PASTRIE: A Corpus of Prepositions Annotated with Supersense Tags in Reddit International English

Related tags

Deep Learningpastrie
Overview

PASTRIE

CC BY-SA 4.0

Official release of the corpus described in the paper:

Michael Kranzlein, Emma Manning, Siyao Peng, Shira Wein, Aryaman Arora, and Nathan Schneider (2020). PASTRIE: A Corpus of Prepositions Annotated with Supersense Tags in Reddit International English [link]. Proceedings of the 14th Linguistic Annotation Workshop.


Overview

PASTRIE is a corpus of English data from Reddit annotated with preposition supersenses from the SNACS inventory.

While the data in PASTRIE is in English, it was produced by presumed speakers of four L1s:

  • English
  • French
  • German
  • Spanish

For details on how L1s were identified, see section 3.1 of Rabinovich et al. (2018).

Annotation Example

Below is an example sentence from the corpus, where annotation targets are bolded and preposition supersenses are annotated with the notation SceneRole↝Function. Together, a scene role and function are known as a construal.


Data Formats

PASTRIE is released in the following formats. We expect that most projects will be best served by one of the JSON formats.

  • .conllulex: the 19-column CoNLL-U-Lex format originally used for STREUSLE.
  • .json: a JSON representation of the CoNLL-U-Lex that does not require a CoNLL-U-Lex parser.
  • .govobj.json: an extended version of the JSON representation that contains information about each preposition's syntactic parent and object.

PASTRIE mostly follows STREUSLE with respect to the data format and SNACS annotation practice. Primary differences in the annotations are:

  • Lemmas, part-of-speech tags, and syntactic dependencies aim to follow the UD standard in both cases. They are gold in STREUSLE, versus automatic with some manual corrections in PASTRIE.
    • PASTRIE does not group together base+clitic combinations, whereas STREUSLE does (multiword tokens—where a single orthographic word contains multiple syntactic words).
    • PASTRIE does not regularly specify SpaceAfter=No to indicate alignment between the tokens and the raw text.
    • In PASTRIE, the raw text string accompanying the sentence may contain two or more consecutive spaces.
    • PASTRIE lacks enhanced dependencies.
  • Multiword expression annotations in PASTRIE are limited to expressions containing a preposition. Depending on the syntactic head, the expression may or may not have a SNACS supersense.
    • Verbal multiword expressions in PASTRIE are not subtyped in the lexcat; they all have a lexcat of V.
  • Noun and verb expressions in PASTRIE do not have supersense labels.
Comments
  • Misc. annotation errors and/or conversion script bugs

    Misc. annotation errors and/or conversion script bugs

    There are some annotations which I'm fairly sure are incorrect and are choking up the JSON conversion script. (These errors occur using the unmodified versions of all scripts taken straight from STRUESLE.) One or two might also be indicative of a bug in the conllulex2json.py file.

    1. vs mistagged as a noun--should be prep

    AssertionError: ('french-fad32caf-e595-e3cb-07bf-aaea891e53cb-02', {'lexlemma': 'versus', 'lexcat': 'CCONJ', 'ss': 'c', 'ss2': 'c', 'toknums': [3]}, {'#': 3, 'word': 'vs', 'lemma': 'versus', 'upos': 'NOUN', 'xpos': 'NN', 'feats': None, 'head': 8, 'deprel': 'nsubj', 'edeps': None, 'misc': None, 'smwe': None, 'wmwe': None, 'lextag': 'O-CCONJ-`c'})

    1. ditto

    AssertionError: ('french-fad32caf-e595-e3cb-07bf-aaea891e53cb-02', {'lexlemma': 'versus', 'lexcat': 'CCONJ', 'ss': 'c', 'ss2': 'c', 'toknums': [3]}, {'#': 3, 'word': 'vs', 'lemma': 'versus', 'upos': 'NOUN', 'xpos': 'NN', 'feats': None, 'head': 8, 'deprel': 'nsubj', 'edeps': None, 'misc': None, 'smwe': None, 'wmwe': None, 'lextag': 'O-CCONJ-`c'})

