such dating unreliable; their opponents defend the diachronic signal. A related discussion asks how old the Tiberian vocalisation really is, with [Aaron D. Hornkohl](https://md.research.elamur.ai/free/en/authors/aaron-d-hornkohl/) and [Benjamin D. Suchard](https://md.research.elamur.ai/free/en/authors/benjamin-suchard/) using Qumran Hebrew and Biblical Aramaic to separate changes in the living language from changes in the reading tradition.

### Rule-based and finite-state systems

Root-and-pattern morphology resists the simple concatenation assumed by classic two-level morphology. Finite-state methods adapted to it by using several tapes or tiers (root, pattern, vocalism) or by compiling large lexicons with replace rules. Finite-state analysers remain important for ancient Hebrew, where training data are scarce.

### Statistical models

From the mid-1990s, probabilistic models took over disambiguation: corpus-estimated priors, hidden Markov models over words or morphemes, classifier combinations with hard linguistic constraints, and probabilistic context-free grammars. Morpheme-level models generally beat word-level ones, provided segmentation errors are controlled.

### Joint and lattice-based models

The defining methodological idea in Hebrew parsing is the **lattice**: represent every possible segmentation and analysis of a sentence as a graph, and let the parser choose a path through it while building syntactic structure. This avoids the error propagation of a segment-then-parse pipeline. Lattice parsing was developed for constituency grammars by [Yoav Goldberg](https://md.research.elamur.ai/free/en/authors/yoav-goldberg/) and [Reut Tsarfaty](https://md.research.elamur.ai/free/en/authors/reut-tsarfaty/), and later extended to graph-based and transition-based dependency parsing.

### Neural models and pre-training

A particular Hebrew question has been the **tokenizer**: subword units learned from frequency do not coincide with Hebrew morphemes, so researchers have added segmentation heads to BERT models, experimented with very large vocabularies, and tested whether explicit morphological knowledge improves pre-training.

### Image and stylometric methods

For ancient manuscripts, pattern recognition on handwriting and stylometry on word choice offer evidence that is independent of philological judgement.

## Datasets, resources and benchmarks

- **The Masoretic Text and its digital editions.** The Leningrad Codex (early eleventh century) underlies most digital Bibles; morphologically tagged editions, the Amsterdam (ETCBC) database and Text-Fabric tooling, and syntactic annotations are the basis for quantitative Biblical Hebrew research. See [The Hebrew Bible as Data](https://md.research.elamur.ai/free/en/papers/roorda-2015-hebrew-bible-data-laboratory-sharing/) and [LAF-Fabric](https://md.research.elamur.ai/free/en/papers/roorda-2014-laf-fabric-data-analysis-tool/).
- **The Hebrew Treebank (HTB)** of newspaper sentences, with its constituency, SPMRL dependency and Universal Dependencies versions; described in [The Hebrew Universal Dependency Treebank](https://md.research.elamur.ai/free/en/papers/sade-2018-hebrew-universal-dependency-treebank-past/).
- **IAHLT WIKI**, a second UD treebank of Hebrew Wikipedia text: [A Second Wave of UD Hebrew Treebanking](https://md.research.elamur.ai/free/en/papers/zeldes-2022-second-wave-ud-hebrew-treebanking/).
- **Ancient Hebrew UD treebanks**: [A Universal Dependencies Treebank of Ancient Hebrew](https://md.research.elamur.ai/free/en/papers/swanson-2022-universal-dependencies-treebank-ancient-hebrew/) and its parallel Hebrew–Greek extension.
- **NEMO**, token- and morpheme-level named-entity annotation of the HTB, and the combined UD-NEMO benchmark.
- **Question answering**: [HeQ](https://md.research.elamur.ai/free/en/papers/cohen-2023-heq-large-diverse-hebrew-reading/) and [ParaShoot: A Hebrew Question Answering Dataset](https://md.research.elamur.ai/free/en/papers/keren-2021-parashoot-hebrew-question-answering-dataset/).
- **Summarization**: [HeSum](https://md.research.elamur.ai/free/en/papers/argaman-2024-hesum-novel-dataset-abstractive-text/).
- **Speech**: [ivrit.ai](https://md.research.elamur.ai/free/en/papers/marmor-2023-ivrit-ai-comprehensive-dataset-hebrew/) and [HebDB](https://md.research.elamur.ai/free/en/papers/turetzky-2024-hebdb-weakly-supervised-dataset-hebrew/).
- **Parliamentary and social text**: [The Knesset Corpus](https://md.research.elamur.ai/free/en/papers/goldin-2024-knesset-corpus-annotated-corpus-hebrew/).
- **Lexical resources**: [A Computational Lexicon of Contemporary Hebrew](https://md.research.elamur.ai/free/en/papers/itai-2006-computational-lexicon-contemporary-hebrew/), [The Hebrew FrameNet Project](https://md.research.elamur.ai/free/en/papers/hayoun-2016-hebrew-framenet-project/) and [a lexicon of Hebrew verb-noun expressions](https://md.research.elamur.ai/free/en/papers/liebeskind-2016-lexical-resource-hebrew-verb-noun/).
- **Diacritization evaluation**: [a challenge set for Hebrew disambiguation and diacritics](https://md.research.elamur.ai/free/en/papers/shmidman-2020-novel-challenge-set-hebrew-morphological/).
- **Parallel corpora** for Hebrew–Arabic translation, such as [An Arabic-Hebrew parallel corpus of TED talks](https://md.research.elamur.ai/free/en/papers/cettolo-2016-arabic-hebrew-parallel-corpus-ted/), and the multilingual Bible corpus [HELFI](https://md.research.elamur.ai/free/en/papers/jyra-2020-helfi-hebrew-greek-finnish-parallel/).
- **Syriac**: handwritten OCR data for East Syriac script, described in [Ancient but Digitized](https://md.research.elamur.ai/free/en/papers/majeed-2024-ancient-but-digitized-developing-handwritten/).

## Landmark papers in this library

Citations of the papers on this topic, by when they were published Papers published in 2015–2019 gather 37% of the 4,514 citations of the papers on this topic.

