giampierosalvi/affordances_and_speech

Dataset Card for Affordances and Speech Dataset

This dataset was created for the affordances and speech experiments published in:

Language bootstrapping: learning word meanings from perception-action association.
Giampiero Salvi , Luis Montesano, Alexandre Bernardino, José Santos-Victor
IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) (Volume: 42, Issue: 3, June 2012)
DOI: 10.1109/TSMCB.2011.2172420
http://ieeexplore.ieee.org/document/6082460/

Open access version: https://arxiv.org/abs/1711.09714

The goal of the experiments was to explore learning methods that would ground the meaning of spoken words to the perception-action associations learned by a robot through experimentation.

The code for the experiments can be found at https://github.com/giampierosalvi/AffordancesAndSpeech

The recordings started in 2007, where later updated on 2010-01-28, and were finally uploaded to huggingface on 2025-09-22.

Dataset Details

Dataset Description

The main audio directory contains audio recordings (.wav) with the corresponding orthographic transcription (.lab). The recordings are divided into

• words: for isolated words
• sentences: for full sentences
• instructions: instructions to the robot The files are further divided into directory corresponding to recording sessions with the following format: {f|m}xx{langid}{s|w|i}yy, where:
• {f|m}: indicates the speaker's gender
• xx: is the speaker ID
• langid: is a two-letter code indicating the speaker's mother tongue
• {s|w|i}: indicates weather the recording contains a sentence, word, or instruction
• yy: is the recording session ID. For example, f03pts00 is a sentence recorded by a Portuguese mother tongue female speaker.

Information about the recording sessions is stored in the file SESINFO.txt.

Besides the format described above, each recording also contains the experiment ID that is useful to recover the affordance parameters stored in affordance_data.csv. For example, m03its17_213.wav contains a recording corresponding to the experiment number 213 made by an Italian mother tongue male speaker in the recording session number 17. The corresponding affordance parameters are, in this case:

• Action: touch
• Color: green2
• Shape: box
• ObjVel: slow
• ObjHandVel: slow
• HandVel: slow
• Contact: long

The affordance_data.csv file contains affordance parameters for every experiment carried out by the robot. In total there are 254 experiments. Each experiment is narrated by more than one speaker, so the number of spoken descriptions is a multiple of 254.

The file full.dic contains a pronunciation dictionary for each word in the material expressed in a modified version of the SAMPA code for British English. The list of symbols together with their meaning and the TIMA code to use them in LaTeX are described in the file sampa_latex_en.pdf in the column HTK.

• Curated by: Giampiero Salvi
• Funded by [optional]: FCT (Portugal), Vetenskapsrådet (Sweden)
• Shared by [optional]: Giampiero Salvi
• Language(s) (NLP): English
• License: cc-by-nc-sa-4.0

Dataset Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Uses

The dataset can be used to reproduce the human-robot interaction experiments described in the publication above, or to extend them by proposing new learning methods.

Direct Use

Out-of-Scope Use

Dataset Structure

[More Information Needed]

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Data Collection and Processing

Recordings took place in two occasions:

• Sessions 00-16 were recorded at IST, Lisbon, Portugal. Each speaker was free to use their computer and their headset in their office. This produced recordings of poor quality that turned out to be difficult to use at the time, but could still be useful today.
• Sessions 17-21 were recorded at KTH, Stockholm, Sweden in a soundproof room, using a TASCAM US-122 sound card and a high quality balanced close microphone. Sessions 17-21 were the ones used in the publication mentioned at the beginning.

Who are the source data producers?

Giampiero Salvi

Annotations [optional]

Giampiero Salvi

Annotation process

[More Information Needed]

Who are the annotators?

Giampiero Salvi

Personal and Sensitive Information

Speakers have been anonymised and no personal information is included in the data.

Bias, Risks, and Limitations

There is a slight over-representation of male speakers in the data.

Recommendations

Citation [optional]

Language bootstrapping: learning word meanings from perception-action association.
Giampiero Salvi , Luis Montesano, Alexandre Bernardino, José Santos-Victor
IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) (Volume: 42, Issue: 3, June 2012)
DOI: 10.1109/TSMCB.2011.2172420
http://ieeexplore.ieee.org/document/6082460/

Open access version: https://arxiv.org/abs/1711.09714

BibTeX:

@ARTICLE{6082460,
  author={Salvi, Giampiero and Montesano, Luis and Bernardino, Alexandre and Santos-Victor, José},
  journal={IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics)}, 
  title={Language Bootstrapping: Learning Word Meanings From Perception–Action Association}, 
  year={2012},
  volume={42},
  number={3},
  pages={660-671},
  keywords={Speech;Robot sensing systems;Speech recognition;Context;Computational modeling;Humans;Affordances;automatic speech recognition;Bayesian networks;cognitive robotics;grasping;humanoid robots;language;unsupervised learning},
  doi={10.1109/TSMCB.2011.2172420}
}

APA:

G. Salvi, L. Montesano, A. Bernardino and J. Santos-Victor, "Language Bootstrapping: Learning Word Meanings From Perception–Action Association," in IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 42, no. 3, pp. 660-671, June 2012, doi: 10.1109/TSMCB.2011.2172420.

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Dataset Card Authors [optional]

Giampiero Salvi

Dataset Card Contact

Giampiero Salvi