    1. Script complains about "to" in this snippet at ID=23. Not immediately clear to me what the issue is--perhaps that "to" is labeled ADP/IN? For its xpos I think it ought to be TO, not sure about its upos. Snippet:
    13      shit    shit    NOUN    NN      _       16      obl:npmod       _       _       _       _       _       _       _       _       _       _       _
    14      this    this    PRON    DT      _       16      nsubj   _       _       _       _       _       _       _       _       _       _       _
    15      can     can     AUX     MD      _       16      aux     _       _       _       _       _       _       _       _       _       _       _
    16      end     end     VERB    VB      _       4       parataxis       _       _       _       _       _       _       _       _       _       _       _
    17      right   right   ADV     RB      _       18      advmod  _       _       _       _       _       _       _       _       _       _       _
    18      now     now     ADV     RB      _       16      advmod  _       _       _       _       _       _       _       _       _       _       _
    19      if      if      SCONJ   IN      _       21      mark    _       _       _       _       _       _       _       _       _       _       _
    20      I       I       PRON    PRP     _       21      nsubj   _       _       _       _       _       _       _       _       _       _       _
    21      want    want    VERB    VBP     _       16      advcl   _       _       _       _       _       _       _       _       _       _       _
    22      it      it      PRON    PRP     _       21      obj     _       _       _       _       _       _       _       _       _       _       _
    23      to      to      ADP     IN      _       21      obl     _       _       _       _       _       `i      `i      _       _       _       _
    24      .       .       PUNCT   .       _       4       punct   _       _       _       _       _       _       _       _       _       _       _
    

    Error:

    AssertionError: ('french-a17a4340-f9c0-8fef-fa1b-1bf13879399b-02', {'lexlemma': 'to', 'lexcat': 'INF', 'ss': 'i', 'ss2': 'i', 'toknums': [23]}, {'#': 23, 'word': 'to', 'lemma': 'to', 'upos': 'ADP', 'xpos': 'IN', 'feats': None, 'head': 21, 'deprel': 'obl', 'edeps': None, 'misc': None, 'smwe': None, 'wmwe': None, 'lextag': 'O-INF-`i'})

    Relevant span of code:

                if validate_pos and upos!=lc and (upos,lc) not in {('NOUN','N'),('PROPN','N'),('VERB','V'),
                    ('ADP','P'),('ADV','P'),('SCONJ','P'),
                    ('ADP','DISC'),('ADV','DISC'),('SCONJ','DISC'),
                    ('PART','POSS')}:
                    # most often, the single-word lexcat should match its upos
                    # check a list of exceptions
                    mismatchOK = False
                    if xpos=='TO' and lc.startswith('INF'):
                        mismatchOK = True
                    elif (xpos=='TO')!=lc.startswith('INF'):
                        assert upos in ['SCONJ', "ADP"] and swe['lexlemma']=='for',(sent['sent_id'],swe,tok)
                        mismatchOK = True
    
    1. Originator as function:

    (in french-c02823ec-60bd-adce-7327-01337eb9d1c8-02) AssertionError: ('p.Originator should never be function', {'lexlemma': 'you', 'lexcat': 'PRON.POSS', 'ss': 'p.Originator', 'ss2': 'p.Originator', 'toknums': [1]})

    1. lexcat DISC with ADJ:

    AssertionError: In spanish-a25e8289-e04a-f5af-ce56-ead9faca65b1-02, single-word expression 'like' has lexcat DISC, which is incompatible with its upos ADJ

    1. "her" tagged with Possessor is incorrectly parsed as iobj and tagged as PRP instead of PRP$. Relevant snippet:
    1       My      my      PRON    PRP$    _       2       nmod:poss       _       _       _       _       _       SocialRel       Gestalt _       _       _       _
    2       grandma grandma NOUN    NN      _       3       nsubj   _       _       _       _       _       _       _       _       _       _       _
    3       had     have    VERB    VBD     _       0       root    _       _       _       _       _       _       _       _       _       _       _
    4       her     she     PRON    PRP     _       3       iobj    _       _       _       _       _       Possessor       Possessor       _       _       _       _
    5       super   super   ADV     RB      _       6       advmod  _       _       _       _       _       _       _       _       _       _       _
    6       thick   thick   ADJ     JJ      _       8       amod    _       _       _       _       _       _       _       _       _       _       _
    7       floor   floor   NOUN    NN      _       8       compound        _       _       _       _       _       _       _       _       _       _       _
    8       mats    mat     NOUN    NNS     _       3       obj     _       _       _       _       _       _       _       _       _       _       _
    9       *       *       PUNCT   NFP     _       8       punct   _       _       _       _       _       _       _       _       _       _       _
    10      over    over    ADP     IN      _       13      case    _       _       _       _       _       Locus   Locus   _       _       _       _
    11      *       *       PUNCT   NFP     _       13      punct   _       _       _       _       _       _       _       _       _       _       _
    12      the     the     DET     DT      _       13      det     _       _       _       _       _       _       _       _       _       _       _
    13      accelerator     accelerator     NOUN    NN      _       3       obl     _       _       _       _       _       _       _       _       _       _       _
    14      ,       ,       PUNCT   ,       _       3       punct   _       _       _       _       _       _       _       _       _       _       _
    