**The numbers**

| Published | Papers and their citations |
| --- | --- |
| 2025–2029 | 73 papers, 27 citations |
| 2020–2024 | 169 papers, 1,157 citations |
| 2015–2019 | 104 papers, 1,651 citations |
| 2010–2014 | 55 papers, 525 citations |
| 2005–2009 | 34 papers, 651 citations |
| 2000–2004 | 23 papers, 356 citations |
| 1995–1999 | 8 papers, 80 citations |
| 1990–1994 | 9 papers, 65 citations |
| 1985–1989 | 1 papers, 2 citations |
| 1980–1984 | 1 papers, 0 citations |
| 1970–1974 | 1 papers, 0 citations |
| 1965–1969 | 2 papers, 0 citations |

### Vowel restoration with a hidden Markov model (2002)

[An HMM Approach to Vowel Restoration](https://md.research.elamur.ai/free/en/papers/gal-2002-hmm-approach-vowel-restoration-arabic/), by Ya'akov Gal, treated each fully pointed word as a hidden state that emits its bare consonantal spelling, and decoded the most likely sequence with Viterbi search. Trained on the Hebrew Bible and on the Qur'an, it beat a most-frequent-reading baseline by a clear margin, and it identified words never seen in training as the main obstacle.

### Multitape finite-state morphology for Syriac (2000)

[Multitiered Nonlinear Morphology Using Multitape Finite Automata](https://md.research.elamur.ai/free/en/papers/kiraz-2000-multitiered-nonlinear-morphology-multitape-finite/), by [George Anton Kiraz](https://md.research.elamur.ai/free/en/authors/george-anton-kiraz/), separated root, pattern and vocalism into distinct lexical tiers and compiled them, together with rewrite rules, into machines that read several tapes at once. Its worked examples are Syriac verb stems and Arabic nominals.

### Unsupervised morpheme-based disambiguation (2006)

[An Unsupervised Morpheme-Based HMM](https://md.research.elamur.ai/free/en/papers/adler-2006-unsupervised-morpheme-based-hmm-hebrew/), by [Meni Adler](https://md.research.elamur.ai/free/en/authors/meni-adler/) and [Michael Elhadad](https://md.research.elamur.ai/free/en/authors/michael-elhadad/), learned segmentation and tagging together from untagged newspaper text, using only an analyser to propose candidates. Because morphemes are hidden inside words, the model ran Baum-Welch and Viterbi over a lattice of morpheme slots. The morpheme model outperformed a word-level one, and the error analysis singled out participles and unknown words, problems the field returned to for years.

### Joint morphology and syntax in parsing (2006)

[Integrated Morphological and Syntactic Disambiguation](https://md.research.elamur.ai/free/en/papers/tsarfaty-2006-integrated-morphological-syntactic-disambiguation-modern/), by [Reut Tsarfaty](https://md.research.elamur.ai/free/en/authors/reut-tsarfaty/), produced the first statistical Modern Hebrew parser that did not rely on hand-disambiguated morphology. It connected an analyser, a tagger and a PCFG parser and asked how much morphological information to pass between them, setting the agenda for the joint models that followed.

### Lattice parsing of raw Hebrew (2011, 2013)

The 2011 [lattice parser for raw Hebrew](https://md.research.elamur.ai/free/en/papers/goldberg-2011-joint-hebrew-segmentation-parsing-pcfgla/) by Goldberg and Elhadad let a latent-annotation PCFG read a lattice built from a lexicon-based analyser, so that unsegmented text could be parsed directly. It clearly beat both a pipeline and the earlier best result on raw text. The journal version, [Word Segmentation, Unknown-word Resolution, and Morphological Agreement](https://md.research.elamur.ai/free/en/papers/goldberg-2013-word-segmentation-unknown-word-resolution/), added an external lexicon for unseen words and a separate agreement filter, and argued that the design carries over to other morphologically rich languages.

### Lattice dependency parsing (2015) and joint transition systems (2019)

[A Graph-based Lattice Dependency Parser](https://md.research.elamur.ai/free/en/papers/seeker-2015-graph-based-lattice-dependency-parser/), by [Wolfgang Seeker](https://md.research.elamur.ai/free/en/authors/wolfgang-seeker/) and [Özlem Çetinoğlu](https://md.research.elamur.ai/free/en/authors/ozlem-cetinoglu/), chose a lattice path and a dependency tree together using dual decomposition, on Turkish and Hebrew. [Joint Transition-Based Models for Morpho-Syntactic Parsing](https://md.research.elamur.ai/free/en/papers/more-2019-joint-transition-based-models-morpho/), by [Amir More](https://md.research.elamur.ai/free/en/authors/amir-more/), [Amit Seker](https://md.research.elamur.ai/free/en/authors/amit-seker/), Victoria Basmova and Tsarfaty, merged a morphological transition system with arc-based dependency systems under one beam search. It showed that syntax improves morphological choices in realistic conditions and set a new end-to-end standard for Hebrew; the system was packaged as YAP, presented in [What's Wrong with Hebrew NLP?](https://md.research.elamur.ai/free/en/papers/tsarfaty-2019-what-s-wrong-hebrew-nlp/).

### Hebrew BERT models (2021–2023)

[HeBERT & HebEMO](https://md.research.elamur.ai/free/en/papers/chriqui-2021-hebert-hebemo-hebrew-bert-model/), by Avihay Chriqui and Inbal Yahav, released one of the first Hebrew BERT models together with an emotion-recognition tool trained on reader comments. [AlephBERT](https://md.research.elamur.ai/free/en/papers/seker-2022-alephbert-language-model-pre-training/) (with an earlier [preprint](https://md.research.elamur.ai/free/en/papers/seker-2021-alephbert-hebrew-large-pre-trained/)), from Tsarfaty's group, trained on a larger corpus and, crucially, evaluated the model on every level of the Hebrew pipeline, adding a character-level head that turns wordpiece vectors into morphemes. [Large Pre-Trained Models with Extra-Large Vocabularies](https://md.research.elamur.ai/free/en/papers/gueta-2022-large-pre-trained-models-extra/) introduced AlephBERTGimmel and showed that keeping more words whole in the vocabulary helps, while [DictaBERT](https://md.research.elamur.ai/free/en/papers/shmidman-2023-dictabert-state-art-bert-suite/) from the Dicta team combined careful data handling with released heads for segmentation, morphology and question answering.

### Morpheme-aware named entities (2021)

[NEMO2](https://md.research.elamur.ai/free/en/papers/bareket-2021-neural-modeling-named-entities-morphology/), by [Dan Bareket](https://md.research.elamur.ai/free/en/authors/dan-bareket/) and Tsarfaty, annotated the Hebrew Treebank with entities at both the token and the morpheme level. Labelling morphemes worked better, but only with good segmentation; their hybrid design uses token-level entity predictions to prune the lattice first. The paper became the standard Hebrew NER benchmark.

### Argument ellipsis in Modern Hebrew (2018)

[Missing objects in Hebrew](https://md.research.elamur.ai/free/en/papers/landau-2018-missing-objects-hebrew-argument-ellipsis/), by Idan Landau, reconsidered the evidence that Hebrew null objects arise from ellipsis of a verb phrase whose verb has moved out. Through examples with non-verbal antecedents, island contexts and mismatched verbs, it argued that only the object is elided.

### Reflexive verbs and the root (2017)

[Reflexive verbs in Hebrew](https://md.research.elamur.ai/free/en/papers/kastner-2017-reflexive-verbs-hebrew-deep-unaccusativity/), by [Itamar Kastner](https://md.research.elamur.ai/free/en/authors/itamar-kastner/), analysed the *hitpael* template, which hosts reflexive, reciprocal and anticausative verbs, without any dedicated reflexive morphology. The reading, it argued, depends on the semantics of the root rather than on the template. Kastner developed the account further in [Voice at the Interfaces](https://md.research.elamur.ai/free/en/papers/kastner-2020-voice-at-interfaces-syntax/).

### Linguistic dating and Qumran (2008)

[Late Biblical Hebrew and Pesher Habakkuk](https://md.research.elamur.ai/free/en/papers/young-2008-late-biblical-hebrew-qumran-pesher/), by Ian Young, measured how many distinct late features cluster in a fixed-size sample of text. The Qumran commentary on Habakkuk, written centuries after the exile, turned out to look no "later" than classical biblical prose. Young concluded that early and late Biblical Hebrew were coexisting styles, a claim that has shaped the dating debate since.

### Recovering authorial layers (2011)

[Unsupervised Decomposition of a Document](https://md.research.elamur.ai/free/en/papers/koppel-2011-unsupervised-decomposition-document-into-authorial/), by [Moshe Koppel](https://md.research.elamur.ai/free/en/authors/moshe-koppel/), Navot Akiva, Idan Dershowitz and [Nachum Dershowitz](https://md.research.elamur.ai/free/en/authors/nachum-dershowitz/), artificially merged pairs of biblical books and tried to separate them without labels. Clustering by choice among synonyms worked far better than ordinary word counts, and a second stage trained on the most confident chapters nearly solved the task. Applied to the Pentateuch, the method's split agreed largely with the scholarly distinction between Priestly and non-Priestly material.

### Two scribes of the Great Isaiah Scroll (2021)

[AI-based writer identification for the Dead Sea Scrolls](https://md.research.elamur.ai/free/en/papers/popovic-2020-artificial-intelligence-based-writer-identification/), by [Mladen Popovic](https://md.research.elamur.ai/free/en/authors/mladen-popovic/), [Maruf A. Dhali](https://md.research.elamur.ai/free/en/authors/maruf-a-dhali/) and [Lambert Schomaker](https://md.research.elamur.ai/free/en/authors/lambert-schomaker/), extracted handwriting features from every column of the scroll and found a sharp transition near its middle, consistent with two scribes writing in a closely matched hand. The same team later combined new radiocarbon dates with handwriting analysis in [Dating ancient manuscripts using radiocarbon and writing style](https://md.research.elamur.ai/free/en/papers/popovic-2024-dating-ancient-manuscripts-using-radiocarbon/).

### Decipherment through Hebrew (2019)

[Neural Decipherment via Minimum-Cost Flow](https://md.research.elamur.ai/free/en/papers/luo-2019-neural-decipherment-minimum-cost-flow/), by [Jiaming Luo](https://md.research.elamur.ai/free/en/authors/jiaming-luo/), [Yuan Cao](https://md.research.elamur.ai/free/en/authors/yuan-cao/) and [Regina Barzilay](https://md.research.elamur.ai/free/en/authors/regina-barzilay/), used Hebrew as the known relative against which Ugaritic vocabulary was aligned, finding cognates by a character-level network and a flow-based matching. The method also worked for Linear B against Greek.

## People

Authors with the most papers on this topic Reut Tsarfaty leads with 26 papers; the 12 authors shown wrote 158 of the papers on this topic between them, counting each paper once per author.

**The numbers**

| Author | Papers here · their citations |
| --- | --- |
| [Reut Tsarfaty](https://md.research.elamur.ai/free/en/authors/reut-tsarfaty/) | 26 papers · 562 citations |
| [Avi Shmidman](https://md.research.elamur.ai/free/en/authors/avi-shmidman/) | 22 papers · 144 citations |
| [Michael Elhadad](https://md.research.elamur.ai/free/en/authors/michael-elhadad/) | 14 papers · 343 citations |
| [Moshe Koppel](https://md.research.elamur.ai/free/en/authors/moshe-koppel/) | 14 papers · 186 citations |
| [Shuly Wintner](https://md.research.elamur.ai/free/en/authors/shuly-wintner/) | 13 papers · 216 citations |
| [Shaltiel Shmidman](https://md.research.elamur.ai/free/en/authors/shaltiel-shmidman/) | 13 papers · 119 citations |
| [Yoav Goldberg](https://md.research.elamur.ai/free/en/authors/yoav-goldberg/) | 11 papers · 259 citations |
| [Alexander Andrason](https://md.research.elamur.ai/free/en/authors/alexander-andrason/) | 11 papers · 89 citations |
| [Nachum Dershowitz](https://md.research.elamur.ai/free/en/authors/nachum-dershowitz/) | 9 papers · 58 citations |
| [Chaya Liebeskind](https://md.research.elamur.ai/free/en/authors/chaya-liebeskind/) | 9 papers · 21 citations |
| [Amit Seker](https://md.research.elamur.ai/free/en/authors/amit-seker/) | 8 papers · 312 citations |
| [Martijn Naaijer](https://md.research.elamur.ai/free/en/authors/martijn-naaijer/) | 8 papers · 23 citations |