    Error:

    AssertionError: In spanish-ebba3c73-2431-c216-8f4d-d469ee8d5564-01, single-word expression 'her' has lexcat P, which is incompatible with its upos PRON

    1. "NA" is misannotated--this is NA as in North America, i.e. a PROPN/NP, but it's lemmatized as "no", and its tags are weird.

    AssertionError: ('german-35000895-1d78-c18a-01ed-f7410b9c0581-01', {'lexlemma': 'no', 'lexcat': 'ADV', 'ss': None, 'ss2': None, 'toknums': [5]}, {'#': 5, 'word': 'NA', 'lemma': 'no', 'upos': 'PART', 'xpos': 'TO', 'feats': None, 'head': 6, 'deprel': 'mark', 'edeps': None, 'misc': None, 'smwe': None, 'wmwe': None, 'lextag': 'O-ADV'})

    opened by lgessler 6
  • Prepositional supersense annotations on non-preposition targets

    Prepositional supersense annotations on non-preposition targets

    Is it OK for a verb-headed SMWE to have a prepositional supersense? The validator complains about it. Offending SMWE:

    21	give	give	VERB	VB	_	10	conj	_	_	2:1	_	give up on	p.Theme	p.Theme	_	_	_	_
    22	up	up	ADP	RP	_	21	compound:prt	_	_	2:2	_	_	_	_	_	_	_	_
    23	on	on	ADP	IN	_	24	case	_	_	2:3	_	_	_	_	_	_	_	_
    
    opened by lgessler 5
  • Prepositions unannotated for supersense

    Prepositions unannotated for supersense

    Token 6:

    # sent_id = french-f57dd6ab-5263-4c8a-e360-8ec683e6a37a-02
    # text = Once you have the hang of it it s pretty fast ( and does n't eat your clutch ) .
    1	Once	once	SCONJ	IN	_	3	mark	_	_	_	_	_	_	_	_	_	_	_
    2	you	you	PRON	PRP	_	3	nsubj	_	_	_	_	_	_	_	_	_	_	_
    3	have	have	VERB	VBP	_	11	advcl	_	_	_	_	_	_	_	_	_	_	_
    4	the	the	DET	DT	_	5	det	_	_	_	_	_	_	_	_	_	_	_
    5	hang	hang	NOUN	NN	_	3	obj	_	_	_	_	_	_	_	_	_	_	_
    6	of	of	ADP	IN	_	7	case	_	_	_	_	_	_	_	_	_	_	_
    7	it	it	PRON	PRP	_	5	nmod	_	_	_	_	_	_	_	_	_	_	_
    8	it	it	PRON	PRP	_	11	nsubj	_	_	_	_	_	_	_	_	_	_	_
    9	s	be	AUX	VBZ	_	11	cop	_	_	_	_	_	_	_	_	_	_	_
    10	pretty	pretty	ADV	RB	_	11	advmod	_	_	_	_	_	_	_	_	_	_	_
    11	fast	fast	ADJ	JJ	_	0	root	_	_	_	_	_	_	_	_	_	_	_
    12	(	(	PUNCT	-LRB-	_	16	punct	_	_	_	_	_	_	_	_	_	_	_
    13	and	and	CCONJ	CC	_	16	cc	_	_	_	_	_	_	_	_	_	_	_
    14	does	do	AUX	VBZ	_	16	aux	_	_	_	_	_	_	_	_	_	_	_
    15	n't	not	PART	RB	_	16	advmod	_	_	_	_	_	_	_	_	_	_	_
    16	eat	eat	VERB	VB	_	11	conj	_	_	_	_	_	_	_	_	_	_	_
    17	your	you	PRON	PRP$	_	18	nmod:poss	_	_	_	_	_	Possessor	Possessor	_	_	_	_
    18	clutch	clutch	NOUN	NN	_	16	obj	_	_	_	_	_	_	_	_	_	_	_
    19	)	)	PUNCT	-RRB-	_	11	punct	_	_	_	_	_	_	_	_	_	_	_
    20	.	.	PUNCT	.	_	11	punct	_	_	_	_	_	_	_	_	_	_	_
    

    I assumed that all preps were supposed to be annotated, but perhaps not?

    opened by lgessler 3
  • Apostrophes removed in preprocessing?

    Apostrophes removed in preprocessing?

    Looking through the data, there are a LOT of sentences where clitics are tokenized off but lack an apostrophe. Is that just the genre or did they get lost in preprocessing?

    opened by nschneid 2
  • Dataset requested

    Dataset requested

    Hi all,

    I would like to request the PASTRIE dataset accompanying the paper "PASTRIE: A Corpus of Prepositions Annotated with Supsersense Tags in Reddit International English".

    Thanks for reply.

    opened by fj-morales 2
  • SNACS supersense tags should start with

    SNACS supersense tags should start with "p."

    For compatibility with STREUSLE, it should be p.Locus, p.Theme, etc.

    Special labels like `i `d `c `$ ?? should not start with p.. In fact, the backtick labels from annotation are not represented as such in STREUSLE—they are reflected in the LEXCAT column of the data.

    opened by nschneid 0
  • Questionable adpositional MWEs

    Questionable adpositional MWEs

    • in_male_term — from "in male terms"; should be in_term (at most)
    • in_the_first_place
    • in_my_hand — from "in my hands"; should be in_hand (at most)
    • for_quite_some_time — just Duration for, weak MWE?
    • at_all_time — from what should have been "at all times". OK?
    • on_a_smaller_scale — omit adjective?
    • withouth — typo
    • see_as — "seeing as" (deverbal MWE acting like a preposition)
    opened by nschneid 0
  • Some undersegmentation of sentences

    Some undersegmentation of sentences

    Despite manual editing there are still places where a long sentence ought to be split up (esp. where it consists of a blockquoted sentence with > followed by a response). Looking for multiple consecutive spaces in the raw text uncovers some of these (as well as some discourse appendages like emoticons, which should probably remain in the same UD sentence).

    It would be nice to write a script to help clean these up—the tricky part is updating offsets in each parse.

    opened by nschneid 0
Releases(v2.0.1)
  • v2.0.1(Nov 21, 2021)

  • v2.0(Oct 22, 2021)

    • Switch to full .conllulex format following STREUSLE
      • add lexcats (#3), morphological features, newdoc directives
    • Scripts for validation and format conversion
    • Clean up various annotation issues, including:
      • restore apostrophes and fixing other conversion problems (#6, #9)
      • include pretokenized raw text (#12)
    Source code(tar.gz)
    Source code(zip)
  • v1.0.1(Dec 14, 2020)

    • Added .json file format
    • Switched lemmatization and pos tagging from StanfordNLP 0.2.0 to Stanza 1.1.1
    • Corrected rare encoding issue from v1.0
    Source code(tar.gz)
    Source code(zip)
Owner
NERT @ Georgetown
NERT @ Georgetown
A repository built on the Flow software package to explore cyber-security attacks on intelligent transportation systems.

A repository built on the Flow software package to explore cyber-security attacks on intelligent transportation systems.

George Gunter 4 Nov 14, 2022
[NeurIPS 2021] Better Safe Than Sorry: Preventing Delusive Adversaries with Adversarial Training

Better Safe Than Sorry: Preventing Delusive Adversaries with Adversarial Training Code for NeurIPS 2021 paper "Better Safe Than Sorry: Preventing Delu

Lue Tao 29 Sep 20, 2022
Back to Basics: Efficient Network Compression via IMP

Back to Basics: Efficient Network Compression via IMP Authors: Max Zimmer, Christoph Spiegel, Sebastian Pokutta This repository contains the code to r

IOL Lab @ ZIB 1 Nov 19, 2021
HyperCube: Implicit Field Representations of Voxelized 3D Models