- **[Reut Tsarfaty](https://md.research.elamur.ai/free/en/authors/reut-tsarfaty/)** (Bar-Ilan University; also Google) is the most prolific author on this topic. Her work runs from the first joint morpho-syntactic parser for Modern Hebrew, through agreement modelling in [Modeling Morphosyntactic Agreement](https://md.research.elamur.ai/free/en/papers/tsarfaty-2010-modeling-morphosyntactic-agreement-constituency-based/) with [Khalil Sima'an](https://md.research.elamur.ai/free/en/authors/khalil-sima-an/), to YAP, NEMO and AlephBERT.
- **[Avi Shmidman](https://md.research.elamur.ai/free/en/authors/avi-shmidman/)** (Bar-Ilan University) leads, with **[Shaltiel Shmidman](https://md.research.elamur.ai/free/en/authors/shaltiel-shmidman/)**, much of the Dicta work: the Nakdan diacritizer, DictaBERT, the DictaLM generative models, and tools for rabbinic and historical Hebrew such as [Identification of Parallel Passages](https://md.research.elamur.ai/free/en/papers/shmidman-2016-identification-parallel-passages-across-large/).
- **[Michael Elhadad](https://md.research.elamur.ai/free/en/authors/michael-elhadad/)** (Ben-Gurion University of the Negev) co-authored the unsupervised morpheme-based HMM, the lattice parsers with Goldberg, Hebrew noun-phrase chunking, and multilingual summarization corpora that include Hebrew.
- **[Moshe Koppel](https://md.research.elamur.ai/free/en/authors/moshe-koppel/)** (Bar-Ilan University), known for authorship attribution, contributed the unsupervised decomposition of biblical texts and is a co-author of the Dicta models.
- **[Shuly Wintner](https://md.research.elamur.ai/free/en/authors/shuly-wintner/)** (University of Haifa) worked on finite-state Hebrew morphology, classifier-based disambiguation, root identification with [Ezra Daya](https://md.research.elamur.ai/free/en/authors/ezra-daya/) and [Dan Roth](https://md.research.elamur.ai/free/en/authors/dan-roth/), and the transfer-based Hebrew–English system with [Alon Lavie](https://md.research.elamur.ai/free/en/authors/alon-lavie/).
- **[Yoav Goldberg](https://md.research.elamur.ai/free/en/authors/yoav-goldberg/)** (Bar-Ilan University and the Allen Institute for AI) built Hebrew dependency and constituency parsers in his early career, including easy-first and lattice parsing, before becoming a leading figure in neural NLP generally.
- **[Nachum Dershowitz](https://md.research.elamur.ai/free/en/authors/nachum-dershowitz/)** (Tel Aviv University) has applied computation to biblical and Judeo-Arabic texts: authorial decomposition, transliteration of Judeo-Arabic into Arabic script, and segmentation of ancient and historic Hebrew.
- **[Chaya Liebeskind](https://md.research.elamur.ai/free/en/authors/chaya-liebeskind/)** (Jerusalem College of Technology) works on Hebrew lexical resources, thesaurus construction, multiword expressions and, with [Marina Litvak](https://md.research.elamur.ai/free/en/authors/marina-litvak/) and [Natalia Vanetik](https://md.research.elamur.ai/free/en/authors/natalia-vanetik/) of Sami Shamoon College of Engineering, offensive-language detection.
- **[Alexander Andrason](https://md.research.elamur.ai/free/en/authors/alexander-andrason/)** (University of Cape Town) writes on the Biblical Hebrew verb, from [Biblical Hebrew Wayyiqtol: A Dynamic Definition](https://md.research.elamur.ai/free/en/papers/andrason-2011-biblical-hebrew-wayyiqtol-dynamic-definition/) to interjections and serial verbs, using grammaticalization theory.
- **[Jacobus A. Naudé](https://md.research.elamur.ai/free/en/authors/jacobus-a-naude/)** and **[Cynthia L. Miller-Naudé](https://md.research.elamur.ai/free/en/authors/cynthia-l-miller-naude/)** (University of the Free State) study Biblical and Qumran Hebrew syntax: verbless clauses, dislocation, negation, and the place of Qumran in the history of the language.
- **[Geoffrey Khan](https://md.research.elamur.ai/free/en/authors/geoffrey-khan/)** (Cambridge) is the leading authority on the Tiberian reading tradition and on North-Eastern Neo-Aramaic, as shown by [Left Dislocation in NENA dialects](https://md.research.elamur.ai/free/en/papers/khan-2016-left-dislocation-north-eastern-neo/) and [Studies in the Grammar and Lexicon](https://md.research.elamur.ai/free/en/papers/khan-2021-studies-grammar-lexicon-neo/).
- **[Martijn Naaijer](https://md.research.elamur.ai/free/en/authors/martijn-naaijer/)** (University of Zurich) and **[Dirk Roorda](https://md.research.elamur.ai/free/en/authors/dirk-roorda/)** represent the Amsterdam tradition of the Hebrew Bible as data, from LAF-Fabric to quantitative studies of clause structure and a transformer parser for Syriac morphology.
- **[Daniel G. Swanson](https://md.research.elamur.ai/free/en/authors/daniel-g-swanson/)** (Indiana University Bloomington), with [Francis M. Tyers](https://md.research.elamur.ai/free/en/authors/francis-tyers/), built the Universal Dependencies treebanks and finite-state analysers for Ancient Hebrew, and began entity and coreference annotation of the Bible.
- **[Naama Friedmann](https://md.research.elamur.ai/free/en/authors/naama-friedmann/)** (Tel Aviv University) uses Hebrew to study syntactic movement in children, in hearing impairment and in aphasia.
- **[Alon Itai](https://md.research.elamur.ai/free/en/authors/alon-itai/)** (Tel Aviv University and the Hebrew University of Jerusalem) was a pioneer of corpus-based Hebrew morphology and of computational Hebrew lexicons.

## Institutions and places

Israeli universities dominate the computational side. [Bar-Ilan University](https://md.research.elamur.ai/free/en/universities/bar-ilan-university/) is home to the Tsarfaty group, the Dicta team and Goldberg; [Ben-Gurion University of the Negev](https://md.research.elamur.ai/free/en/universities/ben-gurion-university-of-the-negev/) to Elhadad's lab; the [University of Haifa](https://md.research.elamur.ai/free/en/universities/university-of-haifa/) to Wintner; [Tel Aviv University](https://md.research.elamur.ai/free/en/universities/tel-aviv-university/) to Dershowitz and Friedmann; and the [Hebrew University of Jerusalem](https://md.research.elamur.ai/free/en/universities/hebrew-university-of-jerusalem/) and the [Technion – Israel Institute of Technology](https://md.research.elamur.ai/free/en/universities/technion-israel-institute-of-technology/) to long traditions in both linguistics and computer science. Smaller institutions such as the [Jerusalem College of Technology](https://md.research.elamur.ai/free/en/universities/jerusalem-college-of-technology/) and [Sami Shamoon College of Engineering](https://md.research.elamur.ai/free/en/universities/sami-shamoon-college-of-engineering/) contribute applied Hebrew NLP. [Google](https://md.research.elamur.ai/free/en/companies/google/) and the [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) appear through their researchers.