HyperCube: Implicit Field Representations of Voxelized 3D Models Authors: Magdalena Proszewska, Marcin Mazur, Tomasz Trzcinski, Przemysław Spurek [Pap

Magdalena Proszewska 3 Mar 09, 2022
Fine-grained Post-training for Improving Retrieval-based Dialogue Systems - NAACL 2021

Fine-grained Post-training for Multi-turn Response Selection Implements the model described in the following paper Fine-grained Post-training for Impr

Janghoon Han 83 Dec 20, 2022
Implementation of CVPR'21: RfD-Net: Point Scene Understanding by Semantic Instance Reconstruction

RfD-Net [Project Page] [Paper] [Video] RfD-Net: Point Scene Understanding by Semantic Instance Reconstruction Yinyu Nie, Ji Hou, Xiaoguang Han, Matthi

Yinyu Nie 162 Jan 06, 2023
Official DGL implementation of "Rethinking High-order Graph Convolutional Networks"

SE Aggregation This is the implementation for Rethinking High-order Graph Convolutional Networks. Here we show the codes for citation networks as an e

Tianqi Zhang (张天启) 32 Jul 19, 2022
Offcial repository for the IEEE ICRA 2021 paper Auto-Tuned Sim-to-Real Transfer.

Offcial repository for the IEEE ICRA 2021 paper Auto-Tuned Sim-to-Real Transfer.

47 Jun 30, 2022
Datasets, Transforms and Models specific to Computer Vision

vision Datasets, Transforms and Models specific to Computer Vision Installation First install the nightly version of OneFlow python3 -m pip install on

OneFlow 68 Dec 07, 2022
Code release for NeurIPS 2020 paper "Co-Tuning for Transfer Learning"

CoTuning Official implementation for NeurIPS 2020 paper Co-Tuning for Transfer Learning. [News] 2021/01/13 The COCO 70 dataset used in the paper is av

THUML @ Tsinghua University 35 Sep 23, 2022
Implementation of the HMAX model of vision in PyTorch

PyTorch implementation of HMAX PyTorch implementation of the HMAX model that closely follows that of the MATLAB implementation of The Laboratory for C

Marijn van Vliet 52 Oct 13, 2022
An experiment on the performance of homemade Q-learning AIs in Agar.io depending on their state representation and available actions

Agar.io_Q-Learning_AI An experiment on the performance of homemade Q-learning AIs in Agar.io depending on their state representation and available act

1 Jun 09, 2022
The repository contains reproducible PyTorch source code of our paper Generative Modeling with Optimal Transport Maps, ICLR 2022.

Generative Modeling with Optimal Transport Maps The repository contains reproducible PyTorch source code of our paper Generative Modeling with Optimal

Litu Rout 30 Dec 22, 2022
Train/evaluate a Keras model, get metrics streamed to a dashboard in your browser.

Hera Train/evaluate a Keras model, get metrics streamed to a dashboard in your browser. Setting up Step 1. Plant the spy Install the package pip

Keplr 495 Dec 10, 2022
A voice recognition assistant similar to amazon alexa, siri and google assistant.

kenyan-Siri Build an Artificial Assistant Full tutorial (video) To watch the tutorial, click on the image below Installation For windows users (run th

Alison Parker 3 Aug 19, 2022
An implementation of the proximal policy optimization algorithm

PPO Pytorch C++ This is an implementation of the proximal policy optimization algorithm for the C++ API of Pytorch. It uses a simple TestEnvironment t

Martin Huber 59 Dec 09, 2022
Codebase for "ProtoAttend: Attention-Based Prototypical Learning."

Codebase for "ProtoAttend: Attention-Based Prototypical Learning." Authors: Sercan O. Arik and Tomas Pfister Paper: Sercan O. Arik and Tomas Pfister,

47 2 May 17, 2022
QRec: A Python Framework for quick implementation of recommender systems (TensorFlow Based)

Introduction QRec is a Python framework for recommender systems (Supported by Python 3.7.4 and Tensorflow 1.14+) in which a number of influential and

Yu 1.4k Dec 30, 2022
A Free and Open Source Python Library for Multiobjective Optimization

Platypus What is Platypus? Platypus is a framework for evolutionary computing in Python with a focus on multiobjective evolutionary algorithms (MOEAs)

Project Platypus 424 Dec 18, 2022
Matthew Colbrook 1 Apr 08, 2022