Biblical Hebrew linguistics has a strong centre in South Africa, at the [University of the Free State](https://md.research.elamur.ai/free/en/universities/university-of-the-free-state/), the [University of Cape Town](https://md.research.elamur.ai/free/en/universities/university-of-cape-town/) and Stellenbosch, whose journals and repositories carry many of the papers here. In Europe, the [University of Cambridge](https://md.research.elamur.ai/free/en/universities/university-of-cambridge/) leads work on reading traditions and Neo-Aramaic, the Netherlands on Bible databases and Dead Sea Scrolls handwriting, and the [University of Zurich](https://md.research.elamur.ai/free/en/universities/university-of-zurich/) on quantitative approaches. In the United States, [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/) hosted early Hebrew machine translation, and [Indiana University](https://md.research.elamur.ai/free/en/universities/indiana-university/) and [Indiana University Bloomington](https://md.research.elamur.ai/free/en/universities/indiana-university-bloomington/) host ancient Hebrew treebanking.

Where the authors on this topic work now Bar-Ilan University is the current home of authors of 59 papers on this topic; counted are the authors with a page here, by their latest institution.

**The numbers**

| Institution | Papers |
| --- | --- |
| [Bar-Ilan University](https://md.research.elamur.ai/free/en/universities/bar-ilan-university/) | 59 papers |
| [Google](https://md.research.elamur.ai/free/en/companies/google/) | 28 papers |
| [Hebrew University of Jerusalem](https://md.research.elamur.ai/free/en/universities/hebrew-university-of-jerusalem/) | 28 papers |
| [Tel Aviv University](https://md.research.elamur.ai/free/en/universities/tel-aviv-university/) | 26 papers |
| [Ben-Gurion University of the Negev](https://md.research.elamur.ai/free/en/universities/ben-gurion-university-of-the-negev/) | 21 papers |
| [University of Haifa](https://md.research.elamur.ai/free/en/universities/university-of-haifa/) | 18 papers |
| [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) | 13 papers |
| [Technion – Israel Institute of Technology](https://md.research.elamur.ai/free/en/universities/technion-israel-institute-of-technology/) | 13 papers |
| [University of the Free State](https://md.research.elamur.ai/free/en/universities/university-of-the-free-state/) | 13 papers |
| [Living Tongues Institute for Endangered Languages](https://ror.org/00gens972) | 11 papers |

Countries of the authors on this topic Israel comes first: authors working there now wrote 145 papers on this topic.

**The numbers**

| Country | Papers |
| --- | --- |
| [Israel](https://md.research.elamur.ai/free/en/countries/israel/) | 145 papers |
| [United States](https://md.research.elamur.ai/free/en/countries/united-states/) | 94 papers |
| [South Africa](https://md.research.elamur.ai/free/en/countries/south-africa/) | 28 papers |
| [Germany](https://md.research.elamur.ai/free/en/countries/germany/) | 18 papers |
| [Netherlands](https://md.research.elamur.ai/free/en/countries/netherlands/) | 14 papers |
| [United Kingdom](https://md.research.elamur.ai/free/en/countries/united-kingdom/) | 13 papers |
| [France](https://md.research.elamur.ai/free/en/countries/france/) | 9 papers |
| [Switzerland](https://md.research.elamur.ai/free/en/countries/switzerland/) | 9 papers |
| [Italy](https://md.research.elamur.ai/free/en/countries/italy/) | 7 papers |
| [Belgium](https://md.research.elamur.ai/free/en/countries/belgium/) | 5 papers |

Venues mirror the two communities: *Glossa* for Modern Hebrew theory; the *Journal of Hebrew Scriptures*, *Journal of Semitic Studies* and *Folia Orientalia* for Biblical Hebrew; ACL venues, LREC and the Semitic and ancient-language workshops for computation.

Where the papers on this topic were published Journal of Hebrew Scriptures carries the most, 34 of 482.

**The numbers**

| Venue | Papers |
| --- | --- |
| [Journal of Hebrew Scriptures](https://md.research.elamur.ai/free/en/venues/journal-of-hebrew-scriptures/) | 34 papers |
| [HIPHIL Novum](https://md.research.elamur.ai/free/en/venues/hiphil-novum/) | 26 papers |
| [Language Resources and Evaluation Conference (LREC)](https://md.research.elamur.ai/free/en/venues/lrec/) | 25 papers |
| [Folia Orientalia](https://md.research.elamur.ai/free/en/venues/folia-orientalia/) | 20 papers |
| [Glossa a journal of general linguistics](https://md.research.elamur.ai/free/en/venues/glossa-a-journal-of-general-linguistics/) | 20 papers |
| [Findings of the Association for Computational Linguistics](https://md.research.elamur.ai/free/en/venues/findings/) | 18 papers |
| [Annual Meeting of the Association for Computational Linguistics (ACL)](https://md.research.elamur.ai/free/en/venues/acl/) | 10 papers |
| [HTS Teologiese Studies / Theological Studies](https://md.research.elamur.ai/free/en/venues/hts-teologiese-studies-theological-studies/) | 10 papers |
| [Workshop on Language Technologies for Historical and Ancient Languages @ LREC](https://md.research.elamur.ai/free/en/venues/lt4hala/) | 8 papers |
| [Old Testament Essays](https://md.research.elamur.ai/free/en/venues/old-testament-essays/) | 7 papers |

## Open questions

- **Tokenization for Semitic morphology.** Pre-trained models still split words into units that cut across morphemes. Whether larger vocabularies, morphology-aware tokenizers or character-level models are the best answer is unsettled, as recent work such as [Explicit Morphological Knowledge Improves Pre-training](https://md.research.elamur.ai/free/en/papers/gueta-2023-explicit-morphological-knowledge-improves-pre/) and [Modelling the Morphology of Verbal Paradigms](https://md.research.elamur.ai/free/en/papers/samo-2026-modelling-morphology-verbal-paradigms-case/) shows.
- **Do models really disambiguate?** Studies such as [a study of Hebrew homographs in contextual models](https://md.research.elamur.ai/free/en/papers/shmidman-2023-do-pretrained-contextual-language-models/) probe whether contextual representations separate the readings of an unpointed word.
- **Joint versus pipeline in the neural era.** Whether end-to-end neural models make explicit lattices unnecessary is still debated; see [A Truly Joint Neural Architecture](https://md.research.elamur.ai/free/en/papers/levi-2024-truly-joint-neural-architecture-segmentation/) and [MRL Parsing Without Tears](https://md.research.elamur.ai/free/en/papers/shmidman-2024-mrl-parsing-without-tears-case/).
- **Linguistic dating of biblical texts.** The disagreement between chronological and stylistic accounts of Early and Late Biblical Hebrew remains open, and statistical methods such as [a probabilistic study of syntactic variation](https://md.research.elamur.ai/free/en/papers/bijl-2019-probabilistic-approach-syntactic-variation-biblical/) are only beginning to test it at scale.
- **The Biblical Hebrew verbal system.** Tense, aspect, mood and discourse function of the verb forms still lack a consensus account.
- **Word order.** Whether Biblical Hebrew is basically verb-subject or subject-verb, and how information structure governs deviations, is argued in papers like [The Typological Classification of Genesis](https://md.research.elamur.ai/free/en/papers/holmstedt-2011-typological-classification-hebrew-genesis-subject/) and [a model of word order in Classical Biblical Hebrew](https://md.research.elamur.ai/free/en/papers/khan-2020-towards-comprehensive-model-interpreting-word/).
- **Low-resource historical varieties.** Syriac, Jewish Babylonian Aramaic, Neo-Aramaic and medieval Hebrew have few annotated corpora; synthetic data, as in [Deep Aramaic](https://md.research.elamur.ai/free/en/papers/aioanei-2023-deep-aramaic-towards-synthetic-data/), and cross-lingual projection are partial remedies.
- **Speech without vowels.** Hebrew text-to-speech must infer pronunciation from unpointed text; [Phonikud](https://md.research.elamur.ai/free/en/papers/kolani-2025-phonikud-overcoming-phonetic-underspecification-hebrew/) and [diacritic-free Hebrew TTS](https://md.research.elamur.ai/free/en/papers/roth-2024-language-modeling-approach-diacritic-free/) take different routes.
- **Generative models for a mid-resource language.** How to adapt large models to Hebrew efficiently, and how to evaluate them fairly, is the newest open front.

## Neighbouring topics

Hebrew and Aramaic sit inside the larger field of [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/), with which almost every paper here is shared; that page covers Arabic, Akkadian, Ethiopic and comparative Semitics. The closest technical neighbour is [Morphology](https://md.research.elamur.ai/free/en/topics/morphology/), followed by [Tokenization & segmentation](https://md.research.elamur.ai/free/en/topics/tokenization/), [Part-of-speech tagging](https://md.research.elamur.ai/free/en/topics/part-of-speech-tagging/), [Dependency parsing](https://md.research.elamur.ai/free/en/topics/dependency-parsing/), [Treebanks](https://md.research.elamur.ai/free/en/topics/treebanks/) and [Universal Dependencies](https://md.research.elamur.ai/free/en/topics/universal-dependencies/). Hebrew encoders and DictaLM link the topic to [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) and [Evaluation & benchmarks](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/); Hebrew–English and Hebrew–Arabic systems to [Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/) and [Multilingual & cross-lingual NLP](https://md.research.elamur.ai/free/en/topics/multilingual-nlp/); applied work to [Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/), [Summarization](https://md.research.elamur.ai/free/en/topics/summarization/), [Named entity recognition](https://md.research.elamur.ai/free/en/topics/named-entity-recognition/), [Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/), [Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) and [Idioms & multiword expressions](https://md.research.elamur.ai/free/en/topics/multiword-expressions/).

On the humanities side, biblical studies connect to [Historical linguistics & language change](https://md.research.elamur.ai/free/en/topics/language-change/), [Discourse analysis](https://md.research.elamur.ai/free/en/topics/discourse-analysis/) and [Syntax & grammar formalisms](https://md.research.elamur.ai/free/en/topics/syntax-grammar/); parallel Hebrew–Greek treebanks and Septuagint studies to [Ancient Greek](https://md.research.elamur.ai/free/en/topics/ancient-greek/); and child-language studies to [Psycholinguistics & language acquisition](https://md.research.elamur.ai/free/en/topics/psycholinguistics/). Under-resourced Aramaic varieties belong also to [Low-resource languages](https://md.research.elamur.ai/free/en/topics/low-resource-languages/). Nearly all papers here stand on the [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) shelf, a substantial share on [Natural language processing](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/), and smaller groups on [Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/), [History & humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/history/) and [Law](https://md.research.elamur.ai/free/en/shelves/social-sciences/law/).

Neighbouring topics: papers shared Semitic languages shares the most: 392 of the 482 papers here (81%) stand on both.

**The numbers**

| Topic | Papers shared |
| --- | --- |
| [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) | 392 shared · 81% |
| [Biblical Hebrew](https://md.research.elamur.ai/free/en/topics/biblical-hebrew/) | 254 shared · 53% |
| [Modern Hebrew](https://md.research.elamur.ai/free/en/topics/modern-hebrew/) | 153 shared · 32% |
| [Aramaic](https://md.research.elamur.ai/free/en/topics/aramaic/) | 63 shared · 13% |
| [Morphology](https://md.research.elamur.ai/free/en/topics/morphology/) | 53 shared · 11% |
| [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) | 15 shared · 3% |
| [Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/) | 14 shared · 3% |
| [Syntax & grammar formalisms](https://md.research.elamur.ai/free/en/topics/syntax-grammar/) | 14 shared · 3% |
| [Historical linguistics & language change](https://md.research.elamur.ai/free/en/topics/language-change/) | 11 shared · 2% |
| [Universal Dependencies](https://md.research.elamur.ai/free/en/topics/universal-dependencies/) | 11 shared · 2% |

The shelves this topic stands on Language and Linguistics: 440 of the 482 papers here.

**The numbers**

| Shelf | Papers |
| --- | --- |
| [Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) | 440 papers |
| [Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) | 218 papers |
| [Religious studies](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/religious-studies/) | 193 papers |
| [Literature and Literary Theory](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/literature-and-literary-theory/) | 72 papers |
| [Sociology and Political Science](https://md.research.elamur.ai/free/en/shelves/social-sciences/sociology-and-political-science/) | 27 papers |
| [Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) | 26 papers |
| [Developmental and Educational Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/developmental-and-educational-psychology/) | 23 papers |
| [Archeology](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/archeology/) | 19 papers |
| [Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) | 18 papers |
| [History](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/history/) | 12 papers |
| [Political Science and International Relations](https://md.research.elamur.ai/free/en/shelves/social-sciences/political-science-and-international-relations/) | 10 papers |
| [Cognitive Neuroscience](https://md.research.elamur.ai/free/en/shelves/neuroscience/cognitive-neuroscience/) | 9 papers |

## Key works in this library

- [HMM vowel restoration for Arabic and Hebrew](https://md.research.elamur.ai/free/en/papers/gal-2002-hmm-approach-vowel-restoration-arabic/) — the reference model for restoring vowels.
- [Multitiered Nonlinear Morphology Using Multitape Finite Automata](https://md.research.elamur.ai/free/en/papers/kiraz-2000-multitiered-nonlinear-morphology-multitape-finite/) — finite-state morphology for Syriac roots and patterns.
- [Morpho-lexical probabilities from untagged text](https://md.research.elamur.ai/free/en/papers/levinger-1995-learning-morpho-lexical-probabilities-untagged/) — corpus priors for ambiguous words.
- [Architectures for Hebrew segmentation and tagging](https://md.research.elamur.ai/free/en/papers/haim-2005-choosing-optimal-architecture-segmentation-pos/) — word-level versus morpheme-level tagging.
- [A Finite-State Morphological Grammar of Hebrew](https://md.research.elamur.ai/free/en/papers/yona-2005-finite-state-morphological-grammar-hebrew/) — the HAMSAH analyser and generator.
- [Classifier combination for Hebrew disambiguation](https://md.research.elamur.ai/free/en/papers/shacham-2007-morphological-disambiguation-hebrew-case-study/) — per-feature classifiers plus constraints.
- [Identifying Semitic Roots](https://md.research.elamur.ai/free/en/papers/daya-2008-identifying-semitic-roots-machine-learning/) — learning to extract roots.
- [Easy-First Dependency Parsing of Modern Hebrew](https://md.research.elamur.ai/free/en/papers/goldberg-2010-easy-first-dependency-parsing-modern/) — non-directional parsing with agreement features.
- [A pointer network for Hebrew morphology](https://md.research.elamur.ai/free/en/papers/seker-2020-pointer-network-architecture-joint-morphological/) — neural selection over symbolic lattices.
- [Neural sentiment analysis for Modern Hebrew](https://md.research.elamur.ai/free/en/papers/amram-2018-representations-architectures-neural-sentiment-analysis/) — a Hebrew sentiment set and model comparison.
- [Nakdan: Professional Hebrew Diacritizer](https://md.research.elamur.ai/free/en/papers/shmidman-2020-nakdan-professional-hebrew-diacritizer/) — interactive vowel pointing for several genres.
- [BEREL](https://md.research.elamur.ai/free/en/papers/shmidman-2022-introducing-berel-bert-embeddings-rabbinic/) — a model for rabbinic Hebrew.
- [DictaLM 2.0](https://md.research.elamur.ai/free/en/papers/shmidman-2024-adapting-llms-hebrew-unveiling-dictalm/) — adapting a generative model to Hebrew.
- [At the interface of syntax and prosody](https://md.research.elamur.ai/free/en/papers/naude-2018-interface-syntax-prosody-differentiating-left/) — accents as evidence for biblical clause structure.
- [Qumran and historical Hebrew linguistics](https://md.research.elamur.ai/free/en/papers/naude-2016-contribution-qumran-historical-hebrew-linguistics/) — participial negation as a diachronic marker.
- [Dead Sea Scrolls Hebrew and the Tiberian tradition](https://md.research.elamur.ai/free/en/papers/hornkohl-2020-hebrew-dead-sea-scrolls-tiberian/) — dating the reading tradition.
- [Sound changes in the Masoretic reading tradition](https://md.research.elamur.ai/free/en/papers/suchard-2019-sound-changes-pre-masoretic-reading/) — Aramaic as a control for Hebrew sound change.
- [The Qumran Scrolls of Judges](https://md.research.elamur.ai/free/en/papers/rezetko-2013-qumran-scrolls-book-judges-literary/) — textual criticism against linguistic dating.
- [Attributive Constructions in NENA](https://md.research.elamur.ai/free/en/papers/gutman-2018-attributive-constructions-north-eastern/) — inheritance and contact in Neo-Aramaic noun phrases.
- [No case for Case in locality](https://md.research.elamur.ai/free/en/papers/friedmann-2017-no-case-case-locality-case/) — the object marker and intervention in acquisition and aphasia.

## Papers (482)

**1** [2](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/2/) … [6](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/6/) [next →](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/2/)

1. **[Colorless Green Recurrent Networks Dream Hierarchically](https://md.research.elamur.ai/free/en/papers/gulordava-2018-colorless-green-recurrent-networks-dream/) [Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)](https://md.research.elamur.ai/free/en/venues/naacl/)**Kristina Gulordava, Piotr Bojanowski, Edouard Grave, et al.*2018 · 760 citations*
2. [**An HMM Approach to Vowel Restoration in Arabic and Hebrew**Ya'akov Gal*2002 · 106 citations*](https://md.research.elamur.ai/free/en/papers/gal-2002-hmm-approach-vowel-restoration-arabic/)
3. [**An Unsupervised Morpheme-Based HMM for Hebrew Morphological Disambiguation**Meni Adler, Michael Elhadad*2006 · 99 citations*](https://md.research.elamur.ai/free/en/papers/adler-2006-unsupervised-morpheme-based-hmm-hebrew/)
4. **[Cross-Linguistic Syntactic Evaluation of Word Prediction Models](https://md.research.elamur.ai/free/en/papers/mueller-2020-cross-linguistic-syntactic-evaluation-word/) [Annual Meeting of the Association for Computational Linguistics (ACL)](https://md.research.elamur.ai/free/en/venues/acl/)**Aaron Mueller, Garrett Nicolai, Panayiota Petrou-Zeniou, et al.*2020 · 97 citations*
5. [**Multitiered Nonlinear Morphology Using Multitape Finite Automata: A Case Study on Syriac and Arabic**George Anton Kiraz*2000 · 96 citations*](https://md.research.elamur.ai/free/en/papers/kiraz-2000-multitiered-nonlinear-morphology-multitape-finite/)
6. [**Missing objects in Hebrew: Argument ellipsis, not VP ellipsis**Idan Landau*2018 · 89 citations*](https://md.research.elamur.ai/free/en/papers/landau-2018-missing-objects-hebrew-argument-ellipsis/)
7. [**The Tiberian Pronunciation Tradition of Biblical Hebrew**Geoffrey Khan*2020 · 75 citations*](https://md.research.elamur.ai/free/en/papers/khan-2020-tiberian-pronunciation-tradition-biblical/)
8. [**HeBERT & HebEMO: a Hebrew BERT Model and a Tool for Polarity Analysis and Emotion Recognition**Avihay Chriqui, Inbal Yahav*2021 · 68 citations*](https://md.research.elamur.ai/free/en/papers/chriqui-2021-hebert-hebemo-hebrew-bert-model/)
9. [**Late Biblical Hebrew and the Qumran Pesher Habakkuk**Ian Young*2008 · 67 citations*](https://md.research.elamur.ai/free/en/papers/young-2008-late-biblical-hebrew-qumran-pesher/)
10. [**Integrated Morphological and Syntactic Disambiguation for Modern Hebrew**Reut Tsarfaty*2006 · 67 citations*](https://md.research.elamur.ai/free/en/papers/tsarfaty-2006-integrated-morphological-syntactic-disambiguation-modern/)
11. [**A Graph-based Lattice Dependency Parser for Joint Morphological Segmentation and Syntactic Analysis**Wolfgang Seeker, Özlem Çetinoğlu*2015 · 66 citations*](https://md.research.elamur.ai/free/en/papers/seeker-2015-graph-based-lattice-dependency-parser/)
12. **[Typology of case alignments in NENA dialects](https://md.research.elamur.ai/free/en/papers/barotto-2014-typology-case-alignments-nena-dialects/) [RiCOGNIZIONI. Rivista di Lingue e Letterature straniere e Culture moderne](https://md.research.elamur.ai/free/en/venues/ricognizioni-rivista-di-lingue-e-letterature-straniere-e-culture-moderne/)**Alessandra Barotto*2014 · 66 citations*
13. [**Multi-Tape Two-Level Morphology: A Case Study in Semitic Non-Linear Morphology**George Anton Kiraz*1994 · 62 citations*](https://md.research.elamur.ai/free/en/papers/kiraz-1994-multi-tape-two-level-morphology/)
14. [**Joint Transition-Based Models for Morpho-Syntactic Parsing: Parsing Strategies for MRLs and a Case Study from Modern Hebrew**Amir More, Amit Seker, Victoria Basmova, et al.*2019 · 60 citations*](https://md.research.elamur.ai/free/en/papers/more-2019-joint-transition-based-models-morpho/)
15. [**AlephBERT: Language Model Pre-training and Evaluation from Sub-Word to Sentence Level**Amit Seker, Elron Bandel, Dan Bareket, et al.*2022 · 57 citations*](https://md.research.elamur.ai/free/en/papers/seker-2022-alephbert-language-model-pre-training/)
16. [**AlephBERT: A Hebrew Large Pre-Trained Language Model to Start-off your Hebrew NLP Application With**Amit Seker, Elron Bandel, Dan Bareket, et al.*2021 · 56 citations*](https://md.research.elamur.ai/free/en/papers/seker-2021-alephbert-hebrew-large-pre-trained/)
17. [**Reflexive verbs in Hebrew: Deep unaccusativity meets lexical semantics**Itamar Kastner*2017 · 52 citations*](https://md.research.elamur.ai/free/en/papers/kastner-2017-reflexive-verbs-hebrew-deep-unaccusativity/)
18. [**Artificial intelligence based writer identification generates new evidence for the unknown scribes of the Dead Sea Scrolls exemplified by the Great Isaiah Scroll (1QIsaᵃ)**Mladen Popovic, Maruf A. Dhali, Lambert Schomaker*2020 · 51 citations*](https://md.research.elamur.ai/free/en/papers/popovic-2020-artificial-intelligence-based-writer-identification/)
19. [**What's Wrong with Hebrew NLP? And How to Make it Right**Reut Tsarfaty, Shoval Sadde, Stav Klein, et al.*2019 · 50 citations*](https://md.research.elamur.ai/free/en/papers/tsarfaty-2019-what-s-wrong-hebrew-nlp/)
20. [**Unsupervised Decomposition of a Document into Authorial Components**Moshe Koppel, Navot Akiva, Idan Dershowitz, et al.*2011 · 46 citations*](https://md.research.elamur.ai/free/en/papers/koppel-2011-unsupervised-decomposition-document-into-authorial/)
21. [**Choosing an Optimal Architecture for Segmentation and POS-Tagging of Modern Hebrew**Roy Bar-Haim, Khalil Sima'an, Yoad Winter*2005 · 46 citations*](https://md.research.elamur.ai/free/en/papers/haim-2005-choosing-optimal-architecture-segmentation-pos/)
22. [**Neural Decipherment via Minimum-Cost Flow: From Ugaritic to Linear B**Jiaming Luo, Yuan Cao, Regina Barzilay*2019 · 43 citations*](https://md.research.elamur.ai/free/en/papers/luo-2019-neural-decipherment-minimum-cost-flow/)
23. [**Learning Morpho-Lexical Probabilities from an Untagged Corpus with an Application to Hebrew**Moshe Levinger, Uzzi Ornan, Alon Itai*1995 · 43 citations*](https://md.research.elamur.ai/free/en/papers/levinger-1995-learning-morpho-lexical-probabilities-untagged/)
24. [**Rapid prototyping of a transfer-based Hebrew-to-English machine translation system**Alon Lavie, Erik Peterson, Katharina Probst, et al.*2004 · 42 citations*](https://md.research.elamur.ai/free/en/papers/lavie-2004-rapid-prototyping-transfer-based-hebrew/)
25. [**The Hebrew of the Dead Sea Scrolls and the Tiberian Reading Tradition**Aaron D. Hornkohl*2020 · 40 citations*](https://md.research.elamur.ai/free/en/papers/hornkohl-2020-hebrew-dead-sea-scrolls-tiberian/)

**1** [2](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/2/) … [6](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/6/) [next →](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/page/2/)

Updated 24 October 2026

- [← previous topic Grammatical error correction & learner language](https://md.research.elamur.ai/free/en/topics/grammatical-error-correction/)
- [next topic → Historical linguistics & language change](https://md.research.elamur.ai/free/en/topics/language-change/)
- [↑ back to topics](https://md.research.elamur.ai/free/en/topics/)